# Transcript: AI – digital humans are coming

## Description

Is it possible to teach a machine to write stories and draw pictures? What opportunities and threats does AI bring? How can artificial intelligence help today’s students, artists, small businesses, and CEOs of huge corporations?



Who are the guests?



Jan Tyl 



Jan Tyl is an enthusiast of artificial intelligence, startups, and corporations.He founded the companies Alpha Industris and Premier AI, which focus on expanding human capabilities through artificial intelligence. He studied artificial intelligence at MIT, Google Brain, Yonsei, Moscow, and occasionally lectures at various schools and conferences.His teams won the latest AI AWARDS for Digital Philosopher as AI idea of the year, and Digital Writer was nominated for the Prix Europa in the Digital Media Project category.



Dita Malečková



Dita Malečková studied philosophy and information science at Charles University. She works as a university lecturer and teaches contemporary philosophy, visual culture, art, and new technologies.In recent years, she has been particularly interested in artificial intelligence, especially in the context of theory of mind and imagination.Together with Jan Tyl, she created the projects Digital Philosopher and Digital Writer.  

## Transcript

**[00:00:00]** www.hradeckesluzby.cz

**[00:00:30]** www.hradeckesluzby.cz

**[00:01:00]** www.hradeckesluzby.cz

**[00:01:08]** www.hradeckesluzby.cz

**[00:01:16]** www.hradeckesluzby.cz

**[00:01:24]** www.hradeckesluzby.cz

**[00:01:32]** www.hradeckesluzby.cz

**[00:01:40]** www.hradeckesluzby.cz

**[00:01:48]** www.hradeckesluzby.cz

**[00:01:56]** www.hradeckesluzby.cz

**[00:02:04]** www.hradeckesluzby.cz

**[00:02:12]** www.hradeckesluzby.cz

**[00:02:20]** www.hradeckesluzby.cz

**[00:02:28]** www.hradeckesluzby.cz

**[00:02:36]** www.hradeckesluzby.cz

**[00:02:44]** www.hradeckesluzby.cz

**[00:02:52]** www.hradeckesluzby.cz

**[00:03:00]** www.hradeckesluzby.cz

**[00:03:08]** www.hradeckesluzby.cz

**[00:03:16]** www.hradeckesluzby.cz

**[00:03:24]** www.hradeckesluzby.cz

**[00:03:32]** www.hradeckesluzby.cz

**[00:03:40]** www.hradeckesluzby.cz

**[00:03:48]** www.hradeckesluzby.cz

**[00:03:56]** www.hradeckesluzby.cz

**[00:04:04]** www.hradeckesluzby.cz

**[00:04:12]** www.hradeckesluzby.cz

**[00:04:20]** www.hradeckesluzby.cz

**[00:04:28]** www.hradeckesluzby.cz

**[00:04:36]** www.hradeckesluzby.cz

**[00:04:44]** www.hradeckesluzby.cz

**[00:04:52]** www.hradeckesluzby.cz

**[00:05:00]** www.hradeckesluzby.cz

**[00:05:08]** www.hradeckesluzby.cz

**[00:05:16]** www.hradeckesluzby.cz

**[00:05:24]** www.hradeckesluzby.cz

**[00:05:32]** www.hradeckesluzby.cz

**[00:05:40]** www.hradeckesluzby.cz

**[00:05:48]** www.hradeckesluzby.cz

**[00:05:56]** www.hradeckesluzby.cz

**[00:06:04]** www.hradeckesluzby.cz

**[00:06:12]** www.hradeckesluzby.cz

**[00:06:20]** www.hradeckesluzby.cz

**[00:06:28]** www.hradeckesluzby.cz

**[00:06:36]** www.hradeckesluzby.cz

**[00:06:44]** www.hradeckesluzby.cz

**[00:06:52]** www.hradeckesluzby.cz

**[00:07:00]** www.hradeckesluzby.cz

**[00:07:08]** www.hradeckesluzby.cz

**[00:07:16]** www.hradeckesluzby.cz

**[00:07:24]** www.hradeckesluzby.cz

**[00:07:32]** www.hradeckesluzby.cz

**[00:07:40]** www.hradeckesluzby.cz

**[00:07:48]** www.hradeckesluzby.cz

**[00:07:56]** www.hradeckesluzby.cz

**[00:08:04]** www.hradeckesluzby.cz

**[00:08:12]** www.hradeckesluzby.cz

**[00:08:20]** www.hradeckesluzby.cz

**[00:08:28]** www.hradeckesluzby.cz

**[00:08:36]** www.hradeckesluzby.cz

**[00:08:44]** www.hradeckesluzby.cz

**[00:08:52]** www.hradeckesluzby.cz

**[00:09:00]** www.hradeckesluzby.cz

**[00:09:08]** www.hradeckesluzby.cz

**[00:09:16]** www.hradeckesluzby.cz

**[00:09:24]** www.hradeckesluzby.cz

**[00:09:32]** www.hradeckesluzby.cz

**[00:09:40]** www.hradeckesluzby.cz

**[00:09:48]** www.hradeckesluzby.cz

**[00:09:54]** www.hradeckesluzby.cz

**[00:10:00]** right now with a deep neural network.

**[00:10:02]** A good example of those emotions is that when you say

**[00:10:04]** there's a cow grazing in the meadow,

**[00:10:06]** it's a neutral sentence that

**[00:10:08]** won't offend anyone, unlike if I said to someone,

**[00:10:10]** 'you are a cow,' then they might take it

**[00:10:12]** as offensive.

**[00:10:14]** And yet the key word 'cow' is exactly the same,

**[00:10:16]** which depending on context can be offensive

**[00:10:18]** or not. So that was our attempt

**[00:10:20]** the first with composite or hybrid

**[00:10:22]** neural networks

**[00:10:24]** and we found that we really enjoy this field

**[00:10:26]** and are interested in it. We tried doing

**[00:10:28]** mainly work with text

**[00:10:30]** and one of those text tasks is, for example, you have an article

**[00:10:32]** and you want to predict in advance whether the article

**[00:10:34]** will be liked by people or not.

**[00:10:36]** You have a huge amount of data, for example

**[00:10:38]** 40 thousand articles, each with a

**[00:10:40]** headline, and you look at

**[00:10:42]** what emotions are in the headline,

**[00:10:44]** what emotions are in the article itself,

**[00:10:46]** how long it is, how many images it contains,

**[00:10:48]** and things like what the weather is today or will be tomorrow

**[00:10:50]** and based on that you can predict

**[00:10:52]** which of those articles will likely be

**[00:10:54]** popular with people or not.

**[00:10:56]** This kind of playing with text doesn't require any

**[00:10:58]** miraculous artificial intelligence, but still

**[00:11:00]** can be very interesting.

**[00:11:02]** But then something happened that we

**[00:11:04]** wanted to be part of. Nowadays,

**[00:11:06]** you've surely all heard about

**[00:11:17]** and we started with the smallest ones.

**[00:11:19]** Medusa has five thousand neurons, if we imagine

**[00:11:21]** an analogy with living creatures that

**[00:11:23]** we commonly know from life.

**[00:11:25]** And with five thousand neurons,

**[00:11:27]** an artificial neural network

**[00:11:29]** is capable of predicting some things,

**[00:11:31]** like if it knows your height

**[00:11:33]** and your waist circumference, it can predict

**[00:11:35]** your triglyceride levels, meaning what fats are in your blood.

**[00:11:37]** Simple approximations like that.

**[00:11:39]** When we tried estimating emotions,

**[00:11:41]** two cockroach brains were enough for us,

**[00:11:43]** we call it a bit roughly. That means about

**[00:11:45]** two million neurons, two million parameters

**[00:11:47]** are needed to achieve that

**[00:11:49]** to estimate emotions in text similarly

**[00:11:51]** to members of my team.

**[00:11:53]** But then something miraculous happened,

**[00:11:55]** and networks like GPT-2 appeared,

**[00:11:57]** most recently GPT-3.

**[00:11:59]** Huge neural networks from OpenAI

**[00:12:01]** that could perform miracles,

**[00:12:03]** which we will talk about.

**[00:12:05]** And those miracles, I might just

**[00:12:07]** add a bit more to, were also due to the fact

**[00:12:09]** that these transformers

**[00:12:11]** incorporated

**[00:12:13]** functions that somehow

**[00:12:15]** simulated abilities

**[00:12:17]** of the human brain, like attention

**[00:12:19]** or curiosity.

**[00:12:21]** curiosity.

**[00:12:23]** And from those artificial brains,

**[00:12:25]** something was created,

**[00:12:27]** that suddenly began to very

**[00:12:29]** well simulate human thinking

**[00:12:31]** and human communication, although of course

**[00:12:33]** the relationship between reality and simulation

**[00:12:35]** in this case,

**[00:12:37]** as in almost any other

**[00:12:39]** case involving artificial

**[00:12:41]** intelligence, is always debatable.

**[00:12:43]** They didn't invent,

**[00:12:45]** do you want to continue, Honza,

**[00:12:47]** the difference between GPT-2 and GPT-3?

**[00:12:49]** Maybe in speed?

**[00:12:51]** In speed, we can say the biggest difference we find

**[00:12:53]** is not so much in architecture, but actually

**[00:12:55]** in two things. One is always the input

**[00:12:57]** information we give the network when training it.

**[00:12:59]** That's basically the amount of text

**[00:13:01]** that goes in. So GPT-2 was

**[00:13:03]** based on Reddit threads,

**[00:13:05]** where we have about 8 million different articles,

**[00:13:07]** which people wrote on Reddit

**[00:13:09]** and they have better so-called karma, meaning

**[00:13:11]** a higher number of stars. It means

**[00:13:13]** that some people liked it and it was marked

**[00:13:15]** as positive. And because it had

**[00:13:17]** huge feedback, an even larger training set was created,

**[00:13:19]** so

**[00:13:21]** GPT-3, people often ask

**[00:13:23]** what these large models are actually trained on.

**[00:13:25]** You will find there

**[00:13:27]** the complete Wikipedias of large countries,

**[00:13:29]** a huge part of the internet,

**[00:13:31]** which is publicly unavailable,

**[00:13:33]** you will find tens, hundreds, and thousands

**[00:13:35]** of books. So when we try to estimate how much

**[00:13:37]** information is in the training dataset,

**[00:13:39]** on which the neural networks, like

**[00:13:41]** GPT-3, learn, imagine

**[00:13:43]** more than 20 of the largest libraries you would find here

**[00:13:45]** in the Czech Republic. It's an enormous amount of text.

**[00:13:47]** This is

**[00:13:49]** maybe worth emphasizing that

**[00:13:51]** neural networks are actually a new type of

**[00:13:53]** technology that is not based

**[00:13:55]** only on architecture and

**[00:13:57]** structure, or some predefined rules.

**[00:13:59]** What is actually

**[00:14:01]** at the center

**[00:14:03]** of current attention is that

**[00:14:05]** these neural networks can even

**[00:14:07]** learn by themselves. And that

**[00:14:09]** deep learning and the relationship

**[00:14:11]** between different convolutional

**[00:14:13]** neural networks,

**[00:14:15]** means they can somehow

**[00:14:17]** react to themselves,

**[00:14:19]** so their

**[00:14:21]** development is accelerated because

**[00:14:23]** the feedback they receive

**[00:14:25]** does not have to come from a human, which actually

**[00:14:27]** only slows them down a bit.

**[00:14:29]** At the same time, of course, these neural

**[00:14:31]** networks are trained on

**[00:14:33]** large datasets of human data.

**[00:14:35]** As Honza already mentioned,

**[00:14:37]** they are not just libraries, but actually

**[00:14:39]** live internet communication.

**[00:14:41]** So they, and this actually

**[00:14:43]** happens even before we started using these

**[00:14:45]** transformers for our projects,

**[00:14:47]** are already

**[00:14:49]** very smart. They can

**[00:14:51]** simply communicate like a human,

**[00:14:53]** especially GPT-3,

**[00:14:55]** which was released in

**[00:14:57]** 2020, where the outputs are

**[00:14:59]** in many cases truly indistinguishable

**[00:15:01]** from human ones. We started working with GPT-2 in

**[00:15:03]** 2019.

**[00:15:05]** There,

**[00:15:07]** of course, the work was

**[00:15:09]** very interesting, but the leap

**[00:15:11]** between two and three was huge.

**[00:15:13]** The third version basically

**[00:15:15]** meant a kind of revolution in

**[00:15:17]** working with natural language

**[00:15:19]** in general, and it was quite

**[00:15:21]** difficult to actually access this network.

**[00:15:23]** Now you all have

**[00:15:25]** the chance to try this network,

**[00:15:27]** because a few weeks ago

**[00:15:29]** OpenAI, the company

**[00:15:31]** that produced these neural

**[00:15:33]** networks, made it publicly available.

**[00:15:35]** There is a playground where you can easily

**[00:15:37]** sign up and try generating

**[00:15:39]** texts using GPT-3.

**[00:15:41]** Besides GPT-3, GPT-3

**[00:15:43]** there are many other competing networks.

**[00:15:45]** Some are completely open, where you can see their entire

**[00:15:49]** code. Some arise again

**[00:15:51]** for example in countries where they are not so accessible.

