Luke Robert Mason: You’re lis­ten­ing to the Futures Podcast with me, Luke Robert Mason.

On this episode, I speak to Emeritus pro­fes­sor of his­to­ry and phi­los­o­phy of sci­ence at University College London, Arthur I. Miller.

We should not judge the work of AI on the basis of whether it can be dis­tin­guished from work done by us, because what’s the point. We want AI to pro­duce work that we present­ly can­not even imag­ine. That may seem to us meaningless—and even nonsensical—but it may be bet­ter than what we can produce.
Arthur I. Miller, excerpt from interview.

Arthur shared his insights on machine gen­er­at­ed work, the impact of arti­fi­cial intel­li­gence on the cul­tur­al land­scape, and how com­put­ers are chal­leng­ing our under­stand­ing of what it means to be creative.

Luke Robert Mason: So, Arthur Miller. Your new book, The Artist and the Machine is a philo­soph­i­cal inter­ro­ga­tion of this thing called cre­ativ­i­ty. Specifically, it asks whether non-human enti­ties such as machines and AI are capa­ble of express­ing this seem­ing­ly only human qual­i­ty of cre­ativ­i­ty. I feel like the first ques­tion I should ask you is, where did your inter­est in the notion of cre­ativ­i­ty first orig­i­nate? Is it fair to say that it began back in the Bronx?

Arthur I. Miller: Yes, by all means it did. It began in a pub­lic library in the Bronx, where I dis­cov­ered clas­si­cal music. I then grad­u­al­ly worked my way back from Tchaikovsky, first, and then back into Bach. What always stuck in my mind is: How did these peo­ple do it? How did they think of those melodies? What is the nature of cre­ativ­i­ty? But in the Bronx, at that time I was grow­ing up—if you were smart, you went into physics, which I did. I became a the­o­ret­i­cal physi­cist doing research in ele­men­tary par­ti­cle physics, but those What is the nature of…? ques­tions stayed in my mind, and so I decid­ed to switch it to his­to­ry and phi­los­o­phy of physics. I read the orig­i­nal German lan­guage papers in rel­a­tiv­i­ty and quan­tum the­o­ry, and what jumped out at me was the role of visu­al imagery in the great works of Einstein and Niels Bohr, and Verner Heisenberg. I want­ed to know: How did those images orig­i­nate in the mind? Were they stored there for fur­ther use, or what? At that time, there was a so-called image con­tro­ver­sy in cog­ni­tive sci­ence con­cern­ing whether images have a causal effect on think­ing. To me, they did. I took part in that con­tro­ver­sy, and it turned out that a tool that was very use­ful was to con­sid­er the brain as an infor­ma­tion pro­cess­ing sys­tem. In oth­er words, as an addi­tion­al com­put­er. What imme­di­ate­ly popped into my mind was, well, can com­put­ers be creative?

So over the decades, and since the 1980s, I wrote papers on cre­ativ­i­ty. Since the 1980s, I’ve touched on machines as well. In my present book, I deal with cre­ativ­i­ty in humans, and cre­ativ­i­ty in machines—with focus on machines.

Mason: There’s some­thing that’s the under­ly­ing the­sis of this entire book—or at least the under­ly­ing assump­tion in this entire book, which real­ly relies on the idea that the human brain is sim­i­lar to a com­put­er or sim­i­lar to an infor­ma­tion pro­cess­ing sys­tem. I just won­der how much do you actu­al­ly believe that that is a use­ful metaphor for under­stand­ing the human brain?

Miller: It’s extreme­ly use­ful, in that in the human brain, we have a mem­o­ry. Machines have mem­o­ries, also. We have long term and short term mem­o­ries, machines have a mem­o­ry. We process infor­ma­tion, machines process infor­ma­tion. What peo­ple seem to for­get about is that we’re essen­tial­ly machines—biological machines—that work on chem­i­cal and elec­tri­cal reac­tions, which can be changed by drugs; chang­ing our per­son­al­i­ty, for exam­ple. So there’s no rea­son to con­sid­er for exam­ple, that machines can­not be cre­ative. We can be cre­ative, machines can be cre­ative. A push­back on that is: How can any­thing made up of wires and tran­sis­tors be cre­ative? But we are made up of wet stuff, of fil­a­ments of veins and the organs and eccetera, and we can be cre­ative, too.

Mason: The only thing that makes me a lit­tle con­cerned about that assump­tion is under­stand­ing the human brain as a com­put­er. It feels to me a lit­tle bit reduc­tion­ist. It kind of takes away some­thing very spe­cial about what it means to be human. Some have called that a very mech­a­nis­tic, mate­ri­al­ist view of the brain. I just won­der if we are so will­ing to under­stand the human as a form of machine, does it obscure, per­haps, the true nature of cre​ativ​i​ty​.Is there some­thing hid­ing with­in human beings that’s very spe­cial, that gives rise to cre­ativ­i­ty, that maybe machines can’t replicate?

