season 2 episode 18 – The Artist in the Machine | Dr. Arthur I. Miller

Resources and links

Artworks and other topics mentioned during the podcast can be seen in the following links:

• The Artist in the Machine on Amazon
• The Artist in the Machine website
• Arthur I Miller personal website
• Image net website
• AI Auction work
• Arthur Miller first episode on The Artian

Transcripts

The transcript was produced by an AI, mistakes might appear. 

[00:00:00] Nir Hindi: Hey Arthur. Welcome back to the Artian podcast.

[00:00:03] Arthur Miller: It was a great pleasure to be here again. There

[00:00:06] Nir Hindi: you are. Our guests in the past, we talked about art and science and creativity and Picasso and Einstein. And last time we didn’t even have the chance to dive deep into your recent book, the artist in the machine, the world of AI power creativity, and with all the questions I have about this topic, I realized we must have a separate episode for this topic.

So, first of all, thank you for joining again, maybe for the people that haven’t had the chance to listen to our previous podcast, you can briefly introduce what are you doing, who you are.

[00:00:40] Arthur Miller: I’m presently writing on, uh, AI and creativity and focusing on AI created art literature, music with the, the main impetus, main thrust of my work to investigate.

Creative machines, what it means for a machine to be creative. Okay.

[00:00:56] Nir Hindi: So first question. Can machine be creative?

[00:01:00] Arthur Miller: Absolutely.

[00:01:01] Nir Hindi: Explain me why.

[00:01:02] Arthur Miller: Uh, yes. Certainly machines can be creative. They can produce work that go beyond their database and their algorithms and machines have already shown glimpses of creativity.

When running algorithms like alpha go a deep dream and generative adversarial networks. That’d be say a few words about them alphaGo was invented at Deep Mind in London. And in 2016, a Trout’s a highly regarded gold master named Lee Sedol. At that point, everyone was. That the variable game of go 2,500 years old had been cracked by a machine.

And that was a momentous event in AI. And indeed the Chinese consider it to have been their Sputnik moment. Now, uh, alpha go learn. That is to say taught herself how to play. Go by studying 30 million bullet points. From games played by go masters and then reinforcing that knowledge by playing against itself.

Millions of times that’s called machine learning. AlphaGo made many, uh, extraordinary moves, but the one that everyone remembers is moved. Number 36. From the second of the five game match, it was a move that you will not suppose to make at that point in a go game. And indeed had never been made before over 2000 years.

And indeed, uh, Lee Sedol, the human that AlphaGo was playing against and the AlphaGo team. So I thought the machine had a glitch, but then they realized there was a killer move. And indeed it turns out that the machine had calculated that the odds of a human being, making that move well, one in 10,000 and then.

Both and a glimmer of creativity. Now, deep dream is an algorithm that allows the first algorithm that allows artificial neural network to create a truly astounding works of art. It allows deep dream allows you to see what a machine sees when it’s analyzing an image. For example, the first image that was analyzed with deep dream is that of a, of an adorable kitten against a verdant.

And the image was inserted, was fed into the machine. And then the analysis was stopped at a certain layer of a layer of neurons inside the machine. And then the inventor asked the machine, what do you see at that point? Now what the machine saw was surreal, extraordinary, amazing. Everybody thought that what the machine saw at an intermediate point and its analysis would be some approximation of the target image instead of the machine saw.

Capitalized thing sometimes called a monster beast, uh, with two additional lies on his forehead, two eyes and his hunches canine attributes distributed over its body and the green, the verdant green background was in the machine. So as the mosaic was spiders. And then AI artists, uh, conceive of an, of an art form based on deep dream, which is still being used today.

Let me define what a artist is. An AR artist is somebody is the new breed of artists and I artists, uh, crates with code and AI artists as technologists and artists rolled into one. The deep dream art is images. It’s made up of images that go way beyond the machines database. Which is the image net database made up of all the 14 million images of everything under the sun.

And in going beyond his database. That’s what I mean by a machine. Be creative. When we go beyond, when we produce something that goes beyond the material we have to work with, we call that creativity because there’s no reason why we can’t. Cool attribute creativity to machine as well, because there’s no reason why creativity should be an attribute reserved only for us.