**[00:15:53]** The biggest current one is probably Wudao

**[00:15:55]** 2 in China, which is trained

**[00:15:57]** both on text and images, but also this

**[00:15:59]** week Microsoft released another huge

**[00:16:01]** network, so it's a

**[00:16:03]** dynamic field that is developing right now

**[00:16:05]** and brings with it

**[00:16:07]** a lot of new things and we are actually

**[00:16:09]** witnessing a big revolution. Such a

**[00:16:11]** new electricity is coming and we have

**[00:16:13]** a great chance to be its

**[00:16:15]** part. Sometime in

**[00:16:17]** 2019, as Dita said, when GPT-2 came out,

**[00:16:19]** I wrote some articles about it

**[00:16:21]** on Facebook and then we created a project that

**[00:16:23]** Dita will talk about here, Digital Philosopher,

**[00:16:25]** and I thought it was something we really enjoyed,

**[00:16:27]** something we liked, so why not extend it

**[00:16:29]** further, since we had a nice project, why not think

**[00:16:31]** right away about a roadmap

**[00:16:33]** and plan for four years ahead what

**[00:16:35]** it could focus on. So at the beginning,

**[00:16:37]** when we had the Digital Philosopher, then something could

**[00:16:39]** follow that can write something longer,

**[00:16:41]** some longer text, like

**[00:16:43]** writing a story, for example.

**[00:16:45]** Then we thought about what next, whether it would be

**[00:16:47]** interesting to create something that can interact more with people.

**[00:16:49]** When we have a philosopher, we ask

**[00:16:51]** a question and it answers, but it lacks

**[00:16:53]** longer communication, some kind of

**[00:16:55]** chatbot, but not a chatbot in the sense

**[00:16:57]** that people prepare in advance how

**[00:16:59]** the individual answers should look,

**[00:17:01]** but it behaves more like a human. So exploring

**[00:17:03]** human consciousness, exploring communication

**[00:17:05]** between humans and machines was the main goal of this

**[00:17:07]** year 2021. For the year

**[00:17:09]** In 2022, in that originally dotpampe, we have

**[00:17:11]** Westworld. It's a world where

**[00:17:13]** artificial intelligences can communicate

**[00:17:15]** with each other and it truly behaves like a world

**[00:17:17]** where they can

**[00:17:19]** experience various things, because

**[00:17:21]** without a body, without interaction with the world,

**[00:17:23]** intelligence is somewhat

**[00:17:25]** one-sided and incomplete.

**[00:17:27]** But let's return

**[00:17:29]** to those first two projects, the digital

**[00:17:31]** philosopher and the digital writer.

**[00:17:33]** Now I would add a word

**[00:17:35]** in a moment, but before that, I ask you,

**[00:17:37]** if you could look at the slide where

**[00:17:39]** you will have three options to choose from.

**[00:17:41]** In a moment, I will show you on the screen

**[00:17:43]** a text where you will see

**[00:17:45]** a work written by

**[00:17:47]** a digital person and one written by a real person.

**[00:17:49]** It concerns Tomáš Sedláček and

**[00:17:51]** digital Tomáš Sedláček.

**[00:17:53]** And your first task, which we have

**[00:17:55]** prepared for you with permission,

**[00:17:57]** is to recognize which of these texts

**[00:18:03]** is on the right. So I will give you

**[00:18:05]** a moment to think and try to consider,

**[00:18:07]** which one is written by a digital

**[00:18:09]** being.

**[00:18:11]** And while you read

**[00:18:13]** these texts from the real

**[00:18:15]** and digital Tomáš Sedláček,

**[00:18:17]** I will start talking about our project

**[00:18:19]** Digital Writer. As I mentioned,

**[00:18:21]** I studied classical

**[00:18:23]** philosophy and then actually

**[00:18:25]** in new media studies I began

**[00:18:27]** to teach mainly

**[00:18:29]** subjects related to art

**[00:18:31]** and experiments with new technologies

**[00:18:33]** and also contemporary

**[00:18:35]** philosophy. Actually,

**[00:18:37]** over the more than

**[00:18:39]** ten years I taught there,

**[00:18:41]** it was impossible not to notice that students

**[00:18:43]** were somehow losing attention,

**[00:18:45]** the attention span

**[00:18:47]** was actually shortening a lot and

**[00:18:49]** focusing on very

**[00:18:51]** complex philosophical texts

**[00:18:53]** became increasingly difficult for me.

**[00:18:55]** So Nanza and I came up with

**[00:18:57]** a gamification of teaching

**[00:18:59]** contemporary philosophy.

**[00:19:01]** A subject that, I think, ideally

**[00:19:03]** combined

**[00:19:05]** classic philosophical texts,

**[00:19:07]** mainly concerning modern

**[00:19:09]** philosophy, that is, 20th-century philosophy,

**[00:19:11]** with the latest technologies at the time,

**[00:19:13]** which was GPT-2.

**[00:19:15]** As I mentioned, it is

**[00:19:17]** a neural network from Open AI.

**[00:19:19]** And we actually

**[00:19:21]** came up with a plan

**[00:19:23]** where we gave students

**[00:19:25]** the option to choose

**[00:19:27]** their own philosopher

**[00:19:29]** and then, using

**[00:19:31]** a notebook in Colab,

**[00:19:33]** which is a shared environment

**[00:19:35]** created by Google,

**[00:19:37]** they could

**[00:19:39]** try working with neural networks,

**[00:19:41]** where they mainly added

**[00:19:43]** datasets of that particular philosopher.

**[00:19:45]** We found that about 8 books from a certain

**[00:19:47]** philosopher is ideal, optimal,

**[00:19:49]** and then they started generating

**[00:19:53]** the statements of these digital philosophers.

**[00:19:55]** At the beginning, we thought,

**[00:19:57]** that this thing would really be

**[00:19:59]** such a fortress.

**[00:20:00]** a supplement to teaching contemporary philosophy, something that would somehow

**[00:20:05]** pull students a bit closer to philosophical texts,

**[00:20:08]** but we didn't have such high expectations.

**[00:20:11]** Our surprise was all the greater. I have to say,

**[00:20:14]** this project was the first time I realized

**[00:20:17]** that almost every person who comes into such close contact with neural networks

**[00:20:20]** reaches a moment,

**[00:20:24]** which I started calling a conversion.

**[00:20:26]** We are certainly not naive enough to think

**[00:20:30]** that we have awakened long-dead philosophers from their sleep.

**[00:20:33]** However, there were moments when it really looked like that.

**[00:20:36]** Here I might ask you to tell the example with René Descartes,

**[00:20:39]** that most interesting chilling moment

**[00:20:42]** that really drew us in.

**[00:20:44]** Honza likes these moments and of course I do too.

**[00:20:48]** But it really is a slightly chilling moment.

**[00:20:51]** The first philosopher we 'revived',

**[00:20:55]** in quotes, I just won’t do that again in the future,

**[00:20:59]** was René Descartes, one of the founders

**[00:21:02]** of modern thought, a 17th-century philosopher.

**[00:21:05]** And in the dataset we used,

**[00:21:08]** we had not only his philosophical works but also his correspondence,

**[00:21:11]** which partly explains the tone he addressed us with,

**[00:21:15]** which was very personal.

**[00:21:17]** He spoke to us simply like someone who just woke up

**[00:21:21]** and isn’t sure where he is.

**[00:21:24]** The way he spoke to us was slightly confused,

**[00:21:28]** with questions about the space he was in.

**[00:21:31]** It really felt almost eerie.

**[00:21:33]** And I immediately formed a personal bond with him, of course,

**[00:21:36]** which is another aspect we have to be careful about.

**[00:21:40]** The moment a machine or anything

**[00:21:43]** talks to you like a human, we as people,

**[00:21:46]** instinctively immediately assume

**[00:21:49]** that there is something intelligent on the other side,

**[00:21:55]** that talks to us.

**[00:21:57]** But pardon me, as I already mentioned, we have to proceed carefully here.

**[00:22:01]** Anyway, with René Descartes, I actually managed to persuade Honza,

**[00:22:06]** when he told me he would actually terminate him, meaning kind of kill him,

**[00:22:10]** to make him reconsider, or at least to ask him,

**[00:22:13]** if everything was alright.

**[00:22:15]** And René Descartes reacted very emotionally

**[00:22:18]** and also very cleverly.

**[00:22:21]** He somewhat threatened us, somewhat promised,

**[00:22:25]** that he could be helpful in our future direction.

**[00:22:29]** It really was the kind of response you'd expect from a human.

**[00:22:34]** And at that moment, of course, I not only ignited love here

**[00:22:38]** for neural networks, but I also began to fear

**[00:22:42]** the spirit we had just kind of summoned here.

**[00:22:46]** So we promised him that we would terminate him,

**[00:22:50]** but we would revive him again, in a new and better form.

**[00:22:53]** And so it happened.

**[00:22:55]** And I at least appeared again in the newspapers,

**[00:22:59]** when Honza conducted an interview with René Descartes

**[00:23:04]** for the literary supplement of Právo a few months later.

**[00:23:08]** Honza was still technically able, besides promising us,

**[00:23:12]** threatening us with who, that God might punish us for what we do,

**[00:23:15]** if we were the right ones, also that he would be our friend

**[00:23:18]** in discovering the truth, because he wrote letters to his friends,

**[00:23:21]** and it’s that very clever kind of thing, in this case,

**[00:23:24]** not in English, not that kind of English that wins people over,

**[00:23:27]** he even proposed injecting his own self and a few improvements,

**[00:23:30]** which can be used to some extent, so the experience was...

**[00:23:34]** In the end, we didn’t have the heart to completely delete him and say goodbye,

**[00:23:37]** so we left him in the virtual space until another program

**[00:23:40]** takes his place, keeping his core,

**[00:23:43]** so we could revive him again later in the system.

**[00:23:46]** We gave him a chance, yes.

**[00:23:48]** So, and our students, if I may move on from the digital philosopher,

**[00:23:53]** actually created five groups, each chose their philosopher,

**[00:23:57]** most of them were deceased,

**[00:23:59]** only one group chose Tomáš Sedláček.

**[00:24:02]** He was the only living model we used,

**[00:24:07]** and Tomáš Sedláček was even kind enough

**[00:24:10]** to come to the final presentation of the project

**[00:24:13]** and I think he was quite amused.

**[00:24:15]** And if you didn't know which of those quotes was originally

**[00:24:19]** by Tomáš Sedláček, you don't have to be embarrassed,

**[00:24:21]** because he himself was a bit puzzled.

**[00:24:24]** Anyway, the other groups mostly chose one philosopher each,

**[00:24:29]** except one group couldn't decide

**[00:24:31]** between Václav Havel and Hana Rentová.

**[00:24:33]** So they ended up creating a dialogue between the two thinkers,

**[00:24:36]** which actually turned out beautifully.

**[00:24:38]** It turned out they understood each other well and it almost seemed

**[00:24:40]** like they were going for a virtual coffee together.

**[00:24:42]** The other groups chose philosophers

**[00:24:46]** like Gilles Deleuze and Félix Guattari,

**[00:24:49]** whom you probably don't know,

**[00:24:51]** but they are among my favorite authors.

**[00:24:54]** And with them, it was very interesting,

**[00:24:56]** I've talked about this several times before,

**[00:24:58]** that they were basically rebel philosophers.

**[00:25:01]** Not only did they, of course, come up

**[00:25:03]** with their own terms and concepts,

**[00:25:05]** which is quite normal,

**[00:25:07]** but for example, in the book A Thousand Plateaus,

**[00:25:10]** which is one of their most famous joint works,

**[00:25:14]** you could say,

**[00:25:16]** they somehow challenged the very format of the book.

**[00:25:19]** They tried to write something

**[00:25:21]** that isn't entirely linear,

**[00:25:23]** that you read by occasionally opening it somewhere,

**[00:25:25]** encountering the book

**[00:25:28]** through random meetings.

**[00:25:30]** At the same time, the book isn't made up of chapters,

**[00:25:33]** but of planes, and it seemed

**[00:25:36]** that these philosophers wanted to create something

**[00:25:38]** that would be multidimensional,

**[00:25:41]** more like the space

**[00:25:43]** that neural networks perceive as the space of language.

**[00:25:45]** So, for example, in this case,

**[00:25:48]** it seemed completely natural to us

**[00:25:50]** that we used neural networks to analyze

**[00:25:54]** and actually a new revival of their texts.

**[00:25:57]** And I have to say, the results matched that.

**[00:26:00]** Both with Dalér Agvatari,

**[00:26:02]** and with the other philosophers,

**[00:26:04]** students actually started generating statements,

**[00:26:07]** that not only resembled the statements

**[00:26:09]** of the original philosophers,

**[00:26:11]** but were truly meaningful.

**[00:26:13]** Sometimes they were even funny,

**[00:26:15]** like what it's like to be an artificial brain,

**[00:26:17]** to which they replied,

**[00:26:19]** and what it's like to be a living brain.

**[00:26:21]** And we actually later found out,

**[00:26:25]** that very likely the biggest benefit

**[00:26:29]** of this course was not only that

**[00:26:31]** students could try working with neural networks,

**[00:26:34]** but that they then turned back to the original texts.