Miller: There’s no rea­son to attribute cre­ativ­i­ty only to humans. Yes, it is a very reduc­tion­ist view—we might say mech­a­nis­tic, mate­ri­al­is­tic—to use old jar­gon. But we are—our brain is—essentially made up of elec­trons, pro­tons, neu­trons, gluons…the whole zoo of ele­men­tary par­ti­cles. All of these par­ti­cles can, in prin­ci­ple, be under­stood using the laws of the equa­tions of quan­tum physics. These equa­tions, when put on a com­put­er, become num­bers. So it’s num­bers all the way down. Machines deal with num­bers too. So there’s no rea­son why machines can­not be cre­ative, can­not have emo­tions and can­not have con­scious­ness. It does­n’t take away any of the mys­tery of life. Because all of these ele­men­tary par­ti­cles act dif­fer­ent­ly in dif­fer­ent peo­ple. I mean, it’s obvi­ous­ly not an even sea. Some peo­ple are born smarter than others.

Mason: I just won­der if you can share your def­i­n­i­tion of what cre­ativ­i­ty is, because you help to define that with­in the first cou­ple of pages of this book.

Miller: My research on high cal­i­bre thinkers—highly cre­ative peo­ple, I’m try­ing to say. From that work emerged my the­o­ry of cre­ativ­i­ty. What emerged from that is that cre­ativ­i­ty can be defined as the pro­duc­tion of new knowl­edge from already exist­ing knowl­edge, accom­plished by prob­lem solv­ing. That def­i­n­i­tion includes both process and prod­uct. I look at prob­lem solv­ing with a four stage mod­el of con­scious thought, uncon­scious thought, illu­mi­na­tion and ver­i­fi­ca­tion. In the con­scious thought stage, the researcher sits at their desk and works con­scious­ly on a prob­lem, and the researcher will get stuck some­where along the line. The expe­ri­enced researcher will take a break. But you take a break only con­scious­ly, because the pas­sion and intense desire to solve a prob­lem keeps it alive in your uncon­scious, where it is turned over in ways not acces­si­ble to con­scious thought, owing to the bar­ri­ers and inhi­bi­tions there. Then it’s along these lines that the illu­mi­na­tion or prob­lem solu­tion crys­tallis­es and bub­bles up into con­scious­ness. In that con­scious stage—that’s the ver­i­fi­ca­tion stage—which is impor­tant because the solu­tion to a prob­lem does­n’t emerge in its final form. It has to be fixed up. Deductions have to be drawn from it. Now run­ning through all of this—lacing through all of this—are what I call char­ac­ter­is­tics of cre­ativ­i­ty, which again emerge from case stud­ies. Characteristics such as inspi­ra­tion; suf­fer­ing; per­se­ver­ance; focus; being out there in the world and hav­ing world­ly expe­ri­ences; unpre­dictabil­i­ty; prob­lem discovery—that’s a big one; and find­ing con­nec­tions between dis­ci­plines that every­body thought was uncon­nect­ed. So my view of cre­ativ­i­ty is fine grain and includes the angst of the cre­ative act, the lust for knowl­edge. As Picasso put it well, cre­ativ­i­ty is every­thing.

Now I claim that machines can have those char­ac­ter­is­tics, too. For machines to be ful­ly cre­ative, they must have emo­tions, and con­scious­ness. If you notice, some of these char­ac­ter­is­tics con­tain an emo­tion­al basis. The com­bi­na­tion of emo­tions and unpre­dictabil­i­ty is explo­sive and can give rise to the big idea.

Mason: The thing was so nice with­in those ear­ly chap­ters of the book is he focused on this idea of not just con­scious thought, but uncon­scious thought. We fix­ate so much when we talk about AI on con­scious beings, but you ele­vate the impor­tance of hav­ing this uncon­scious thought, or being able to walk away from a prob­lem, and then for the solu­tions to poten­tial­ly arise. But I strug­gled to come to terms with the idea that machines are capa­ble of uncon­scious thought.