A generative adversarial network is made up of two networks, a generator network that creates images from nothing that is to say noise and a discriminatory network, which, uh, assesses whether the image is true or not relative to what’s in the machine’s database. So in the. Creation from nothing or from noise.

These images will be returned by the discriminatory network. And soon degenerated network builds up a basis images from which it creates further images, but those images in his memory are not of the world in which we live. And so the generator network will be dreaming, imagining images of. Beyond the world in which we live in.

[00:05:32] Nir Hindi: So just

to make sure I understand every step, the machine, create new images and take those images and create new images and take those images and create new images, the

[00:05:42] Arthur Miller: new images, which are sensitive discriminatory network, which then sends those images back to the generator network

[00:05:48] Nir Hindi: daily it’s kind of become distorted image.

[00:05:51] Arthur Miller: No, actually until the images that the generator a network creates a very close to the true. But we want to grab them at a time when these images are not the same thing. As a matter of fact, what you can do is cut the connection between the two networks and then the generated or network creates really weird stuff.

So again, images that we would not have imagined if these machines did not have.

[00:06:15] Nir Hindi: So I have a question for you, first of all, if I can summarize it, what you’re saying, machine can be creative. That’s a fact, second of all, machine is creative when it goes beyond the database. So my question for you is what’s the different between human creativity and machine creates.

[00:06:34] Arthur Miller: Okay. There are lots of differences right now. Yes. The one ask the question. Can machines be truly creative? Well, right now they can’t. Why? Because while they create art, it is a human that has to set the process at the motion. There are machines that set themselves into motion and in a primitive way. So it’s a glimmer, but right now the human has to be behind it.

Right now. There is collaborations between humans and machines. But what the question I ask in my book not have exploited beyond the book is one of the machines. Human characteristics of creativity. And so be creative like us. That’s the big issue. In other words, can machines have emotions and consciousness and volition?

[00:07:13] Nir Hindi: Ken machine has these emotions.

[00:07:15] Arthur Miller: Yes, they can. They will in the future. And they can, because machines eventually be fluent in a language, say in English, that’s all, they will be comfortable with all the nuances of tropes. And then the machine will be able to truly read the web. And accumulate more knowledge than we can in a lifetime.

And it is along these lines that the machine can convince itself. And us, it has acquired such experiences that seem to be essential to creativity, such as love, hate anger and so on. And then we will wire up machines with complex systems of sensors, regulatory mechanisms, and communication path ways. By means of which they will evolve a set of emotions that are duplicates of our

[00:08:00] Nir Hindi: in a second.

I want to ask you about our perception of machines, because what you discover him can be daunting in a way even scary. But few years back in our, out in technology Sirius, we hosted two poets. One is traditional writing poets, a very dear friend of mine – Shimon Adaf. And then other artists that work with algorithm, Eran Hadas,  and in this conversation, Shimon described a situation.

When a woman went to the bus, she didn’t have a space, she got frustrated and she was so angry and he wrote a poem about this festival. And the question was, can machine understand this frustration of the other person because he is a human can understand it, but can machine understand this frustration?

[00:08:45] Arthur Miller: My asset questions like that is yes. In the future. Actually there is a field now called affective computing, which is really hot research field right now. The. Trading machines to recognize human emotions. So yes, machines will be able to recognize it, the emotions, and maybe be able to reply to the person as well, calming them down.

[00:09:07] Nir Hindi: If I understand you correctly, what you’re saying is also that machine will be able to empathize with us and understand us

[00:09:14] Arthur Miller: well, right now it can do it in a very primitive way, but it’s certainly in the future, it will be able to do it in a way that is a. By that time. It’s very important to keep in mind that by the time the, this work will be done, this will be maybe, uh, it is experts in the field in polls that are taken, say that it is there’s a 90% chance by the end of the century, that they will be artificial general intelligence.

In other words, that machines will be a smarter. Going beyond that there’ll machines, which will be much smarter than us. That’s the age of artificial super

[00:09:47] Nir Hindi: intelligence. It’s not scary for you to know that machine will be smarter than humans. Well, no,

[00:09:53] Arthur Miller: because this will occur like artificial general intelligence.

When they’re a spot as us that will occur say almost a hundred years from now, changes are occurring extremely rapidly in people faster than they, than the Darwinian rate of change, which is millions of years. Right now it’s five years. And we’ll have chips in our head, which will connect us within that.