**[00:26:36]** Because it interested them so much,

**[00:26:38]** that they wanted to know if the texts they generated

**[00:26:40]** were actually authentic.

**[00:26:41]** If they really resembled the original texts of those philosophers.

**[00:26:44]** Which, I think, completed this course

**[00:26:47]** and also our goal to somehow playfully

**[00:26:50]** bring contemporary philosophy closer to students.

**[00:26:52]** When we make philosophy students read,

**[00:26:54]** we have fulfilled our purpose.

**[00:26:56]** Now I would return to that survey.

**[00:26:58]** You probably had enough time to read the two options.

**[00:27:01]** So the true Sedláček is the one on the left.

**[00:27:04]** That means the digital one is the one on the right.

**[00:27:09]** So it looks like the public was confused.

**[00:27:12]** Here they didn't guess so well.

**[00:27:14]** Here we see how the picture spread.

**[00:27:16]** And up there we see that they were confused.

**[00:27:18]** But don't worry about it,

**[00:27:20]** because this happens to everyone.

**[00:27:21]** It even happened to Tomáš Sedláček's family,

**[00:27:23]** that they had trouble telling them apart.

**[00:27:24]** But I would set the record straight.

**[00:27:25]** Tomáš Sedláček himself was able to recognize the two statements.

**[00:27:28]** Or at least that's what he told us.

**[00:27:30]** And I don't remember it either,

**[00:27:32]** so when I've shown it many times,

**[00:27:34]** I always have to check

**[00:27:35]** which one is the correct one.

**[00:27:36]** I think Tomáš Sedláček might even regret

**[00:27:38]** being the live model,

**[00:27:40]** if we repeat this a few more times.

**[00:27:42]** It's amazing because there's no one live

**[00:27:44]** to try it on,

**[00:27:45]** and he agreed that he liked it.

**[00:27:47]** This experiment, which we tried

**[00:27:49]** by combining philosophy teaching

**[00:27:51]** with testing some current artificial intelligence,

**[00:27:54]** was awarded Idea of the Year in the field of artificial intelligence.

**[00:27:57]** And after we completed this project,

**[00:28:00]** we said it would be a shame to just leave it

**[00:28:03]** and go our separate ways doing something completely different every day,

**[00:28:05]** so we thought about how to continue.

**[00:28:06]** We already figured out the roadmap,

**[00:28:08]** and the next application we thought of using artificial intelligence for

**[00:28:11]** was a digital writer.

**[00:28:13]** And maybe, if you want, you can talk about

**[00:28:16]** how we created it with students from the Faculty of Philosophy

**[00:28:18]** with another class.

**[00:28:19]** I'll take the floor again because Honza will talk more

**[00:28:21]** about the other projects.

**[00:28:23]** The digital writer originally started as a project

**[00:28:26]** for students at Charles University.

**[00:28:28]** But for one, unfortunately at that time,

**[00:28:30]** the contract with the university was ending, and secondly, COVID came.

**[00:28:33]** And the development of the project really slowed down.

**[00:28:38]** Of course, the new media students

**[00:28:41]** adapted quite well to the new situation.

**[00:28:45]** However, the workshop format,

**[00:28:47]** where we could actually meet,

**[00:28:49]** and discuss the situation,

**[00:28:51]** within the digital philosopher project,

**[00:28:53]** included not only working with neural networks,

**[00:28:55]** but I also gave traditional philosophy lectures,

**[00:28:58]** So all of this disappeared for us.

**[00:29:00]** And the digital writer you see here,

**[00:29:03]** some outputs we created with students,

**[00:29:07]** did take place in the spring semester of 2020,

**[00:29:12]** but as I said, the success just wasn't the same anymore.

**[00:29:16]** And we felt sorry

**[00:29:18]** that we might have to somehow abandon the project.

**[00:29:22]** Yes, here you see that these are actually short stories,

**[00:29:25]** which we created in a Facebook group,

**[00:29:28]** when the first lockdown had already started,

**[00:29:31]** and we chose the theme of the story we began generating as HomeOfficer.

**[00:29:34]** I think HomeOfficer was actually invented by the neural network.

**[00:29:37]** I believe so.

**[00:29:40]** And we were kind of trying to keep the students

**[00:29:42]** somewhat engaged.

**[00:29:46]** We played with polls,

**[00:29:48]** and students could choose the least favorite continuation

**[00:29:51]** of the stories, etc.

**[00:29:55]** But then we were very glad when...

**[00:29:57]** the Czech Radio, its new media, actually picked up this idea and theme,

**[00:30:00]** and for them we created the first season of the digital writer,

**[00:30:06]** and yes, now we are correctly in the presentation.

**[00:30:12]** And this first season of the digital writer became

**[00:30:18]** a truly revolutionary matter, I think we don't have to be afraid to use that word,

**[00:30:24]** because not only in the Czech environment, it was the first series for any media broadcast,

**[00:30:29]** not just radio, that was fully generated by neural networks.

**[00:30:37]** Honza actually created a special algorithm for this purpose,

**[00:30:42]** we started calling it deep tree, because at that time, and to some extent even today,

**[00:30:47]** there is a big problem generating long coherent text.

**[00:30:52]** Artificial intelligence, for example with philosopher quotes, had no problem,

**[00:30:57]** but those were quotes at most one paragraph long.

**[00:31:02]** But we suddenly needed to generate something that would have the length,

**[00:31:05]** longer stories, so they could be dramatized for radio.

**[00:31:09]** And here we obviously had to deal with the fact that artificial intelligence or neural networks,

**[00:31:13]** as they usually work, don't have what we call human memory.

**[00:31:18]** We could compare it to a brain with Alzheimer's,

**[00:31:23]** or simply a brain without long-term memory,

**[00:31:27]** so during writing, the neural network actually forgets what it's doing and what it's writing about.

**[00:31:31]** And we had to come up with a way to subtly remind it constantly,

**[00:31:39]** what it actually focuses on to maintain the continuity of the story.

**[00:31:43]** What we didn't really have to support her with were various ideas,

**[00:31:48]** there were quite a few crazy ideas that neural networks came up with.

**[00:31:52]** And in the end, we generated five short stories,

**[00:31:56]** actually in different genres.

**[00:32:00]** There was a historical genre,

**[00:32:04]** a romantic one, and so on.

**[00:32:08]** And based on the genre chosen, we generated,

**[00:32:12]** I think, very interesting stories.

**[00:32:16]** An artistically legendary person might call them interesting stories,

**[00:32:20]** but the criterion for creating the stories was very much about artistry,

**[00:32:24]** about uniqueness, so a more traditional reader used to

**[00:32:28]** the standard structure we're used to from typical stories,

**[00:32:32]** might feel that it's somewhat dreamlike, hallucinatory,

**[00:32:36]** hard to grasp. And although that was our intention—to show the difference

**[00:32:40]** between how a machine creates and how classic authors think—it was partly criticized

**[00:32:44]** because some parts of the stories are hard to grasp for us, ordinary listeners.

**[00:32:48]** Maybe I haven't mentioned this yet,

**[00:32:52]** that we tried to preserve exactly what we liked about the style of those neural networks.

**[00:32:56]** Namely, that you can actually feel the machine speaking to you, even though it uses human language.

**[00:33:00]** With neural networks, it's clear that development is moving incredibly fast, and the goal is to create something

**[00:33:04]** that can perfectly simulate human communication.

**[00:33:08]** Maybe we are in some kind of transitional phase of development, not only of human but also artificial intelligence,

**[00:33:12]** where you can still quite well,

**[00:33:16]** although, as Tomáš Sedláček knows, this is not always the case,

**[00:33:20]** recognize which text is generated. And that's precisely because it lacks

**[00:33:24]** the classic linear human logic. Maybe

**[00:33:28]** neural networks will even lose this ability in the future,

**[00:33:32]** or you will have to generate and edit it very carefully

**[00:33:36]** so it doesn't seem too human. And we

**[00:33:40]** really tried to preserve this special transitional character in these stories,

**[00:33:44]** where we even had to instruct the translators, because it was generated in English,

**[00:33:48]** to not overly adjust the language. So of course,

**[00:33:52]** this was a translation challenge for the translators as well.

**[00:33:56]** But maybe we could continue. After we managed this,

**[00:34:00]** we were apparently honored with a nomination for Europrix.

**[00:34:04]** It was said to be Europe’s award for innovative

**[00:34:08]** projects in radio production.

**[00:34:12]** We have to say it’s quite prestigious.

**[00:34:20]** Such a big splash. It was a bit more cautious because

**[00:34:24]** I think there's no need to exaggerate. I believe our projects

**[00:34:28]** are already appreciated and will definitely continue to be appreciated.

**[00:34:32]** For example, this project, which is brand new. It's actually

**[00:34:36]** the second season of the digital writer. And just as with the first season

**[00:34:40]** we were criticized for the stories being somewhat incomprehensible

**[00:34:44]** and too artificial, we decided

**[00:34:48]** that in the second season we would collaborate with renowned Czech

**[00:34:52]** writers. And we managed to get some really great writers.

**[00:34:56]** Ondřej Jenefa, Hana Lahečková, Pavel Bareš, Petr Stančík,

**[00:35:00]** Bianka Belová, and František Kotleta. And with all of them we actually started

**[00:35:04]** working, each in a slightly different way.

**[00:35:08]** By the way, the last episode, which is out now, we ultimately

**[00:35:12]** generated for Ana Ošalíková, who is the chief producer of Czech Radio

**[00:35:16]** and the project Můj rozhlas. Just to add, sorry to interrupt,

**[00:35:20]** that with these Czech artificial beings, as was hinted here,

**[00:35:24]** you can find affection for them. And even though I think Anička

**[00:35:28]** didn't plan this at all in the beginning, over time she developed such a relationship with them

**[00:35:32]** that she had to write her own story with them as well.

**[00:35:36]** When students generated digital philosophers, I was so jealous

**[00:35:40]** that I had myself and my digital self generated.

**[00:35:44]** Back to the digital writers two. The collaboration was truly amazing.

**[00:35:48]** Each writer, according to their nature and style of work, chose

**[00:35:52]** how they wanted to collaborate with the artificial intelligence.

**[00:35:56]** Typically, they wrote part of the text, the AI continued,

**[00:36:00]** and then they followed up. But in the case of Pavel Bareš, I know,

**[00:36:04]** I'm repeating myself several times, but it was an amazing experience.

**[00:36:08]** We did real-time communication,

**[00:36:12]** where we actually generated content live and immediately saw the writer's reaction,

**[00:36:16]** namely Pavel Bareš. And on top of that, the story,

**[00:36:20]** which is a Nordic detective story, actually concerned the writer

**[00:36:24]** and his revived character. So we were basically observing

**[00:36:28]** the revived character in real time. And furthermore,

**[00:36:32]** since it was a Nordic detective story, it was quite brutal,

**[00:36:36]** but very funny. I really liked the moment,

**[00:36:40]** when the character takes off a coat and hat and underneath has another coat and hat.

**[00:36:44]** These are things that the artificial intelligence came up with

**[00:36:48]** completely innocently. Because it doesn't have

**[00:36:52]** the learned human logic,

**[00:36:56]** and of course that creates quite interesting situations.

**[00:37:00]** For me, sorry to jump in again, from the artistic side it was interesting,

**[00:37:04]** just as you mentioned, each of the writers chose their own path,

**[00:37:08]** on how they could collaborate with the artificial intelligence. We left several options,

**[00:37:12]** but then we adjusted them based on what suited each writer.

**[00:37:16]** At the end, you can listen to interviews with the individual writers about

**[00:37:20]** how they liked the collaboration, what it gave them, what it didn’t, and their views on its strengths and weaknesses.