Miller: It’s dif­fi­cult for machines to have con­scious thought at the present time because they have no con­scious­ness. You have to feed the prob­lem to the machine. My def­i­n­i­tion of cre­ativ­i­ty goes over to machines very eas­i­ly because machines work on accu­mu­lat­ed knowl­edge and they’re prob­lem ori­ent­ed. So a machine receives a prob­lem, or we put a prob­lem into it. It mulls over this prob­lem. It applies all the data in its data­base to it—that’s uncon­scious thought. Then it can be the illu­mi­na­tion when the machine solves the prob­lem. If you were talk­ing about a human being, you would say that that per­son takes a leap of intu­ition. So, you know, why can’t the machine do that? I’m not using intu­ition, in any mag­i­cal sense. Intuition is a skill which is honed. I mean, we’ve seen or heard about art con­nois­seurs tak­ing a look at a stat­ue that has emerged in an archae­o­log­i­cal dig. Taking one look at it and say­ing, It’s fake, it was plant­ed.” That’s not mag­i­cal in any way. That per­son has made a huge num­ber of mis­takes, and machines are mak­ing mis­takes also—when they are look­ing at a prob­lem and try­ing to fit data to it.

Mason: Again, my issue with this—the uncon­scious thought—is how you describe in the book: A human being can be doing some­thing else com­plete­ly alien or dif­fer­ent from what the prob­lem is that they’re try­ing to solve, and then sud­den­ly in that moment—that embod­ied moment—the solu­tion presents itself. That makes me think that the only way in which AI or machines can have these uncon­scious thoughts is that they have to be embod­ied. They have to be able to exist in the world and to be doing oth­er things that may not allow them to fix­ate on the prob­lem that they were pro­grammed for.

Miller: Well at this point, the machines can only do it in this closed off space. There will come a time when machines will be set into robots, as their brains. Then they will move about the world, and by touch, they’ll improve their notion of What is touch? What is inti­ma­cy? and so on and so forth. Then they’ll be doing some­thing else. But right now they can only do one thing, and that’s pret­ty good when they do that one thing.

Mason: Within that one thing that machines can do, I just won­der how many of those express­ing forms of cre­ativ­i­ty? Now in the book, you look at sev­en hall­marks of cre­ativ­i­ty. I just won­der how close machines are to achiev­ing some of those hall­marks of creativity—the machines that we have today?

Miller: Well, machines today can be inspired in the sense that the machine makes a good move in Go. Then the innards of the machine are adjust­ed, so that when the machine is in that cir­cum­stance again, it will make that move. It will be inspired to make that move. Just as when we make a good move in Chess or Go, we keep that in our mem­o­ry, and then when we see that sit­u­a­tion com­ing again we say, Aha, I can do that.” So we’re inspired to make that move.

Suffering will come along, okay. Suffering goes along with aware­ness, which machines don’t have. Machines don’t know whether they made a great move in chess or com­posed a great melody. But machines have a basic sense of aware­ness in that they’re aware of the prob­lem they’re work­ing on, and they’re aware of their own wiring. When their wiring is tam­pered with, they could give mes­sages, Something’s wrong here.”, and hope­ful­ly they won’t react like HAL does in 2001. Perseverance—machines per­se­vere. They don’t get tired, they go on and on, which is why it’s great to have robot arms assem­bling cars. Robot arms don’t come into work with any dis­tress from a bad week­end. Focus is focused on.

Machines have unpre­dictabil­i­ty, because although there are sep­a­rate parts which are assem­bled accord­ing to Newtonian physics with its deter­min­ism and causal­i­ty, when these parts are put togeth­er into a hole, they can exhib­it chaot­ic behav­iour. That is, say, unpre­dictabil­i­ty. Once machines have emo­tions, then they have true cre­ativ­i­ty. Combine that with unpredictability.

Problem dis­cov­ery and find­ing con­nec­tions between dis­ci­plines that don’t seem to be con­nect­ed. Machines can have that in a prim­i­tive sense, even at this point. Although cer­tain­ly when they have nat­ur­al lan­guage pro­cess­ing; that is to say, when they’re flu­ent in a lan­guage, then a machine can sur­vey the web and, say, look at the field of physics, look at a par­tic­u­lar area of physics and see that it’s clut­tered with redun­dan­cies and incon­sis­ten­cies. That’s usu­al­ly a sign that the wrong prob­lem is being worked on. What a machine can do at that point is to look at all the dis­ci­plines that are con­nect­ed with that field—even tangentially—ones that are all the way out in left field. Then all of a sud­den they can see that: Wow, one of these obscure ones real­ly has some­thing to do with his prob­lem sit­u­a­tion. That can lead you to some­thing new again, you have this leap of intu­ition. Then the machine might com­pose a new prob­lem and work towards solv­ing it. Now that that way of look­ing at sci­en­tif­ic research will have great div­i­dends already, to a cer­tain extent. We have machines that have a huge amount of knowl­edge. When a machine can read the web, it will have all knowl­edge that has ever been set up on the plan­et. Based on that knowl­edge, it can pro­pose hypothe­ses that can be rad­i­cal­ly dif­fer­ent from the human sci­en­tist who works on just accu­mu­lat­ed knowl­edge that he or she knows about. It is in that way that machines can come across knowl­edge that we don’t know any­thing about. The great move that AlphaGo made in its chess match in 2016. That was a move we might say of genius. It changed the game. People play dif­fer­ent­ly now, look­ing at the way machines play.