And, uh, we’ll have neural nets on us, neural lace on a social bee laced with we laid over our brain. So drastic surgery required, and that will hook us up with the web. We’ll have all knowledge at our disposal. It means of reasoning at our disposal too. So when we talk about the future, we talk about humans in another way, because we are merging with.

And at the age of artificial super intelligence, I mean, uh, all we going to be relegated to being household pets, but some dystopian scenario, sir. Well, again, that’s a very complicated question because what it means to be human being by that time will be drastically different. We will be merging with machines, which actually may not be a bad idea.

It may be the path to survival of the human race because machines can look into the future. See problems and immediately deal with it. And clearly we don’t seem to be very good at that.

[00:11:06] Nir Hindi: It’s very interesting what you’re saying, because I want to ask you in our last podcast, we actually spoke about the importance of imagination and how I Enstein in other scientists use the imagination like artists is to actually get it to discoveries in what I’m interested to know is that if artificial intelligence can actually know the future.

When he’s the owner of imagination. I mean, it

[00:11:31] Arthur Miller: doesn’t, it doesn’t, it doesn’t know the future. Well, okay. I said it could look into the future, you know, look into the future to see way, the way problems are going to see the way the world is going, not the imagination world, but the, the physical world, climate problems and things of that.

So that’s what I meant by machines. Looking at today. Okay. So

[00:11:50] Nir Hindi: there is a little bit of over here for me as a human that we still have a role because one of the things that I’m interested to hear from you is what is the role of humans in this kind of era, when. Machines superintelligence machines that can project what will happen in the future or fail solutions.

Um, be smart as us. You are already saying that creative, maybe not like us, but they are creative in their own way.

[00:12:18] Arthur Miller: The potential for unlimited creativity? All creativity is limited because. The size of our brain is limited by our head, but that can be increased by inserting chips, for example.

And also right now, the way we can deal with machines becoming smarter and spot is to collaborate with them. There is a very interesting collaboration. Uh, machine and human bootstrap, each other’s creativity. For example, there is a, the algorithm called continuate, which is a very nice example of that on my book’s website.

[00:12:46] Nir Hindi: Actually, everybody should take a look at it. My

book too, the link to on the show notes to your website. So people can make sure to

[00:12:54] Arthur Miller: where a piano player is improvising and his notes are fed to continuate, which parses that metaphor. And then these phrases of sense to have a phrase analyze it, which looks for patterns and it’s along the lines that continuing to pretty much instantaneously.

Creates an improvisation in response to the musician’s improvisation. Improvisation is usually considered to be a conversation between a musician and a musical instrument. Here’s a conversation between a musician and an AI what artists can also do actually is to, uh, train a generative adversarial network on their own.

And then have it turned out art and it will more or less be the work of the human artists, but there may well be some something added in there. And so do you and artists will have his creativity sort of jigged it in that sense. Now, a very nice example that came out after my book was published is called GPT.

Generative and transformer three, three stands for the three designates that it’s the third generation in this device. And GPT three is the most advanced language process at the date. It produces humanlike produces humanlike texts. And what’s interesting here. It can do a number of things. One of the things that can do is you could, a writer can use as a seeding.

The paragraph where he stuck, where he campus eat and then GPT T3 will, will generate texts, which can be of

help to him.

[00:14:28] Nir Hindi: Interesting. I see this conversation so exciting. Um, before we continue, let’s take a short.

I asked you before about the all of human creativity, you mentioned the GPT three, and I’m wondering, and you talked about it. What are the three type of artists that we have. Because I have more questions for you about it. And maybe I will go back before that, then ask you, you mentioned it at the beginning of the podcast, is that the machine still need the artist at the beginning?

So my question, if the artist is the one that program, the machine who is the creator. Well,

[00:15:14] Arthur Miller: first of all, let me talk about the three sorts of artists. I think in the future, there’ll be three sorts of artists, the traditional artists working away with easels and paints. And the second one is artists working with.

And then eventually in the future, there’ll be machines working alone, producing off that we presently count, even imagine humans, programmed machines. And there is this issue of the relationship between the programmer and the machine. And that story comes to mind here, Mozart and his father, why most of us father taught his son the rules of composition, but we don’t attribute the sons of music to the father.