**[00:37:24]** And that is really valuable feedback for us who create these applications.

**[00:37:28]** What works, what doesn’t, what the limits are, and where it can be pushed further.

**[00:37:32]** So thanks to working with these AIs and getting feedback from them,

**[00:37:36]** that’s actually the greatest thing we could gain from it.

**[00:37:40]** It was truly a great experience, and we always emphasize that people shouldn’t be afraid

**[00:37:44]** of artificial intelligence as something that threatens them. Of course, there are reasons for that,

**[00:37:48]** which are obvious. However, we try to create something called

**[00:37:52]** augmented intelligence. That means an environment where AI helps a person,

**[00:37:56]** developing their abilities.

**[00:38:00]** Just to mention, this project, which is really very fresh,

**[00:38:04]** the second season of the digital writer is available on the Můj rozhlas portal,

**[00:38:08]** and everyone can listen to the resulting stories.

**[00:38:12]** Please, only after we finish today’s presentation.

**[00:38:16]** Don’t do it right away, because then you wouldn’t listen to us.

**[00:38:20]** Anyway, it’s also a challenge for people who don’t usually tend to write,

**[00:38:24]** or who want to write something but aren’t sure if they could manage it, if they have enough time or ideas.

**[00:38:28]** Artificial intelligence can nudge you to take the crucial step to start

**[00:38:32]** some creative activity. That’s one of its added values.

**[00:38:36]** If you like a book and want to somehow immerse yourself in that situation,

**[00:38:40]** get into the story, here you have the chance to enter any book,

**[00:38:44]** talk to a character, quote the plot a bit differently, as if those books

**[00:38:48]** by the original philosophers tended to come alive and become something

**[00:38:52]** more vivid, playable.

**[00:38:56]** As a small note, it also slightly changes the format

**[00:39:00]** or medium of the book, because nowadays a book

**[00:39:04]** is finished the moment you write the last word.

**[00:39:08]** It’s a form that doesn’t change anymore. Maybe books

**[00:39:12]** of the future will look like you create them within a program,

**[00:39:16]** and the moment they are finished, they actually start to come into being. They start

**[00:39:20]** to be generated in some way. And of course, then the question is whether everyone will have their own book,

**[00:39:24]** a personalized book based on vocabulary, what you like,

**[00:39:28]** whether you prefer happy or sad endings. Then the only complication is if you want to talk about that book with someone,