Looking even fur­ther at this and in the realm of the social sci­ences. When, say, we’re nego­ti­at­ing a treaty—the US is nego­ti­at­ing a treaty with Russia—and we might put some state­ments in the treaty. Then you might want to play the game, but, How will they react? If they react in this way, then we’ll do that.”, and so on. Playing these games—humans can play them but with their lim­it­ed realm of knowl­edge. Machines can play it with their much wider realm of knowl­edge. You can do this with music, too. Looking at dif­fer­ent gen­res of music that don’t seem relat­ed, machines will find the rela­tion. Maybe they’ll dis­cov­er or cre­ate a new kind of genre of music that we knew noth­ing about.

Generally, what I’m get­ting at is that machines can improve our lives. We’ve done a mis­er­able job on this plan­et, we haven’t planned ahead in any way. Machines can help us along those lines. Also we are merg­ing with machines, so it would be good if we went from human­i­ty 2.0 to 3.0, and 4.0 that will help us survive.

For exam­ple, if cli­mate change real­ly gets dif­fi­cult, it may be that we will have to change faster than the Darwinian time­frame. Another rea­son why I wrote this book is to give anoth­er view from the dystopi­an view of machines. The cul­tur­al side of AI.

Mason: It feels like there’s so much in the answer that I don’t know where to go next, but I do want to look specif­i­cal­ly at the case study that you just allud­ed to of AlphaGo, because in the book you look at these gam­ing AIs as exhibit­ing forms of cre­ativ­i­ty. AlphaGo had that move that was con­sid­ered by oth­er AlphaGo play­ers to be a very cre­ative act. What is AlphaGo, IBM Watson, Google’s DeepMind—what are these cur­rent AI prod­ucts teach­ing us about cre­ativ­i­ty today?

Miller: Well, they’re teach­ing us to be on the look­out, essen­tial­ly, for new ways of approach­ing prob­lems. What we’re also learn­ing is look­ing at how this process works. Researchers use essen­tial­ly two dif­fer­ent types of machines for this. They use sym­bol­ic machines, which are loaded with soft­ware data­bas­es and com­plex rules for using the data­bas­es. Our lap­tops, for exam­ple, are sym­bol­ic machines. Deep Blue, the machine that defeat­ed Garry Kasparov and cracked the game of chess is your lap­top on steroids. On the oth­er end of the spec­trum, our arti­fi­cial neur­al net­work machines—which are min­i­mal­ist machines that use a min­i­mum of input algo­rithms of soft­ware. Both pro­po­nents of both machines actu­al­ly claim that they are the way that the human brain works. So what these machines show us is how we use our data­bas­es. That the data­bas­es that we’re trained on allow us to react to the world in which we live.

Mason: What’s inter­est­ing is how you real­ly do argue for the cre­ativ­i­ty with­in the machine. The way in which you open up that pos­si­bil­i­ty is to see not just art as a cre­ative act, but to see sci­ence and to see prob­lem solv­ing also as a cre­ative act. How is art and sci­ence unit­ed through this thing called creativity?

Mason: On machines, AI cre­at­ed art. You have a new species of artists. The per­son who is the artist and tech­nol­o­gist rolled into one. That’s good because in the 21st cen­tu­ry, art and tech­nol­o­gy is merg­ing. Music and tech­nol­o­gy is merg­ing, and so is writ­ing and tech­nol­o­gy. Those are the new fron­tiers, where we’ll see the new works. I mean, even­tu­al­ly, machines will cre­ate art, lit­er­a­ture, music, of the sort that we per­son­al­ly can­not even imagine.

Mason: The book is lit­tered with these won­der­ful case stud­ies of where we’ve seen machine and AI artists, and the agency prob­lem. How do we deal with art­work that is cre­at­ed by machines? Does it always need a sub­jec­tive human view­point to see it as art? Will we ever get to the point where AI will cre­ate art­work pure­ly and only for oth­er AIs?