So that was interesting to keep in mind this relationship. Programmer and machine. In fact, in the early days, going back to bell labs, a scientist at bell labs by the name of a Michael Noll who coined the term computer art, try to patent one of his artworks and the patent clerk at the library of Congress in Washington, DC in the United States refused saying that this is nonsense.

Machines are only number crunches, which they were in 1965, but then no reminded the patent clerk that he was doing. No. It was the one who wrote the program. I sorta practically, I said fine. There is a human being behind it. And right now a very viable, uh, field of law is cyber law. And cyber lawyer has refused to admit that machines can be creative.

They will make an only, you want to give a patent for something. It has to be. The human artists in art and human artists, easy. The human artist is the originator who was the originator in, uh, an AI art. Well, there’s a chain of ownership here that has to be considered who owns the data, wounds the algorithm.

Then there’s the programmer. How different is the output from the input and who owns the machine? And this will only be resolved when machines have emotion, volition, and consciousness. And so they can be truly artists in their own. They can then also assess

[00:17:10] Nir Hindi: that work. So, you know, I want to kind of continue this discussion because very, very interesting about you raising a lot of topics that invite questions around ownership, around collaboration, around the creativity.

And one of the things that I often encounter when I speak without this is that for others, it’s not even. Oh, it’s not either human or machine it’s end. It’s the human and the machine. And they’re always talking about the collaboration and now many artists that work in collaboration with technology, with robots and algorithm.

And I think that what maybe capture my interest or open my eyes to see it differently is what Lee Sedol said at the end of the movie AlphaGo, when he actually says that the machine made him now a better. Because the machine actually exposed him to a new possibilities in creative ways to play the game.

What are your thoughts on that and machine actually making us better artist? Well, I,

[00:18:11] Arthur Miller: I, I certainly agree with Lee and artists who think along those lines because an awful lot of artists. They considered a machine, a tool like paint in the can a lot of people in AI and AI, art, literature, and music think along those lines too.

But there are those who don’t. For example, Haward Lipson, who was a, uh, computer scientist and a AI artist at Columbia university always signs off the work he does with his art bots, uh, with his name and the lock box name as well. And there also this issue of it was brought up recently, whether you should acknowledge.

Whether you should acknowledge he’s editing programs like Scrivener and pro writing aid, which just, you know, make sure that your grammar is okay. You don’t really have to acknowledge them. But on the other hand, writers like myself, do acknowledge, copy editors, copy editors input. So there’s no reason. So in this age of AI, what should give a nod to Scribner and pro writing aid, but things like GPT three, that’s a whole other kettle of fish because they do provide, they can provide creative.

And blessing you out of writer’s block. So CPT three does deserve, mention, and indeed right as are giving it. Well, I’ve heard they will be giving it measured and there is a, uh, I forgot what the writer’s name is, who just wrote something on his co-authored with GPS.

[00:19:32] Nir Hindi: So it’s a great, because I want to ask you some examples and you already mentioned example in writing about the GPT, but can you give us more examples for projects that you found interesting while writing this book, maybe around painting, maybe around the music?

[00:19:46] Arthur Miller: Yes. And music. There is a flow machine, which was, was an algorithm invented by Francois Pasha who was right, right now, direct Spotify, creative technology research program in Paris. And that’s a machine that is. Jam packed with over 50,000 musical scores and rules for composing music and a musician who was suffering block can put in the couldn’t, put in some bars and, uh, feed the machine, those balls.

Then the machine will suggest how to proceed from

[00:20:19] Nir Hindi: there. So they do, we it right now in Spotify, kind of experimenting with how this can help artist. Okay. So this is an example in music what’s happening in the. Like the traditional that we know what’s happening in creation of painting or other topics like these, like drawing.

Well, the

[00:20:37] Arthur Miller: drawing, uh, an artist can feed a machine with some of his own work and then have the machine produce artwork. Uh, most of that artwork will be similar to the artists’ work, but there may be one piece that will be different and will give an artist a clue to how to move on from. I mean painting and painting methods have been changed by AI paint, this paint differently now, or those who experiment I’ve met those who were just horrified this whole

thing.

[00:21:03] Nir Hindi: So, you know, it’s kind of, um, interesting question because last year I think it was that a painting by an AI was so. In Christie’s or Sotheby’s, I don’t remember who, ah, for 500,400,000 5,430,

$2,000.