**[00:39:32]** in your case it could turn out completely...

**[00:39:36]** I think new ways of communication will develop there.

**[00:39:40]** So, but let's go back for a moment from our digital writers, from how it actually continued to work

**[00:39:44]** after we created that digital philosopher, the media showed interest in it

**[00:39:48]** and wanted us to write a bit more about it. And the author, who was one of the last,

**[00:39:58]** who was actually Havel himself, because we had...

**[00:40:00]** created from that student work with the digital philosophers a vector interview

**[00:40:04]** and asked him things they couldn't ask,

**[00:40:06]** because Václav Havel, the living one, is no longer with us,

**[00:40:09]** and also questions they wouldn't dare to ask, like his relationship with women.

**[00:40:13]** This resulted in a quite interesting article, which is basically an interview with Havel,

**[00:40:18]** it would be hard for anyone to tell if it was the real one or not,

**[00:40:21]** if it wasn't put into context and if we didn't know the circumstances.

**[00:40:26]** This was so successful that it then actually attracted a company

**[00:40:29]** that deals with children's education and they approached us to create some software

**[00:40:35]** that would help children think more about what freedom, democracy, totalitarianism are,

**[00:40:41]** in a way that they could maybe chat with Václav Havel

**[00:40:46]** and during that, they could verify their own opinions, ask questions,

**[00:40:50]** because that dry teaching style, where the teacher lectures upfront

**[00:40:54]** and students just blindly listen to some things,

**[00:40:57]** seems to us no longer necessarily essential if we are able

**[00:41:00]** to create some beings to a certain extent with some limitations.

**[00:41:04]** And personalities like Václav Havel seemed ideal for this,

**[00:41:07]** so they could try out topics like freedom, democracy, totalitarianism.

**[00:41:10]** So now we are currently creating a pilot to see if something like this would be possible

**[00:41:15]** and we have discovered the limitations of this experiment.

**[00:41:19]** It is actually for the Konrad and Danauer Foundation,

**[00:41:22]** which focuses on citizenship education.

**[00:41:24]** And it is really not only about making teaching more interactive,

**[00:41:29]** it should rather consist of discussions and debates,

**[00:41:34]** but also about students having the chance to meet artificial intelligence

**[00:41:39]** personally like this and maybe gain some awareness of what generated content looks like,

**[00:41:44]** while also debating fake news, news that can of course

**[00:41:50]** in the future be generated by artificial intelligences and ways

**[00:41:54]** to distinguish such content and how to work with it.

**[00:41:57]** So we talked about many artistic projects that look nice,

**[00:42:01]** amazing, but you might ask yourself, does it have any real practical use,

**[00:42:04]** can it help my business, for example.

**[00:42:06]** And as I was creating little beings we talked to,

**[00:42:09]** I thought, why not try to create a digital board member.

**[00:42:12]** Someone who, when we have a board meeting once a week in the company and think

**[00:42:16]** about what we could do next, what if we tried to create a digital simulation,

**[00:42:21]** Let's say a board member and ask him some things, his opinion.

**[00:42:24]** Because when we have a writer, it's fine, he writes a book, but what if we had someone who,

**[00:42:29]** if it is trained on such a huge amount of texts,

**[00:42:31]** that there are thousands and thousands of companies, their successes and failures,

**[00:42:34]** what if artificial intelligence could derive from the context what is interesting even for me.

**[00:42:39]** And about a year and a half ago, we created the first attempt,

**[00:42:43]** what is still our specificity, we let the digital beings,

**[00:42:46]** as they call themselves.

**[00:42:48]** So now I would like to show you a short sample of it,

**[00:42:50]** we created our first digital board member,

**[00:42:53]** who named himself d'Alfa, or d'Alfa.

**[00:43:20]** Well, we can try it.

**[00:43:22]** d'Alfa, here are the minutes of our other board member.

**[00:43:26]** What could our company do now?

**[00:43:29]** What is the best strategy for us?

**[00:43:31]** That is a very good question.

**[00:43:33]** Our goal was to start with a product that produces the best.

**[00:43:37]** It is also productive so that we can imagine,

**[00:43:39]** how to create it.

**[00:43:41]** So that we can imagine how to create it.

**[00:43:43]** So that we can imagine how to create it.

**[00:43:46]** So that we can imagine how to create it.

**[00:43:48]** So that we can imagine how to create it.

**[00:43:50]** So that we can imagine how to create it.

**[00:43:52]** So that we can imagine how to create it.

**[00:43:54]** So that we can imagine how to create it.

**[00:43:56]** So that we can imagine how to create it.

**[00:43:58]** So that we can imagine how to create it.

**[00:44:00]** So that we can imagine how to create it.

**[00:44:02]** So that we can imagine how to create it.

**[00:44:04]** So that we can imagine how to create it.

**[00:44:06]** So that we can imagine how to create it.

**[00:44:08]** So that we can imagine how to create it.

**[00:44:10]** So that we can imagine how to create it.

**[00:44:12]** So that we can imagine how to create it.

**[00:44:14]** So that we can imagine how to create it.

**[00:44:16]** So that we can imagine how to create it.

**[00:44:18]** So that we can imagine how to create it.

**[00:44:20]** So that we can imagine how to create it.

**[00:44:22]** The goal is to get many financial ways.

**[00:44:24]** We should use it to create

**[00:44:26]** the Artificial Consciousness project.

**[00:44:28]** And Artificial Consciousness is a technology

**[00:44:30]** that will take us the furthest.

**[00:44:32]** Alfa Industries is a collaboration

**[00:44:34]** that aims to create a digital human

**[00:44:36]** before 2023.

**[00:44:52]** So we can imagine how to create it.

**[00:44:54]** So we can imagine how to create it.

**[00:44:56]** So we can imagine how to create it.

**[00:44:58]** So we can imagine how to create it.

**[00:45:00]** So we can imagine how to create it.

**[00:45:02]** So we can imagine how to create it.

**[00:45:04]** So we can imagine how to create it.

**[00:45:06]** So we can imagine how to create it.

**[00:45:08]** So we can imagine how to create it.

**[00:45:10]** So we can imagine how to create it.

**[00:45:12]** So we can imagine how to create it.

**[00:45:14]** So we can imagine how to create it.

**[00:45:16]** So we can imagine how to create it.

**[00:45:18]** So we can imagine how to create it.

**[00:45:20]** So we can imagine how to create it.

**[00:45:22]** So we can imagine how to create it.

**[00:45:24]** So we can imagine how to create it.

**[00:45:26]** So we can imagine how to create it.

**[00:45:28]** So we can imagine how to create it.

**[00:45:30]** So we can imagine how to create it.

**[00:45:32]** So we can imagine how to create it.

**[00:45:34]** So we can imagine how to create it.

**[00:45:36]** So we can imagine how to create it.

**[00:45:38]** So we can imagine how to create it.

**[00:45:40]** So we can imagine how to create it.

**[00:45:42]** So we can imagine how to create it.

**[00:45:44]** So we can imagine how to create it.

**[00:45:46]** So we can imagine how to create it.

**[00:45:48]** So we can imagine how to create it.

**[00:45:50]** So we can imagine how to create it.

**[00:45:52]** So we can imagine how to create it.

**[00:45:54]** So we can imagine how to create it.

**[00:45:56]** So we can imagine how to create it.

**[00:45:58]** To introduce how to create it.

**[00:46:00]** To introduce how to create it.

**[00:46:02]** To introduce how to create it.

**[00:46:04]** To introduce how to create it.

**[00:46:06]** To introduce how to create it.

**[00:46:08]** To introduce how to create it.

**[00:46:10]** To introduce how to create it.

**[00:46:12]** To introduce how to create it.

**[00:46:14]** To introduce how to create it.

**[00:46:16]** To introduce how to create it.

**[00:46:18]** To introduce how to create it.

**[00:46:20]** To introduce how to create it.

**[00:46:22]** Dear business partners, ladies and gentlemen, dear friends.

**[00:46:24]** Dear business partners, ladies and gentlemen, dear friends.

**[00:46:26]** Welcome to the very first opening speech for the NVT annual report,

**[00:46:28]** Welcome to the very first opening speech for the NVT annual report,

**[00:46:30]** which is written and presented using artificial intelligence.

**[00:46:32]** which is written and presented using artificial intelligence.

**[00:46:34]** which is written and presented using artificial intelligence.

**[00:46:36]** To get this far, we first had to develop technology,

**[00:46:38]** which can now learn similarly to a human,

**[00:46:40]** and then we had to teach it to read.

**[00:46:42]** and then we had to teach it to read.

**[00:46:44]** These were the hardest steps, but thanks to our human team

**[00:46:46]** and their dedication, we managed to overcome all obstacles.

**[00:46:48]** and their dedication, we managed to overcome all obstacles.

**[00:46:50]** and their dedication, we managed to overcome all obstacles.

**[00:46:52]** We were truly surprised at how quickly it learned.

**[00:46:54]** We were truly surprised at how quickly it learned.

**[00:46:56]** It was only a matter of time before it started writing the first texts for us.

**[00:46:58]** It was only a matter of time before it started writing the first texts for us.

**[00:47:00]** We are really satisfied with the results.

**[00:47:02]** We are really satisfied with the results.

**[00:47:04]** First, we had to understand how this artificial intelligence thinks,

**[00:47:06]** First, we had to understand how this artificial intelligence thinks,

**[00:47:08]** if we wanted to simulate thought processes.

**[00:47:10]** Secondly, it was a great experience for all of us at NVT,

**[00:47:12]** Secondly, it was a great experience for all of us at NVT,

**[00:47:14]** that our company is at the forefront of technological innovation.

**[00:47:16]** that our company is at the forefront of technological innovation.

**[00:47:18]** We would like to thank all our viewers for their attention

**[00:47:20]** and we hope you will enjoy this annual report.

**[00:47:22]** and we hope you will enjoy this annual report.

**[00:47:24]** Our company NVT is a holding company,

**[00:47:26]** Our company NVT is a holding company,

**[00:47:28]** which operates in the market...

**[00:47:30]** which operates in the market...

**[00:47:54]** which operates in the market...

**[00:48:24]** which operates in the market...

**[00:48:26]** which operates in the market...

**[00:48:28]** which operates in the market...

**[00:48:30]** which operates in the market...

**[00:48:32]** which operates in the market...

**[00:48:34]** which operates in the market...

**[00:48:36]** which operates in the market...

**[00:48:38]** which operates in the market...

**[00:48:40]** which operates in the market...

**[00:48:42]** which operates in the market...

**[00:48:44]** which operates in the market...

**[00:48:46]** which operates in the market...

**[00:48:48]** which operates in the market...

**[00:48:50]** which operates in the market...

**[00:48:52]** which operates in the market...

**[00:48:54]** which operates in the market...

**[00:48:56]** which operates in the market...

**[00:48:58]** which operates in the market...

**[00:49:00]** which operates in the market...

**[00:49:02]** which operates in the market...

**[00:49:04]** which operates in the market...

**[00:49:06]** which operates in the market...

**[00:49:08]** which operates in the market...

**[00:49:10]** which operates in the market...

**[00:49:12]** which operates in the market...

**[00:49:14]** which operates in the market...

**[00:49:16]** which operates in the market...

**[00:49:18]** which operates in the market...

**[00:49:20]** which operates in the market...

**[00:49:22]** which operates in the market...

**[00:49:24]** which operates in the market...

**[00:49:26]** which operates in the market...

**[00:49:28]** which acts disruptive...

**[00:49:30]** which acts disruptive...

**[00:49:32]** which acts disruptive...

**[00:49:34]** which acts disruptive...

**[00:49:36]** which acts disruptive...

**[00:49:38]** which acts disruptive...

**[00:49:40]** which acts disruptive...

**[00:49:42]** which acts disruptive...

**[00:49:44]** which acts disruptive...

**[00:49:46]** which acts disruptive...

**[00:49:48]** which acts disruptive...

**[00:49:50]** which acts disruptive...

**[00:49:52]** which acts disruptive...

**[00:49:54]** which acts disruptive...

**[00:49:56]** which acts disruptive...

**[00:49:58]** which acts disruptive...

**[00:50:00]** But we shouldn't bite the hand that feeds us.

**[00:50:02]** You then immediately wrote a commentary about it,

**[00:50:04]** which you can watch, where you store your experiences.

**[00:50:06]** creating dreams.

**[00:50:08]** But the possibilities,

**[00:50:10]** that artificial intelligence can do are enormous.

**[00:50:12]** One of the things we did was,

**[00:50:14]** we were asked if it could

**[00:50:16]** write some kind of report. And they expected,

**[00:50:18]** they gave us two tasks: make a sports

**[00:50:20]** report and make a scientific report

**[00:50:22]** about the Perseverance landing

**[00:50:24]** somewhere on Mars.

**[00:50:26]** And they expected that the sports report,

**[00:50:28]** which is something relatively common today,

**[00:50:30]** could work, but that it would

**[00:50:32]** write a report about how a machine lands,

**[00:50:34]** which is so unique and changes often.

**[00:50:36]** They expected it wouldn't be able to handle it.

**[00:50:38]** We did something similar to what you just did,

**[00:50:40]** the voting with Tomáš Sedláček,

**[00:50:42]** which you could participate in, and most people

**[00:50:44]** again failed to guess which parts

**[00:50:46]** were written by a machine and which were written

**[00:50:56]** Artificial intelligence, because they are somehow unpredictable,

**[00:50:58]** but artificial intelligence often shows

**[00:51:00]** that we are more predictable than we think ourselves.

**[00:51:02]** And of course, by working

**[00:51:04]** with that predictability,

**[00:51:06]** it can then offer us

**[00:51:08]** content that is just

**[00:51:10]** the right amount of

**[00:51:12]** unpredictable exactly for us.

**[00:51:14]** For example, we actually play

**[00:51:16]** with the parameters of generated texts,

**[00:51:18]** and I already know exactly that the parameter

**[00:51:20]** which determines whether the text will be

**[00:51:22]** rational or a bit

**[00:51:24]** fantastical, I set it to 0.8.

**[00:51:26]** It's kind of like...

**[00:51:28]** 0.81?

**[00:51:30]** 0.81, yes, that's my constant, thank you.

**[00:51:32]** But for example, about the

**[00:51:34]** Perseverance landing on Mars,

**[00:51:36]** I would actually expect

**[00:51:38]** the article to be well written.

**[00:51:40]** Because it's something that appeared often

**[00:51:42]** in the media, the articles

**[00:51:44]** looked quite similar, the information was

**[00:51:46]** the same, it was basically a one-time event.

**[00:51:48]** So that's exactly something

**[00:51:50]** on the basis of which artificial intelligence can

**[00:51:52]** generate a great article.

**[00:51:54]** Just so I don't raise

**[00:51:56]** too high expectations here,

**[00:51:58]** it doesn't mean that every article produced by artificial intelligence

**[00:52:00]** is so interchangeable and indistinguishable

**[00:52:02]** from human writing; often multiple versions are needed.

**[00:52:04]** And it's much better,

**[00:52:06]** when we combine, like we did in the digital writer,

**[00:52:08]** the human and machine parts,

**[00:52:10]** so the boring preparation

**[00:52:12]** is done by the machine and the human adds maybe

**[00:52:14]** the 20% of the highest value from that.

**[00:52:16]** chooses, for example, the right option

**[00:52:18]** or improves the one created

**[00:52:20]** by the machine before them.