Miller: Well, that’s a big and good ques­tion. There’s always the issue of: Machines can cre­ate art, but are machines artists? But what you can do, for exam­ple, as a first approach to that prob­lem is you can put a web­cam on a robot artist. Take it out­doors and it will look around, and some­thing will attract its eyes. Some shape, some par­tic­u­lar geo­met­ri­cal shape. It will want to draw it, want to paint it. That’s a prim­i­tive sense of voli­tion, and freewill as well. That’s part of the issue—that peo­ple argue that machines aren’t out there. They’re not hav­ing world­ly expe­ri­ences. But what machines can do is to have flu­en­cy in a lan­guage, can read the web, and can read about thirst. It can vic­ar­i­ous­ly con­vince itself that it’s thirsty and us too, and then it can read about love and think: Well, that’s cool. It can watch movies and can learn more about love by watch­ing movies, watch­ing TV, eaves­drop­ping on con­ver­sa­tions between us all via the web. Then it can think to itself: Well, I had a good con­ver­sa­tion with a machine down the street. I think we’re in love. Then it can think about inti­ma­cy, and can learn more about inti­ma­cy by read­ing nov­els. Then it can be embod­ied in the robot—it can devel­op a sense of touch and feel more inti­ma­cy there. It will cer­tain­ly come to pass that arti­fi­cial inti­ma­cy will be the real inti­ma­cy. The line will van­ish between the arti­fi­cial and the human intimacy—there will be just inti­ma­cy. I mean, we have sex bots. They’re com­ing online, and imag­ine what that will do to human rela­tion­ships. But by that time, the notion of what a human being is will have been trans­formed, because it’s chang­ing faster than a Darwinian rate.

Mason: But again, it goes back to this agency issue. If machines have access to the wealth of human cre­at­ed knowl­edge, sure­ly the sorts of expe­ri­ences that it’s going to have are going to be defined by the ways in which humans have cod­i­fied their under­stand­ing of these things.

Miller: Right now it is yes. I mean, you could look at the var­i­ous pos­si­bil­i­ties you have. Right now, you don’t need very much tech­ni­cal abil­i­ty to do AI art​.You can buy an algo­rithm off the shelf that’s already been trained on say 50,000 images of peo­ple from the web, and then go home. What is often done is a per­son puts a self­ie into it, and then dials in the style of Van Gogh, and out will come some­thing in the style of Van Gogh, but it will be slight­ly dif­fer­ent­ly in that ele­ments of your face will pick up ele­ments of faces in the data set, and you will be slight­ly changed.

Then you might want to be cre­ative and put that back in and dial in the style of Cubism and then what will come out as a super­po­si­tion of your face as Cubist and Van Gogh. But it will again be changed even fur­ther, through fur­ther inter­ac­tion with fea­tures in your face with what’s in the data set. You could be more cre­ative, do that a few times and then pick out one that you can put into a con­test, say. Okay, so in that case, cre­ativ­i­ty sole­ly has human agency, but then you can go on and you can have sit­u­a­tions where the cre­ativ­i­ty will be shared between the human and the machine. That’s in Deep Dream, where a human wrote down an entire­ly new code and used an arti­fi­cial neur­al net­work trained on a huge image net data set of 14 mil­lion images. What came out is images that were far from what was in the data set. So the machine essen­tial­ly jumped the data set.

Then you can have humans and machines work­ing hand in hand, so to speak, each boot­strap­ping their own creativity—for exam­ple, in music. Francois Pachet, who is a musi­cian and com­put­er sci­en­tist, Director of the Spotify Creative Technical Research Lab has always been inter­est­ed in com­ing up with devices that can aid musi­cians and com­pos­ing, and play­ing music. One of them is a device which can improve musi­cians cre­ativ­i­ty in impro­vi­sa­tion. Pachet is keen on jazz, and he calls this device Continuator. It works along the fol­low­ing lines that the musi­cian begins to play on a piano, and that train of notes is fed to the con­tin­u­a­tor AI device which pars­es it out in phras­es. Then these phras­es are sent down a pipeline to a phrase analyser, which picks out pat­terns. It’s along these lines that the Continuator impro­vis­es in response to the musi­cian. The musi­cian then turns around and impro­vis­es in response to the Continuator. So impro­vi­sa­tion is often defined as a con­ver­sa­tion between musi­cian and musi­cal instru­ment. Here, it’s a con­ver­sa­tion between a musi­cian and AI.

Mason: When I strug­gle with is: Where does the artist begin and end? Is the non­hu­man agent the artist, to a degree? Is the human who’s pro­gram­ming the AI the artist? Or can it be con­sid­ered the only artist? For exam­ple, if you have a painter, it’s very clear that they’re using a paint­brush to manip­u­late paint on the can­vas and there­fore the agency of that artwork—or the agency of the per­son who’s cre­at­ing the artwork—sits with the human artist. But if they were paint­ing with slime mould for exam­ple, and it would land on the can­vas and then grow across the can­vas and cre­ate its own mor­phogenic forms, the ques­tion then becomes: Is the artist the human, or iis the artist the slime mould? In all of those sorts of exam­ples you just gave, how do we give cred­it to the cor­rect artists, or do those things real­ly mat­ter? Is it always going to be a human artist if there is a human in the loop in the cre­ation of the work?