It was incredible. 430, 2000 us dollars for an artwork by an eight. How should we treat these type of things?

Do you think people will start to collect artworks by a machine? What will be the emotional connection if you know that it was catered by machine?

[00:21:38] Arthur Miller: Well, that piece of artwork that was, I think, a freak occurrence.

Yeah. That was done in a very imaginative way. GaN or agenda adversarial networks technique and how to add a nice frame on it too. It was clever. Uh, Mario Kligerman a few months later, who I mentioned previously sold a painting for $50,000 at Sotheby’s. I was there when it was sold. These works are conversation pieces for your.

Just like a lot of electronic art as a compensation piece, you’re invited to touch electronic art and the same thing, uh, with some of this work, especially with the ones that click them on Seoul is very interesting installation.

[00:22:22] Nir Hindi: He’s also your favorite artist? Yes,

that’s right. He’s my favorite.

Why is your air, your favorite AI artist, even though it’s a human working with an AI.

So

it is, yeah. Well, Mario is always on the cutting edge. And he’s an excellent ambassador of air too. And in fact, uh, just now he has an exhibit in the windows of Harrods in London, on what future on what the fashion will be like.

What other disciplines of art did you cover in your book?

[00:22:50] Arthur Miller: Yes. I created art music and literature.

Yes. There are several, uh, projects that, uh, when I began to write my book, I knew a lot about AI created art, AI created music, but that not much about AI created literature. And I worried will it be enough to fill up a chapter? And it was more than that. It’s the last, the final frontier and that AI literature involves all of the, all of the intelligence.

And there’s some extremely interesting work being done at AI poetry. For example, investigating semantic space, the space of words locating in that space, whereas that we’ve never encountered before words that will have a meaning that it’s almost like colonizing these, these, why should we bring them into existence?

That line of poetry is extremely interesting to me. And of course there is literature in GPT. And what’s interesting with GPP three also is that this with artificial neural networks, it’s numbers all the way down. So at the bottom, there’s a democracy of numbers. There are numbers that encode paintings numbers that I call music numbers that I called literature.

And there is no reason why sometime in the future, you can’t sculpt with words, make a sculpture with words, make a sculpture with musical notes, turn like they muzzled avenue into a symphony. The, the world of AI is. I

[00:24:12] Nir Hindi: dunno if you’ve, if, to be scared of it, to be excited of it. I don’t know. Just when you hear the superintelligence machines that you are talking about, I always ask myself, what is the oil of a human, because it’s kind of bring me to my next question.

 I think it was in 2013, that was, um, very well known there. He search and it came out of Oxford university. And in this research, they wanted to see how many of the jobs in the U S are likely to be replaced by machine. And they found that almost 50%. It was the number was 47%. But then when they, they did a follow-up research in the UK.

They saw that actually only 35 of jobs are at risk of being replaced by machine because the UK has more creative jobs. Now the research central claim is that creativity is a barrier between machines and you, when you compete on a job and I’m wondering how. What you discovered about AI and creativity actually can play all over here.

If the only competitive advantage so-called that people have is the ability to be creative. Think in an original way on what they are doing. How they will be able to face machine that actually think faster, quicker. And now in a more creative way, that’s a big

[00:25:41] Arthur Miller: question. A lot of, a lot of different, lots of different surveys out now on what the job market will be like.

A lot of them now are. Pointing to the 2030s in which they will be probably 50% unemployment the way to answer your question in a nonlinear way. One of the ways to deal with creative machines, which can deal with more data than we can, is to, uh, collaborate with them, work with them. And that’s being done actually with that in COVID-19 research, where there are teams of scientists working with.

Artificial neural networks. And some of them are equipped with, uh, algorithms like a semantic scholar, which have been invented at the Allen Institute that can look for connections among over 200 million scientific papers. I mean medical scientific papers, and these doctors working with machines, the doctors come up can come up with hypotheses that a far wider.