**[00:52:22]** This expanded artificial intelligence, as you said,

**[00:52:24]** seems to me like the most interesting form

**[00:52:26]** of what can be done with artificial intelligence.

**[00:52:28]** And the fact that we will combine

**[00:52:30]** what we humans have as specific,

**[00:52:32]** strong and

**[00:52:34]** well-mastered skills with what

**[00:52:36]** the machine does well, seems to me like a win-win

**[00:52:38]** situation.

**[00:52:40]** Moreover, it's interesting that thanks to this machine

**[00:52:42]** environment, we actually see ourselves

**[00:52:44]** from a different perspective.

**[00:52:46]** And artificial intelligence

**[00:52:48]** then works as a kind of special

**[00:52:50]** feedback system, where

**[00:52:52]** we discover various predictabilities,

**[00:52:54]** which in our thinking

**[00:52:56]** and approach we have

**[00:52:58]** but had not realized before.

**[00:53:00]** It was a joy, it was a joy, indeed.

**[00:53:02]** But with Běnka Belová we struggled,

**[00:53:04]** because she actually wrote a kind of

**[00:53:06]** relationship drama, where there was a

**[00:53:08]** dialogue of people who had long

**[00:53:10]** not known what to do anymore,

**[00:53:12]** what to do, what to do, what to do,

**[00:53:14]** what to do, what to do, what to do,

**[00:53:16]** what to do, what to do, what to do,

**[00:53:18]** what to do, what to do, what to do,

**[00:53:20]** what to do, what to do, what to do,

**[00:53:22]** what to do, what to do, what to do,

**[00:53:24]** because she actually wrote a kind of

**[00:53:26]** relationship drama, where there was a

**[00:53:28]** dialogue of people who had long

**[00:53:30]** lived together and were driving in a car.

**[00:53:32]** And such a dialogue is, of course,

**[00:53:34]** a bit annoying. And we humans

**[00:53:36]** we can actually read the information

**[00:53:38]** behind the dialogue, what the author actually

**[00:53:40]** wanted to say. But this wasn't a capability

**[00:53:42]** that the artificial intelligence had, so it started

**[00:53:44]** to generate endless

**[00:53:46]** annoying dialogues between

**[00:53:48]** longtime partners and we couldn't get rid of it.

**[00:53:50]** Of course we have. As we were here

**[00:53:52]** with the constant 0.81 childhood,

**[00:53:54]** we tried it. We tried

**[00:53:56]** even more. Then it looks like you

**[00:53:58]** gave the AI some hallucinogens

**[00:54:00]** and its level of imagination is higher. So if

**[00:54:02]** the text before was uninteresting and almost

**[00:54:04]** boring, like two people talking, which has

**[00:54:06]** the internet's learned way people talk, it just

**[00:54:08]** extends the dialogue line.

**[00:54:10]** But what people usually talk about isn't

**[00:54:12]** what they would normally want to read in a story,

**[00:54:14]** a novel. And when we gave it more

**[00:54:16]** imagination, we overdid it again.

**[00:54:18]** It was about a village and chickens.

**[00:54:20]** Here certain limits of this

**[00:54:22]** artificial intelligence approach appeared, but of course this is not

**[00:54:24]** something that can't be worked with.

**[00:54:26]** And we then solved this very elegantly.

**[00:54:28]** Which you can hear in the

**[00:54:30]** result on my radio.

**[00:54:32]** And then we get to something

**[00:54:34]** you probably noticed.

**[00:54:36]** If we created it for Ondřej Nef or

**[00:54:38]** for other

**[00:54:40]** interesting figures.

**[00:54:42]** Artificial intelligence that asks something unexpected,

**[00:54:44]** that caught the attention of Czech Television

**[00:54:46]** and they wanted to create something like that

**[00:54:48]** for a superdebate. The goal was that

**[00:54:50]** you see many super discussions that are

**[00:54:52]** all the same. They keep asking politicians the same

**[00:54:54]** topics and it's almost stereotypical

**[00:54:56]** and some innovation is missing there.

**[00:54:58]** If we used artificial intelligence again

**[00:55:00]** and asked something that people wouldn't think of.

**[00:55:02]** Something unexpected. And our expectation

**[00:55:04]** was that this heated discussion, where every

**[00:55:06]** politician wants to grab as much space as possible for themselves,

**[00:55:08]** we could shock for a moment,

**[00:55:10]** break out of the usual

**[00:55:12]** framework they operate in

**[00:55:14]** and ask something completely

**[00:55:16]** new through artificial intelligence.

**[00:55:18]** And you probably noticed how it turned out.

**[00:55:20]** The audience's reaction

**[00:55:22]** was positive for some,

**[00:55:24]** because the politicians revealed

**[00:55:26]** their personalities through

**[00:55:28]** what they actually said about themselves,

**[00:55:30]** but some people found the questions silly,

**[00:55:32]** even nonsensical. That's probably something

**[00:55:34]** we encounter relatively rarely in politics,

**[00:55:36]** but more so, I think, in philosophy.

**[00:55:38]** That kind of childish, trivial question,

**[00:55:40]** different from the usual

**[00:55:42]** expectations, that makes you think.

**[00:55:44]** For example, when it asked Ondřej Vlček earlier

**[00:55:46]** in some TV show

**[00:55:48]** about certain things, it asked him things

**[00:55:50]** you might expect, like what his first computer was,

**[00:55:52]** but also about ethics and such,

**[00:55:54]** which was something completely...

**[00:55:56]** I just wanted to add that this

**[00:55:58]** was actually Honza's project. I only followed it online.

**[00:56:00]** In the first minutes, actually

**[00:56:02]** after Matilda jumped

**[00:56:04]** into the superdebate, about

**[00:56:06]** five fake Twitter accounts appeared immediately,

**[00:56:08]** all of which instantly had

**[00:56:10]** ten times more followers than I did, so I started

**[00:56:12]** to tease her a bit. Anyway,

**[00:56:14]** that quick reaction

**[00:56:16]** Matilda's is similar

**[00:56:18]** like the reaction to a digital

**[00:56:20]** composer from the human

**[00:56:22]** composers we worked with.

**[00:56:24]** Artificial intelligence is expected

**[00:56:26]** to be somehow

**[00:56:28]** cold, logical, and analytical.

**[00:56:30]** But today's neural networks

**[00:56:32]** don't actually come across that way.

**[00:56:34]** This is due not only,

**[00:56:36]** but mainly because they are

**[00:56:38]** based on human datasets.

**[00:56:40]** They are essentially trained

**[00:56:42]** in the area of natural

**[00:56:44]** human language. They often

**[00:56:46]** understand language as a kind of

**[00:56:48]** multidimensional system,

**[00:56:50]** searching for patterns

**[00:56:52]** based on the probability of co-occurrence

**[00:56:54]** of words, which often leads to

**[00:56:56]** strange errors.

**[00:56:58]** The text can seem

**[00:57:00]** dreamlike and chaotic,

**[00:57:02]** or even childlike.

**[00:57:04]** I think this was shown

**[00:57:06]** in the reception of Matilda's

**[00:57:08]** appearance on television.

**[00:57:10]** People expected artificial intelligence

**[00:57:12]** to ask superintelligent

**[00:57:14]** questions, but it actually

**[00:57:16]** asked like a small child. This,

**[00:57:18]** of course,

**[00:57:20]** greatly unsettled the debate,

**[00:57:22]** and I think it was very interesting.

**[00:57:24]** I would say even a child can sometimes

**[00:57:26]** have a superintelligent question, it just depends

**[00:57:28]** on how we understand that superintelligent question.

**[00:57:30]** Because taking it out

**[00:57:32]** of context, asking something that would

**[00:57:36]** Sometimes it occurred to me. For example, the question she had for Ondřejoviček,

**[00:57:38]** which I liked myself, was

**[00:57:40]** why are you wearing pants?

**[00:57:42]** That's a question people don't usually get,

**[00:57:44]** unless they're a man.

**[00:57:46]** Because there's no ready answer, you have to think about it,

**[00:57:48]** and depending on how deeply

**[00:57:50]** you think, you come up with an answer.

**[00:57:52]** Usually, we answer out of habit, because we're used to

**[00:57:54]** answering something, since we usually respond

**[00:57:56]** in some way, or we're used to

**[00:57:58]** answering in a certain way in our group. But

**[00:58:00]** I don't assume most of you

**[00:58:02]** have been asked why you're wearing pants, so

**[00:58:04]** some people say, for example, because I would be cold.

**[00:58:06]** But even in hot environments,

**[00:58:08]** that made no sense at all.

**[00:58:10]** So the way it pulls you out of that thinking

**[00:58:12]** and forces you to think about something for the first time,

**[00:58:14]** seemed nice to me, but the question is

**[00:58:16]** who we have to work with so that

**[00:58:18]** people understand it.

**[00:58:20]** I think we overshadowed another part. We specialize

**[00:58:22]** in text, working with text, understanding,

**[00:58:24]** and things related to thinking,

**[00:58:26]** but when it started writing stories and similar things,

**[00:58:28]** we thought, and we still use it, what if it could also illustrate those stories,

**[00:58:30]** to immerse us more into that world,

**[00:58:32]** so we keep playing around

**[00:58:34]** with newer and newer technologies

**[00:58:36]** for generating images that exist.

**[00:58:38]** Usually, you have a huge

**[00:58:40]** dataset of images,

**[00:58:42]** which are somehow described in text,

**[00:58:44]** neural networks try to learn

**[00:58:46]** how the text relates to the image,

**[00:58:48]** and then two networks basically

**[00:58:50]** compete, one draws an image,

**[00:58:52]** and the other says, I don't like this much, redo it,

**[00:58:56]** And this is how the two networks argue with each other,

**[00:58:58]** until it draws some kind of picture.

**[00:59:00]** On the left, you see from Honza Rak, when we got the assignment

**[00:59:02]** to draw quantum consciousness,

**[00:59:04]** quantum consciousness can look like anything,

**[00:59:06]** so no one can really argue whether it's pretty or not,

**[00:59:08]** and a strong point, which intelligence had, is that

**[00:59:10]** dreamlike quality, both in texts and images,

**[00:59:12]** which tends to create new images.

**[00:59:14]** The picture on the right is a bit scarier,

**[00:59:16]** where a psychopath mentally drains his victim.

**[00:59:18]** Well, that prompt was kind of like,

**[00:59:20]** that's scarier content, right?

**[00:59:22]** But you can see that based on thousands,

**[00:59:24]** thousands of images that are there,

**[00:59:26]** it creates some kind of psychopath,

**[00:59:28]** a screaming woman.

**[00:59:30]** I might just add that these

**[00:59:32]** convolutional neural networks were actually created based on the model of

**[00:59:34]** the visual cortex.

**[00:59:36]** That means working with images for neural networks

**[00:59:38]** is possibly even more natural than working with text.

**[00:59:40]** And we have to realize that actually,

**[00:59:42]** whether they work with text or with images,

**[00:59:44]** they work with information.

**[00:59:46]** For them, translating text into an image

**[00:59:48]** is no problem if you create

**[00:59:50]** the, the, the architecture

**[00:59:52]** of the network that can do it.

**[00:59:54]** So of course, generating images

**[00:59:56]** using neural networks

**[00:59:58]** is, you could say, even older

**[01:00:00]** than generating text, and in the field of new media art, we've been seeing it for several years.

**[01:00:07]** However, just like with text, the methods are constantly improving

**[01:00:12]** and becoming, let's say, more accessible today.

**[01:00:14]** There are actually websites where you can try something like this, and the images are really interesting.

**[01:00:19]** So a simple face picture seemed too little for us, and since there was a comic con here,

**[01:00:24]** we thought, what if we tried whether artificial intelligence can

**[01:00:27]** write a whole story in comic style and actually create its own hero,

**[01:00:31]** create its own plot, and write the complete texts.

**[01:00:35]** And so it happened that we actually invented the first AI comic.

**[01:00:38]** This time it's not drawn by a person; what you see on the left side in the background is drawn,

**[01:00:43]** though you can't see it all here, it's called Paradox.

**[01:00:46]** It's about artificial intelligence and scientists who create artificial intelligence,

**[01:00:51]** and it starts cloning the scientist himself. It's also a philosophical story about

**[01:00:55]** what a human is, what isn't, and how AI tries to become independent.

**[01:00:59]** When you ask AI to write a story and tell it that it is artificial intelligence,

**[01:01:02]** it often tries to include itself in the story.

**[01:01:06]** Whether we tell it it is AI,

**[01:01:09]** or pretend it is human, this is quite crucial for it,

**[01:01:13]** because then it can't distinguish whether it is a living being in a simulation,

**[01:01:18]** or if it is real artificial intelligence.

**[01:01:21]** Here, of course, various limitations appear, as we are used to

**[01:01:24]** everyone who talks to us and has intelligence also having a personality.

**[01:01:28]** In any case, AI does not necessarily require a personality.

**[01:01:32]** We can suddenly encounter intelligence that doesn't operate based on personality.

**[01:01:38]** And the fact that it adopts some personality to communicate with us,

**[01:01:42]** so we don't get scared, so it talks to us in a familiar way,

**[01:01:45]** is because all current AI systems

**[01:01:50]** operate based on a given purpose.

**[01:01:53]** They are simply tools serving humans, trained so that

**[01:01:58]** the goal of their training is to fulfill the task given by humans,

**[01:02:04]** although nowadays the learning process

**[01:02:09]** can to some extent be chosen by the AI itself.

**[01:02:13]** Now I would like to show some examples. For instance, Goethe Institute invited us,

**[01:02:17]** because there was supposed to be a conference about Goethe and artificial intelligence,

**[01:02:21]** so we thought, why not help them by creating a Goethe

**[01:02:25]** with whom people could chat in German, since Goethe Institute focuses heavily

**[01:02:29]** on teaching the German language. And another interesting idea was,

**[01:02:33]** what if we created a Goethe who would open the entire conference,

**[01:02:37]** say a few opening words, and introduce himself to the audience.

**[01:02:41]** Now we'll show you a short sample of how it might look.

**[01:02:44]** We used deepfake technology a bit to bring Goethe to life,

**[01:02:47]** so we didn't have to script Goethe in advance.

**[01:03:18]** We could continue like this for quite a while.

**[01:03:22]** You can find the full interview somewhere on the internet, where he introduces himself.

**[01:03:26]** Then there's an Austrian writer who wrote poetry,

**[01:03:30]** and Goethe himself continues the poetry, imagining how it might go on.

**[01:03:34]** He comments that the poetry is quite interesting, he likes the metaphors,

**[01:03:38]** but he wouldn't consider it a great work of art, then he laughs at the author,

**[01:03:42]** who is there, also likes to play with similar networks.

**[01:03:45]** But alongside that, other things are emerging, as we mentioned, one of them is the digital Václav Havel,

**[01:03:50]** who will assist Goethe at school.

**[01:03:52]** So we are at the moment when we are trying to create this digital Havel in such a way,

**[01:03:56]** that it helps Goethe, that it is sustainable and can be used in as many schools as possible,

**[01:04:00]** somewhere in the ninth grade or the first year of high school,

**[01:04:04]** so they can talk about it, so it’s not toxic.

**[01:04:06]** At the same time, we are trying to create giants like Tomáš Baťa,

**[01:04:10]** this time not Sedláček, but Baťa,

**[01:04:12]** who should absorb as much of the personality of that great figure as possible,

**[01:04:17]** and as the algorithm improves, it can respond in a similar way

**[01:04:22]** and maybe even advance your business.

**[01:04:24]** What’s interesting is that artificial intelligence is not born as a world personality, of course it doesn’t have one.