Miller: Sometimes agency is shared between human and machine. In that case, it’s often a sit­u­a­tion where the art­work is signed off with the sig­na­ture of the human and the machine. When machines have voli­tion and free will then they will be real artists. That is not yet total­ly the case. Then the machine will have agency.

Mason: I guess then the prob­lem becomes—once this work’s created—who gets to appre­ci­ate it? Is it always going to be human sub­jec­tiv­i­ty, human audi­ences that are going to define some­thing as art­work, or can the AI also con­tribute to whether they see some­thing as an artis­tic endeav­our or not?

Miller: Well, there will come a time when the AIs will have emo­tions and con­scious­ness and then they can appre­ci­ate the work that they’ve done, and they’ll do that work for us and their brethren.

Mason: Right now today, do we as human beings, and does the art world even appre­ci­ate machine gen­er­at­ed art? Is it the fact that it’s instant­ly replic­a­ble, it seems very easy to do—as you just said, you can plug some­thing into a com­put­er, press a but­ton and out comes the art­work. I guess, will we ever appre­ci­ate machine art? Could there be an argu­ment that we will nev­er appre­ci­ate machine gen­er­at­ed art until there is a mon­e­tary val­ue put on to it?

Miller: I think that whilst we bear in mind the ques­tion, Can machines be cre­ative? Can they cre­ate art?”, we should also bear in mind the ques­tion, Can we learn to appre­ci­ate art that’s been cre­at­ed by machines?”. From a mon­e­tary view, yes there is appre­ci­a­tion. At an auc­tion at Sotheby’s last year, a piece by the very promi­nent and bril­liant AI artist Mario Klingemann was sold for $50,000. Just a few months before that, a much less cre­ative work of art was auc­tioned by Christie’s for $432,000. The drop in price might have been because the unique­ness of AI art wore off and there was a big tumult about this $432,000 sale. But nev­er­the­less, AI art has hit the high end of the art mar­ket, and it is being appre­ci­at­ed up there. I mean, it’s a great con­ver­sa­tion piece for your liv­ing room, for example—to see a work of art con­stant­ly chang­ing. So it is of mon­e­tary val­ue, and it is in appreciation.

Mason: In exam­ples like that. It is the algo­rithm that’s cre­at­ed to gen­er­ate the art­work also seen as part of the artwork—as an art­work with­in itself?

Miller: It’s some­thing that is pro­duc­ing art. As time goes on, as machines become more sophis­ti­cat­ed, they will have voli­tion and will be pro­duc­ing art them­selves. The point is that you want to have the machine come up with a problem—the art prob­lem that it wants to solve. Right now that is inputted. We have to kick off the process.

Mason: Does the prob­lem of true cre­ativ­i­ty get sti­fled by the fact that it’s humans at one end that define the art prob­lem to expect some sort of art out­put?

Miller: That’s right. Precisely as I’ve said, the human has to kick off a sit­u­a­tion. Machines are not yet at the point where it can choose a subject.

Mason: It feels like from read­ing the book, your hope is that even­tu­al­ly the human will no longer need to be in the loop.

Miller: That’s right. As I dis­cussed previously—that chain of first per­son with the self­ies, and then Deep Dream and so on—the last step of that is machine work­ing by itself. Now sim­i­lar­ly, right now it’s most­ly the sit­u­a­tion where we have humans and machines work­ing togeth­er, but even­tu­al­ly machines will go off on their own. In lit­er­a­ture, they may pro­duce lit­er­a­ture that we can­not under­stand what­so­ev­er. They’ll have to explain it to us. And maybe at the end of the day it’ll be bet­ter than what we can do.

Mason: In the book, you focus on three things that AI can cur­rent­ly cre­ate in the cys­tic domain. That’s visu­al art­work, lit­er­a­ture, and music. I just won­der why you focused on those three? Is it because they’re the low­est hang­ing fruit in terms of what AI can actu­al­ly gen­er­ate? Or are they very spe­cif­ic cre­ative endeav­ours that you feel are at risk by these AI entities?