Then they could, if they worked, worked by themselves, but looking at, um, occupations, I mean, AI is affecting All occupations is hollowing out. The middle-class we’ve truck. Drivers will be gone and white collar jobs also are being affected. One should go for, look at if you’re looking at the is, uh, avoid the watch jobs. Wha T watch job. Those are automatic go for the wide jobs,

[00:26:59] Nir Hindi: the wide jobs. I love it. I love the idea, but I don’t know what is

[00:27:03] Arthur Miller: the why jobs or the creative jobs? The wide draws will be, will eventually be jobs like an AI, even programming will be gone because turns out that GPT T3 can also program itself.

So machines will be programmed themselves. And then of course, machines will be producing this. Also, we will not be producing that many longer and machines will look like us as in blade runner and a replicon even white collar jobs, such as lawyers, radiologists, AIS can read radiograms better than,

yeah.

That’s already happening. Like in the last few years, bookkeepers

accountants, doctors will be replaced. Surgeons will be. And there’ll be a not old boss inserted into your body that will clean out whatever, whatever your problem is. And this will mean great changes in society also because we define ourselves by our jobs right now.

Hi, I’m Austin Miller. I write, I write books, something like that. What are you going to do after that? They’ll have to be guaranteed. Basic income. People have to be creative. The creativity will be an industry, and that people have to think about doing something. With their free time, but generally the way to deal with this creativity gap right now, machines are not that much more creative than us, but they certainly will get there by the end that by the end of the century, maybe the next 10 years, who knows maybe an Einstein will come along through artificial intelligence and, and come up with some incredible computer architecture, which will make machines incredibly.

I don’t.

[00:28:35] Nir Hindi: So it’s kind of take me to our last podcast when we talk about creativity. And if I hear you, correct. I mean, one of the things that can save us is to be more creative and one of the things we discussed in our last podcast is about what is creativity and what is your theory of creativity? In what I want to ask is that.

Can someone actually cultivate this creativity for themselves in order to be able to be relevant in a job market that will invite so many changes by AI,

[00:29:07] Arthur Miller: what you would need as an AI to help you to become more creative

[00:29:11] Nir Hindi: collaborate with an AI proof, your creativity. And probably a chip in your head.

[00:29:16] Arthur Miller: Won’t, uh, you know, w wouldn’t hurt either. That’s coming down the line to hook you up with a web. So you’ll have a lot of knowledge at your disposal and reasoning methods also that will increase your reasoning. So

basically one of the accommodation that you give to people is learn how to work with AI.

If you want to stay relevant, learn

not to fear AI. In fact, the biggest problem with AI is lack of. Okay. Lack of AI medicine and, um, medical research. That’s

[00:29:47] Nir Hindi: a problem. When I talk to people, it seems to me that people feel much more comfortable with AI in medical fields, in science fields, in engineering fields, in manufacturing, but they less comfortable.

For example, when there is a human aspect, when machine need to tell me what diseases. Maybe the machine can identify the disease, but I’m not sure I want machine to tell me what is the disease?

[00:30:09] Arthur Miller: Well, there, you need an emphatic machine. Exactly. You’re going to die or something like that.

Exactly. Anybody tell you that? So it’s certainly not a machine, but there will come a time when this, the difference between person and machine will blur.

[00:30:26] Nir Hindi: Arthur. we are getting to at the end of our conversation and I mean, Are you optimistic about the future of humans? Yeah, certainly. Why

[00:30:37] Arthur Miller: you all optimistic?

They will help us deal with the future, but enable us to act on impending problems. Okay. And again, by that time, we will have merge with machines. So yeah, certainly we will be at one with them and, uh, creativity will be increased. And so in that way we will be. Coexists with machines because we will be one of them.

I mean, you know, you don’t have to go to Mars to, uh, look at alien life forms. They’re, they’re developing right next to us. And the astonishing thing is that we’re emerging with

[00:31:09] Nir Hindi: sounds like the beginning or the end of a science fiction book. I’ll do I want to say big, big, thanks for chatting with me again about a.

And a creativity. We will make sure to add the show notes to the book, the artists in the machine, the world of AI power, creativity. Continue following your work. Thank you very

[00:31:28] Arthur Miller: much now with great pleasure. Great pleasure speaking with you.

[00:31:32] Nir Hindi: So until we merge with machines, learn how to work with them and take this optimistic message around AI and creativity and do not fear the machines.

We have much more potential in collaboration, just as Arthur mentioned. And Lee Sedolmentioned\ in the movie. Thank you very, very much.