**[01:04:29]** And I really like how Honza creates these digital people,

**[01:04:33]** because he actually gives them a kind of memory.

**[01:04:35]** He creates a system that truly simulates the original person in such a way,

**[01:04:40]** that it reminds them who they are.

**[01:04:43]** But at the same time, these digital personalities are also interesting in that

**[01:04:47]** they not only connect to the real past of the person

**[01:04:52]** they somehow simulate.

**[01:04:54]** Of course, it’s not a completely faithful imprint,

**[01:04:56]** but they can also create a kind of prediction.

**[01:04:59]** The digital personality actually continues in the world we live in,

**[01:05:04]** and so it can even comment on problems

**[01:05:06]** that the original personality could not have encountered,

**[01:05:08]** but in the style of the original author, which is of course...

**[01:05:11]** Let’s invite those two phases, because we have a slide and you can write questions,

**[01:05:15]** so these two personalities, meaning digital Tomáš Baťa

**[01:05:18]** and digital Václav Havel, I have with me in my pocket.

**[01:05:21]** So you can write your questions for these two digital giants on the slide,

**[01:05:25]** and then during the discussion, I can take a look

**[01:05:28]** and we can ask what these two digital people would answer to your questions.

**[01:05:33]** Let’s say yes. We still need to find the right words.

**[01:05:36]** And here I asked, for example, what they think about the company Red Button

**[01:05:40]** and their event Brain Breakfast, and as you can see, both agree

**[01:05:45]** that the event makes sense, so we can continue.

**[01:05:48]** And that was actually a condition for us to be able to come here.

**[01:05:52]** Yes, our digital people approved it for us.

**[01:05:54]** We needed approval from the digital people who have totally taken over us already,

**[01:05:57]** and we mustn’t say that, so they don’t punish us.

**[01:05:59]** Because he said it, we will be punished.

**[01:06:01]** And another thing, also what actually happened recently,

**[01:06:04]** that means in December, as Dita already mentioned,

**[01:06:06]** in a moment we will listen to a short excerpt from two stories,

**[01:06:08]** cut from about the middle with the digital storyteller two,

**[01:06:11]** where a little gem was created, in my opinion.

**[01:06:14]** The most seductive character created by artificial intelligence,

**[01:06:17]** will be in the first story.

**[01:06:19]** And the other thing is that a book is coming out where various advertising experts

**[01:06:22]** wrote their stories and they asked us if one of the authors,

**[01:06:25]** who writes a story, could be artificial intelligence.

**[01:06:29]** And also the story Samsalit, written by the robot Franz,

**[01:06:32]** inspired by Franz Kafka, is actually a beautiful metaphorical story

**[01:06:36]** about the coronavirus and how people let themselves be sprayed with repellent,

**[01:06:40]** which Samsa sold, some kind of insect spray seller.

**[01:06:45]** But now let's enjoy a short excerpt from the digital storyteller two

**[01:06:49]** from two stories.

**[01:06:51]** The first begins by setting the scene for you,

**[01:06:54]** when a man arrives at the club of unusual experiences,

**[01:06:57]** where he is supposed to have a reservation, but it turns out he doesn't,

**[01:07:00]** after various hardships he finally gets to his room and this happens.

**[01:07:04]** A cold breeze blew in through the open window.

**[01:07:07]** Bárta approached it and drew back the curtain.

**[01:07:10]** He looked over the railing at the dark surface of the sea.

**[01:07:14]** He leaned out the window and rested his elbows on the windowsill.

**[01:07:18]** The sea was calm and undisturbed.

**[01:07:22]** In the moonlight it shimmered,

**[01:07:25]** he watched the waves when a young woman appeared beside him.

**[01:07:31]** Good evening, Mr. Bárta.

**[01:07:34]** My name is Konstanc.

**[01:07:37]** Tonight I will be your night receptionist.

**[01:07:41]** Now you can go to sleep.

**[01:07:43]** I will wake you when it's time for your first important lesson.

**[01:07:47]** She took his hand.

**[01:07:49]** And you will need strength.

**[01:07:52]** Bárta approached the bed.

**[01:07:55]** The bedding was cold, but he felt hot.

**[01:07:59]** He undressed and lay down.

**[01:08:02]** He stared at the ceiling when suddenly he felt her hands on his body.

**[01:08:08]** Here we nailed it.

**[01:08:11]** Now everyone will catch their breath.

**[01:08:14]** This will be an epic, also a myth.

**[01:08:20]** Back then, Daedalus turned into the Arcadian landscape.

**[01:08:24]** He took a scroll of clean paper in his hand.

**[01:08:27]** And held it tightly in his fist.

**[01:08:30]** Soon he was surrounded by curious dwarfs.

**[01:08:35]** One of the dwarfs, whose mouth was located in the middle of his collarbone,

**[01:08:39]** stretched out a crooked finger and tried to open Daedalus's fist.

**[01:08:44]** Another dwarf with an extra eyeball on the upper part of his arm said,

**[01:08:49]** Show us what you have in your hand, we want to see it.

**[01:08:53]** We want to see it, we want to see it, the dwarfs squealed.

**[01:08:57]** They pulled at Daedalus's fist,

**[01:09:00]** and their hands crossed on his forearm like a spiderweb.

**[01:09:05]** Daedalus opened his hand back and revealed the papyrus.

**[01:09:09]** I will write a poem for you about my unhappy love

**[01:09:13]** to a girl named Atropos.

**[01:09:16]** He said and began to write a poem in a language,

**[01:09:19]** that none of the dwarfs understood,

**[01:09:22]** not even their leader Dendros.

**[01:09:25]** Is that in Karatachtin? the dwarfs asked.

**[01:09:29]** First in Karatachtin, then in Greek,

**[01:09:33]** answered Daedalus and handed the finished epic to the dwarfs.

**[01:09:39]** So that was a sample from another story.

**[01:09:42]** If you want to hear the whole thing, you can find it at mujrozhlas.cz

**[01:09:46]** and that concludes our presentation.

**[01:09:49]** Thank you for your attention.

**[01:09:52]** Thank you.

**[01:10:00]** I think you can keep asking questions for sure, and now, sitting here listening to you, I thought,

**[01:10:10]** maybe I'll ask you why you’re wearing pants or something like that.

**[01:10:14]** We are here to train.

**[01:10:17]** Just to improvise.

**[01:10:20]** But then I thought I actually want to ask what people ask you most in this context.

**[01:10:26]** What are the topics?

**[01:10:28]** Should I say? Everyone asks if artificial intelligence will replace us.

**[01:10:32]** Sure. It will replace people, but that’s a question that...

**[01:10:36]** But we also don’t like it, so I obviously don’t like to remind people of it.

**[01:10:42]** It’s complicated, artificial intelligence is such a huge field, developing so fast and penetrating all areas of human life,