Miller: No, not at all. I dis­cussed those three because an aim of my book is to dis­cuss the upside of AI—the cul­tur­al side of AI—which is some­thing that’s not dis­cussed very often. Most of the peo­ple in AI are engi­neers who deal with dri­ver­less cars and drones, and things of that sort, which is all well and good. A small sub­set deals with AI cre­ativ­i­ty. Also if you speak with engi­neers in AI who are not involved in AI cre­ativ­i­ty, they haven’t the fog­gi­est idea what goes on in that area, and they’re not inter­est­ed in it either. Unfortunately, dystopi­an books are very pop­u­lar with the pub­lic, and these dystopi­an sce­nar­ios aren’t going to take place for maybe 100 years. These days, it’s dan­ger­ous to pre­dict more than five years ahead owing to how fast things are chang­ing. So 100 years from now, we’re going to be very dif­fer­ent. Our rela­tion to machines is going to be very dif­fer­ent. A good many peo­ple on the plan­et will be most­ly machines. I mean, I think there’ll be three sorts of life liv­ing on this plan­et. Human beings who are not merg­ing with machines, who pre­fer not to; human beings that are; and machines.

Mason: If we were to have that sub-speciation of cyborgs, of ani­mals and of humans and then machines, and AI—do you think they would each live in dif­fer­ent artis­tic and aes­thet­ic domains? Do you think that each would have their own artis­tic lan­guage? Or do you think there’ll be some new emerg­ing aes­thet­ic lan­guage to describe this co-living of dif­fer­ent forms of species?

Miller: So there will cer­tain­ly be a dif­fer­ent aes­thet­ic lan­guage, what­ev­er aes­thet­ics means. But it may be that these three dif­fer­ent species may be liv­ing in dif­fer­ent areas—they may not coex­ist. Putting aside the humans who don’t want to be transhuman—they’re going to be left behind, I think. What we want is to be in good rela­tions with machines, all the way through. Teaching machines to be cre­ative can be extreme­ly use­ful here. But again, we can’t pre­dict what our rela­tions with machines will be. Indeed, there may be rogue machines—there are rogue peo­ple too. We cre­ate machines. You have to begin somewhere—machines have our cre­ativ­i­ty. There will be—perhaps not too dis­tant future—what’s called arti­fi­cial gen­er­al intel­li­gence. That will be where machines are as intel­li­gent as us. They will have evolved a set of emo­tions, con­scious­ness and cre­ativ­i­ty which are dupli­cates of ours. So there will be the line between arti­fi­cial intel­li­gence, which is an oxy­moron, and nat­ur­al intel­li­gence will disappear—there will just be intel­li­gence. Then the next step—who knows how long that will be, maybe not too long—there’ll be arti­fi­cial super intel­li­gence, where machines may have evolved a set of emo­tions and con­scious­ness which are total­ly dif­fer­ent from ours, what­ev­er that might be. But one thing is for sure that machines at that point will have a cre­ativ­i­ty that far out­strips us, because machines have the poten­tial for unlim­it­ed creativity.

Mason: Will machines even val­ue cre­ativ­i­ty? It feels like all of the dis­cus­sions around super intel­li­gence are mov­ing towards ideas of the sin­gu­lar­i­ty. It’s mov­ing towards greater effi­cien­cy and greater eco­nom­ic effi­cien­cy, and machine-like effi­cien­cy on how we oper­ate and live in the world. Will the cre­ation of art­work be seen by these super intel­li­gences as a point­less endeav­our, perhaps?

Miller: Oh no, they will val­ue aes­thet­ics. They will val­ue cre­ativ­i­ty high­ly. They’ll have super cre­ativ­i­ty. As a mat­ter of fact, they can work along with us and we can put our own unique cre­ativ­i­ty in there, as well. And again, we’ll be entire­ly dif­fer­ent. You can’t even pre­dict what sort of mind­set we will have. But cre­ativ­i­ty will still be val­ued. The works of art they pro­duce, the lit­er­a­ture and music…again, we can’t even imag­ine what they will be. But I picked those three sub­jects, because they’re their cul­tur­al sub­jects. And I want­ed to indi­cate a cul­tur­al side of AI.

Mason: Let’s focus on one of those cul­tur­al ele­ments, which is lit­er­a­ture. Because there’s some­thing so love­ly that you say in the book about the fact that AI lit­er­a­ture actu­al­ly chal­lenges our notions of what we per­ceive to be non­sense and what we per­ceive to be poet­ry. In actu­al fact, some of the AI lit­er­a­ture that we’re see­ing and in some of the case stud­ies that you give with­in the book, we’re actu­al­ly see­ing the sorts of out­puts that are very sim­i­lar to the big poets or to William Burroughs, and that seems very hope­ful. That there’s a degree of cre­ativ­i­ty, at least in the lit­er­a­ture space.