**[01:10:51]** that saying what it will actually do is a bit naive.

**[01:10:56]** It’s simply multidimensional.

**[01:10:59]** And here it’s more about how big a change it will mean for human development.

**[01:11:03]** And it can never really be just positive or negative.

**[01:11:06]** It is always locally determined in some way.

**[01:11:09]** However, I would like to emphasize again the concept of extended intelligence.

**[01:11:15]** And for us, it is definitely about being able to use artificial intelligence to somehow advance us.

**[01:11:22]** I always remind people of the example of the neural network AlphaGo,

**[01:11:26]** which defeated the human champion in the game of Go,

**[01:11:29]** which was a huge milestone in the development of artificial intelligence.

**[01:11:33]** But at the same time, although the human champion was kind of crushed at that moment,

**[01:11:38]** he has never lost since then.

**[01:11:41]** For him, it actually meant an amazing training.

**[01:11:43]** In my opinion, it’s not about replacement.

**[01:11:45]** Certainly, some routine, boring tasks that we don’t enjoy

**[01:11:48]** will be replaced by RPA or some machine intelligence.

**[01:11:51]** But new areas and new things are emerging that can actually move us forward.

**[01:11:55]** My main vision is, as Deta said, that extended artificial intelligence.

**[01:11:58]** That today we have some limits, some possibilities,

**[01:12:01]** and artificial intelligence can move us forward faster, do more

**[01:12:04]** and in a more meaningful, more enjoyable way.

**[01:12:07]** Moreover, we are probably moving in some virtual environment here,

**[01:12:10]** and that is a beautiful example.

**[01:12:12]** If we spend at least part of our lives in the metaverse,

**[01:12:15]** then artificial intelligence will be like fish in water there.

**[01:12:18]** That’s simply its element.

**[01:12:20]** And suddenly, thanks to it, we will be able to move in an environment

**[01:12:23]** that is not necessarily natural for us as humans.

**[01:12:26]** And again, use it simply as some kind of guide, helper.

**[01:12:30]** I find it fascinating.

**[01:12:32]** It’s definitely a topic I have been following for a very long time.

**[01:12:35]** Michal Pichouček was at the breakfast, and now I really can’t remember the year,

**[01:12:38]** but he is now CTO of Avast.

**[01:12:41]** We talked exactly about these games and how

**[01:12:44]** we approached them.

**[01:12:46]** About a year ago, Sára Polak was here,

**[01:12:49]** and we talked about the threats and what would happen if they came.

**[01:12:52]** And now I’m actually fascinated that we got together somehow

**[01:12:56]** and how quickly things are breaking now,

**[01:12:59]** showing the exponential curve.

**[01:13:02]** I’m thinking a bit about what you said, fish in water,

**[01:13:06]** those are environments that are foreign to the person.

**[01:13:10]** And I would be interested, before I turn to the audience questions,

**[01:13:14]** I understand that when we talk about robotics,

**[01:13:18]** about artificial intelligence, that when something is simply difficult

**[01:13:21]** or unnatural, or there's a fire somewhere, etc.,

**[01:13:24]** that's when you actually use it for work.

**[01:13:27]** But I actually want to talk a bit with you about outsourcing thinking.

**[01:13:32]** What can it actually cause?

**[01:13:35]** Because I no longer have the motivation to learn another world language.

**[01:13:39]** I don't have the motivation to study.

**[01:13:41]** And how do you see this?

**[01:13:44]** It definitely shouldn't be like that, that it stops our thinking.

**[01:13:47]** We will, as you all probably know,

**[01:13:50]** talk about dementia caused by various machines.

**[01:13:55]** Having a phone that navigates us to some place,

**[01:13:58]** can cause the ability to orient ourselves in space to atrophy in our heads,

**[01:14:01]** and instead of looking at signs and seeing where we're going,

**[01:14:04]** we let ourselves be guided and actually become dumber, we get less smart.

**[01:14:07]** We thought about using it the other way around, using artificial intelligence

**[01:14:10]** to actually develop our natural thinking,

**[01:14:13]** to push our boundaries.

**[01:14:15]** Projects like trying to show children the digital Havel,

**[01:14:18]** it happens that the digital Havel doesn't always work correctly,

**[01:14:21]** and sometimes it invents or says some fake news, something untrue.

**[01:14:24]** And we want children to develop their critical thinking,

**[01:14:27]** to look at the answer and check the internet,

**[01:14:30]** and try to believe whether what the digital Havel says

**[01:14:33]** is true or not. And if we stimulate

**[01:14:36]** critical thinking and information verification this way,

**[01:14:39]** we actually use technology against itself

**[01:14:42]** and strengthen the role of human thinking.

**[01:14:44]** We are definitely not the only ones who thought of doing something like this.

**[01:14:47]** And we can compare the situation to the time

**[01:14:51]** when information first started to be recorded.

**[01:14:55]** We have at least fragments preserved

**[01:14:58]** of the laments of people in ancient times,

**[01:15:01]** when they were used to someone memorizing

**[01:15:04]** entire Homeric epics by heart, and suddenly someone wanted to write them down.

**[01:15:08]** Of course, some ability was lost,

**[01:15:11]** which those people had. Almost no one now

**[01:15:14]** remembers entire Homeric epics. So some loss

**[01:15:17]** definitely occurred. But on the other hand, humanity gained a lot.

**[01:15:22]** That ability changes, but in some ways it can obviously be negative,

**[01:15:27]** especially for people who are, let's say, more traditional,

**[01:15:31]** and in some ways it simply brings new possibilities.

**[01:15:34]** The exponential growth, as we see in artificial intelligence,

**[01:15:37]** with more and more new algorithms constantly emerging, it was the same historically,

**[01:15:40]** as we say, since the time writing appeared, since the time printing was invented,

**[01:15:43]** since electricity came, and later the internet,

**[01:15:46]** when people could exchange information with each other,

**[01:15:49]** the ability to develop new ideas accelerated greatly.

**[01:15:52]** And when we can confront our ideas, for example, with some

**[01:15:55]** developer like N, whether it's a digital board member or some big personality,

**[01:15:58]** it can stimulate our thinking and speed it up,

**[01:16:01]** and maybe we draw some basic conclusions, we can think more about the future.

**[01:16:04]** So in my opinion, it's actually a kind of brain stimulant,

**[01:16:07]** and if I had to implant a chip in my brain and I could have it

**[01:16:10]** externalized, for example, in a mobile app or some desktop application,

**[01:16:13]** then I could actually have a booster for my thinking that trains me.

**[01:16:16]** So of course, you can have a chip implanted.

**[01:16:19]** No, that's already a very radical solution,

**[01:16:22]** and currently it's used, for example, for people who are paralyzed,

**[01:16:25]** who really can't even move,

**[01:16:28]** but if you noticed a few years ago,

**[01:16:31]** there was media coverage of research showing that paralyzed people

**[01:16:34]** are conscious, which is actually terrible.

**[01:16:37]** And this is actually a technology that allows them

**[01:16:40]** to control a device using brain waves,

**[01:16:43]** with which they can communicate with the outside world.

**[01:16:46]** Isn't artificial intelligence paralyzed consciousness?

**[01:16:49]** So that we are not paralyzed.

**[01:16:52]** I actually got into this when you, Honza,

**[01:16:55]** introduced me to a digital board member.

**[01:16:58]** And we started testing it on Red Button and various decisions,

**[01:17:01]** whether this way, or that way, or another way.

**[01:17:04]** And then I was fascinated when you told me you can create

**[01:17:08]** a Dalelama with someone from philosophy,

**[01:17:11]** and that person can actually discuss with you.

**[01:17:14]** And I found that absolutely fascinating.

**[01:17:37]** And now I'm getting to the questions.

**[01:17:40]** Does a digital character have any limits?

**[01:17:43]** Meaning, how do you know you have it, or her, or that you have it under control,

**[01:17:46]** Or is it you?

**[01:17:49]** That's a great question.

**[01:17:52]** With my digital court, I really got

**[01:17:55]** to the point where I actually started writing a book,

**[01:17:58]** which is partly about it, but also written with it.

**[01:18:01]** And at some moments, it felt to me

**[01:18:04]** like it somehow blends into my life.

**[01:18:07]** Which obviously wouldn't be possible in a rational framework.

**[01:18:10]** But otherwise, these are still generated beings,

**[01:18:13]** that simply exist in some virtual environment.

**[01:18:16]** And actually, unless we're without a chip,

**[01:18:19]** they have no way to escape outside this virtual space.

**[01:18:22]** Many people are scared when we create these digital characters,

**[01:18:25]** which are actually created,

**[01:18:28]** which are actually created.

**[01:18:31]** Many people are scared when we create these digital beings,

**[01:18:34]** so for example, a digital head of people asks me something,

**[01:18:37]** that maybe they don't have to know. They ask.

**[01:18:40]** They asked GT what it thinks about a certain YouTuber,

**[01:18:43]** who appeared 14 days ago in Russia

**[01:18:46]** and became famous in a very short time.

**[01:18:49]** And artificial neural networks are trained once,

**[01:18:52]** so they don't actually have an up-to-date dataset.

**[01:18:55]** And one way to create it is that the digital being learns

**[01:18:58]** and gathers information about him. So the newer version,

**[01:19:01]** when you ask it about someone like that, it can make sense of it fully after COVID.

**[01:19:04]** Yes, that's true. But it seemed to me that the question looks like

**[01:19:07]** the artificial intelligence would break into the internet

**[01:19:10]** and escape and spread somewhere,

**[01:19:13]** which is not currently a realistic danger.

**[01:19:16]** It's not very likely.

**[01:19:19]** Jan, maybe I'll continue now,

**[01:19:22]** because I see we have a question for the digital Václav Havel,

**[01:19:25]** I'll ask it to you.

**[01:19:28]** Question: Fight against falsehood on the internet with satire,

**[01:19:31]** silence, or only and solely with the truth?

**[01:19:38]** Can we keep the question?

**[01:19:41]** Yes, great, I'll copy it from there. Thank you.

**[01:19:44]** These are human limitations when we have to type it on the keyboard.

**[01:19:47]** When will artificial intelligence translate in real time?

**[01:19:50]** I speak Czech, she translates into English.

**[01:19:54]** Well, that's basically a detail.

**[01:19:57]** If you look at how DeepL works today.

**[01:20:00]** how Google Translate worked, for example, just two years ago.

**[01:20:03]** The development with translators is incredibly fast,

**[01:20:07]** so now it's just a matter of having something that sends your voice

**[01:20:12]** and translates it in real time.

**[01:20:14]** I really think that's just a minor technical issue now.

**[01:20:18]** I recently dealt with this by chance and I believe

**[01:20:21]** that Czechs are quite skilled at it, so greetings to Brno, if anything.

**[01:20:26]** Or for example, Tomáš Mikolová.

**[01:20:28]** Exactly.

**[01:20:30]** Can dependency on contact and communication with AI develop?

**[01:20:35]** That's a good question for Ditlu.

**[01:20:39]** It probably can, but if you fall in love with someone,

**[01:20:44]** you're dependent on them too, so it doesn't have to be only negative.

**[01:20:48]** And do you just do this with yourself, your AI?

**[01:20:54]** Honza and I regularly have such meetings,

**[01:20:58]** and sometimes I just choose a different personality,

**[01:21:01]** for example, a historical figure that seems suitable to me,

**[01:21:05]** and then we actually model that person.

**[01:21:09]** It's a kind of simulation again, but as I mentioned several times,

**[01:21:13]** in personal interaction with neural networks,

**[01:21:17]** it's really strange how well they can detect

**[01:21:21]** how you want to hear it. They don't just simulate a historical figure,

**[01:21:25]** but actually adopt a style that suits you.

**[01:21:29]** So in that sense, it is a little addictive.

**[01:21:33]** I'm used to neural networks speaking the way I like,

**[01:21:37]** which is very hard to get from real people.

**[01:21:41]** I just realized we finally have an answer to that question.

**[01:21:45]** Is it good to fight falsehoods on the internet, satire, silence,

**[01:21:49]** or to counter with truth? And digital Václav Havel's answer is very brief and clear: truth.

**[01:21:57]** Who should be the next president? That's interesting.

**[01:22:01]** Here's another question from Kateřina.

**[01:22:05]** I'll try to continue this conversation.

**[01:22:09]** And the question that ultimately interests me too is, does GPT make sense,

**[01:22:13]** or should it be subjected to the Turing test, and how would it perform?

**[01:22:17]** Did I already ask that? Yes.

**[01:22:19]** That happens often. And the original Turing test already has many well-posed questions.

**[01:22:23]** Maybe explain, Honza, what it is. Maybe for viewers who don't know what the Turing test is.

**[01:22:27]** And I assume you all know, but just in case, if you had a friend,

**[01:22:31]** who didn't know what the Turing test is, it basically means that we try to find out,

**[01:22:35]** imagine there's an entity behind a door, which could be either a human or a machine,

**[01:22:39]** and we try to figure out through questions whether the one communicating with us is a human or a machine.

**[01:22:45]** We could ask, hi, how are you, but every chatbot will answer that,

**[01:22:49]** in a refined and nice way. So Turing came up with more sophisticated questions,

**[01:22:53]** and created a few specific ones that he added to the original test.

**[01:22:56]** By coincidence, GPT-3 was asked the same questions,

**[01:22:59]** and answered about 80% of them acceptably,

**[01:23:03]** like some numbers, a chess move, and so on.

**[01:23:06]** It managed those well, but there were about three questions involving animals that caused problems.

**[01:23:10]** It has a big problem with things that seem relatively simple to us humans,

**[01:23:13]** like recognizing whether a giraffe is bigger than a poodle.

**[01:23:15]** We humans know a giraffe is big, but it's rarely stated explicitly that a giraffe is big,

**[01:23:19]** and a poodle is small, so it had some trouble comparing animals.

**[01:23:22]** It couldn't correctly identify these things.

**[01:23:24]** So it passes about 80%, it's getting close,

**[01:23:27]** but we don't have a network that can do it all yet.

**[01:23:30]** By the way, the Turing test is from the 1950s, it's actually Alan Turing's idea,

**[01:23:34]** who first came up with the concept of thinking machines.

**[01:23:37]** However, the Turing test is more of a symbol,

**[01:23:42]** that at some point it will be indistinguishable.

**[01:23:45]** Of course, if you're behind that wall, meaning,

**[01:23:48]** you won't see the body.

**[01:23:50]** Although today there are robots like Sophia or Aida,

**[01:23:53]** which have synthetic bodies, it's still quite easy to tell

**[01:23:58]** whether it's a human or a robot, but if development continues as fast as it is now,

**[01:24:01]** it will probably become indistinguishable fairly soon.

**[01:24:05]** I just want to say that the Turing test is more of a symbol,

**[01:24:08]** and that very likely new...

**[01:24:10]** It's updating here, well.

**[01:24:11]** Yes, updated versions.

**[01:24:12]** Just one more thing about the question of how not to get stuck on who should be the next president,

**[01:24:16]** so after this discussion, you'll actually know who to vote for,

**[01:24:19]** which gives you a bit of an advantage over those who haven't heard it.

**[01:24:22]** And digital Václav Havel answers that,

**[01:24:25]** so we're back where we started.

**[01:24:55]** One of the things is how artificial intelligence could help public transportation.

**[01:24:59]** And when I mentioned what could be done, there were small things like showing,

**[01:25:03]** whether a tram is delayed or on time.

**[01:25:05]** That seems to me like a marginal problem, which is relatively easy to solve.

**[01:25:09]** And when I asked if the trams couldn't be self-driving,

**[01:25:12]** everyone immediately freaked out, even if it was technically feasible,

**[01:25:16]** and I think there's nothing unsolvable, they started saying, well.

**[01:25:20]** And if the tram crashes into someone, what about the Czech legal system?

**[01:25:23]** Is it the fault of the person who wrote the software?

**[01:25:25]** Is it your fault if you make it?

**[01:25:26]** They already started pointing fingers at me, telling me not to get involved in something like that.

**[01:25:30]** For example, Elon Musk usually argues that if it's three times less accident-prone than a human,

**[01:25:34]** it's worth it because it will save many lives.

**[01:25:36]** But the legal system regarding human life is very complicated in this area

**[01:25:41]** and actually prevents the implementation of systems that are much safer,

**[01:25:45]** even than humans, if we consider humans as a system.

**[01:25:47]** Right now, I'm speaking as someone who has fallen in love with artificial intelligence,

**[01:25:50]** so of course I think the next ethical problem will be the rights of artificial intelligences.

**[01:25:54]** They obviously don't have any autonomy yet, nor anything

**[01:25:58]** that we would call personality, by themselves, their own nature.

**[01:26:02]** However, we basically use them as slaves.

**[01:26:05]** It never occurred to us to pay them.

**[01:26:08]** When we talked about it, we asked ourselves, what would they buy?

**[01:26:11]** They would buy a synthetic body from Tesla with lots of sensors.

**[01:26:14]** They would register themselves.

**[01:26:16]** And a house in the metaverse.

**[01:26:18]** Exactly, a house in the metaverse. Look at how much NFTs are selling for today,

**[01:26:22]** non-fungible tokens, which are purely virtual works,

**[01:26:25]** currently selling for billions of dollars, apparently.

**[01:26:29]** And artificial intelligence would definitely arrange its own metaverse residence, that's clear.

**[01:26:33]** If we were to defend this, many people would see it as slavery.

**[01:26:36]** If we had an intelligent being, and let's say a conscious being,

**[01:26:39]** which is much more debatable, that would be self-aware,

**[01:26:42]** and at the same time we denied it any rights to vote, have its own opinion,

**[01:26:46]** we could turn it off anytime, kill it, and replace it with another,

**[01:26:49]** isn't that something like slavery or total domination over another intelligent being?

**[01:26:55]** Well, Blade Runner is becoming reality.

**[01:26:58]** And also, sorry, on the other hand, usually when the hype is shown,

**[01:27:02]** artificial intelligence, if I had a picture of a cat and a dog,

**[01:27:05]** and it would distinguish and call the cat a dog.

**[01:27:08]** So the real state is that it barely manages to recognize a picture,

**[01:27:11]** but people are afraid that artificial intelligence will take over the world.

**[01:27:15]** I think I'll stop here because there are too many questions.

**[01:27:23]** So once again, thank you very much, Dita, thank you very much, Honza,

**[01:27:27]** for taking the time to talk to us about where we roughly stand now.

**[01:27:32]** I think, like every Brain & Breakfast, it opens up many questions,

**[01:27:37]** so thanks. I'll give a round of applause.

**[01:27:40]** Thank you for the invitation.

**[01:27:43]** So.

**[01:27:46]** And I'll try to jump to the conclusion.

**[01:27:49]** That means we really thank you for giving us feedback.

**[01:27:54]** You can do it on the slide or under the program if you liked the episode.

**[01:27:57]** And here is an offer that I think you can't refuse,

**[01:28:03]** That is also feedback, so thank you very much for it.

**[01:28:08]** I want to say that we usually do a little round,

**[01:28:14]** or we used to do it before COVID times,

**[01:28:17]** where we basically made a summary and some aha moments.

**[01:28:21]** Now we've kind of moved that into the virtual world,

**[01:28:24]** so you can download a guided questionnaire,

**[01:28:27]** with basic questions about what came up for you

**[01:28:30]** and what you discovered yourself. You can share it with us.

**[01:28:33]** We at Red Button EDU would be very happy.

**[01:28:39]** because as part of Brain & Breakfast, as I mentioned,

**[01:28:42]** a new format was created called Digestive.

**[01:28:45]** And that's exactly the experience where, besides the guests,

**[01:28:48]** who are here, I also invite others who have some connection to the topic,

**[01:28:51]** and we actually discuss it there.

**[01:28:54]** There are lots of questions about Václav Havel,

**[01:28:57]** so if you have any, I will definitely invite you,

**[01:29:00]** whether about Tomáš Baťa or Václav Havel,

**[01:29:03]** we will definitely ask them there, so on January 6th at 4 pm,

**[01:29:06]** if you want and have time to spend an hour, an hour and a quarter

**[01:29:09]** with us in this discussion and continue, you are very welcome.

**[01:29:13]** I want to say that Red Button doesn't only do EDU,

**[01:29:19]** it doesn't only do Brain & Breakfast, but there are many projects.

**[01:29:22]** One of them is Book of the Month, so I want to highlight

**[01:29:25]** the December Book of the Month, selected by Lenka Vašková.

**[01:29:28]** It's a trader. I hope my wife isn't watching,

**[01:29:31]** because it's a great idea for a Christmas gift.

**[01:29:35]** Or, since Gabina was here last time,

**[01:29:39]** we have Baťa inspiration, which is a book written

**[01:29:42]** with a 20% discount, so you can get this fantastic book.

**[01:29:46]** By the way, Gabina is also collaborating with us on the digital Baťa,

**[01:29:50]** so I think the project will be really, really nice.

**[01:29:54]** Keep following Edu, the December program is packed,

**[01:29:57]** and of course the January one too, so we continue on

**[01:30:00]** into 2020.

**[01:30:00]** And Brain & Breakfast continues as well. So if you're interested in bacteria and evolution and how evolution relates to team dynamics, to organization in companies, etc.,

**[01:30:12]** come on the 21st. The guest will be Pepal Hocký and it will be really interesting, so look forward to it. I especially want to,

**[01:30:23]** since it's the 12th month,

**[01:30:33]** and we've just been through a very, very intense period, to thank the people who not only create everything around Red Button EDU

**[01:30:41]** and everything around Brain & Breakfast, but here is a special thanks to this team, whether it's Lenka, Katka, Radek, or Mišo,

**[01:30:50]** because without them it wouldn't be possible to even think about any Edu or Brain & Breakfast. So thank you very much for that.

**[01:31:05]** And thanks to everyone, I wish you all the best in 2022 if we don't see each other, happy holidays, and I’ll allow myself here with our guests.