Miller: Yes. and there’s a degree of abstrac­tion, too. Some of these texts are rather nom­ic. They have very lit­tle sub­jec­tive mean­ing, one might say. In that sense, the read­er becomes the writer, okay, so there may be text along those lines. But cer­tain­ly, machines can trans­form the land­scape of lan­guage. They can trans­form the progress of lit­er­a­ture just as they’ve trans­formed the progress of art and music as well. There’s pro­gram­ming involved in some of this, espe­cial­ly when it’s done with sym­bol­ic machines—less so with arti­fi­cial neur­al net­works. Nevertheless, one should not crit­i­cise this work as being just the result of pro­gram­ming, and machines are the ser­vant of the pro­gram­mer. I always like to tell the sto­ry of Leopold Mozart. His father taught him the rules of music, but we don’t cred­it the father with the son’s music. Similarly, we don’t cred­it the AlphaGo team with the great moves that AlphaGo made. So you have machines jump­ing their pro­grammes, jump­ing their algorithms.

Mason: The one thing that we’re still strug­gling with when it comes to at least AI cre­at­ed lit­er­a­ture is the idea of humour, and the idea of sar­casm, and there’s a good argu­ment to be made. The rea­son why we strug­gle to pro­gramme that into AI is because they’re not present in the world. They have no con­text on the human world to make the sorts of com­par­isons that give rise to things like puns. So is there an impor­tant step that we need to take, to take this AI out of its black box where it’s cre­at­ing art­work and find a way to embody it so that it can expe­ri­ence the lived real­i­ty, the lived world? To tru­ly cre­ate art­work in a way sim­i­lar to the ways in which human beings cre­ate art­work, by tak­ing inputs from the world through our sen­so­ry organs, and then—through uncon­scious thought or otherwise—we’re then able to process that and out­put this thing that we’ve notion­al­ly called artwork?

Miller: Absolutely. I mean, what one needs is machines to be tru­ly out there in the world as the brains of robots. Experiencing their con­tact build­ing up, sort of like the way the psy­chol­o­gist Jean Piaget thought of a child con­struct­ing real­i­ty; the robot moves around the world and sees the rela­tion­ship between objects and then builds up a notion of space and a notion of time. What one needs per­haps is an Einstein for AI, who will say, No, you’re all work­ing on the wrong prob­lem. What you need is…” What we know we need is a real­ly rev­o­lu­tion­ary archi­tec­ture. Artificial neur­al net­works are doing a great job now, but there may be some­thing bet­ter than that. After all, you look back at Leonardo da Vinci and oth­ers who were study­ing flight, and they put wings on their arms and flight was…well, if you emu­late birds, you’ll get there. But a 747 aero­plane does not fly like a bird. So it may be the case that progress in AI will not pro­ceed and will take off from what’s going on. But actu­al­ly what’s being researched these days is called hybrid machines—machines that are part sym­bol­ic machines, part arti­fi­cial neur­al net­work machines. In fact, some very inter­est­ing work has gone on in Deep Mind where they’ve suc­ceed­ed in hav­ing an arti­fi­cial neur­al net­work evolve a symbolic-like algo­rithm. Because the holy grail at Google is for end-to-end train­ing using only arti­fi­cial neur­al net­works, and noth­ing else. It’s doing a great job now, but it may be the case that some­body will come along and say, Well, look. Here we have this very inter­est­ing archi­tec­ture.” It may be that an archi­tec­ture of the future will be part wet, part human cells and part machine.

Mason: Should we be look­ing at not arti­fi­cial intel­li­gence for where the future artists may come from, but arti­fi­cial life? Should we be look­ing at the sorts of works that are being cre­at­ed in col­lab­o­ra­tion with non-human agents like bac­te­ria or slime mould? Are we clos­er to hav­ing these ful­ly realised art­works that emerge from the cre­ation of arti­fi­cial life than we are the cre­ation of arti­fi­cial intel­li­gence? Are we look­ing in the wrong place? Should we be look­ing at vital­i­ty itself rather than intel­li­gence as the thing that we put on the


Miller: We don’t know. The point is, the great progress right now is being made in AI with both sym­bol­ic machines and arti­fi­cial neur­al net­works. Artificial life has giv­en rise to some inter­est­ing music com­po­si­tions such as in the hands of Eduardo Miranda at Plymouth University.

Mason: There are so many won­der­ful case stud­ies in the book of exam­ples where machines are gen­er­at­ing art­work. Is there one, par­tic­u­lar­ly, that real­ly encom­pass­es all of the things that you’re try­ing to grap­ple with when it comes to this idea of cre­ativ­i­ty? Is there an art­work which is exem­plary in terms of show­ing all of those dif­fer­ent forms of creativity?

Miller: Perhaps what comes to mind is Ahmed Elgammal’s cre­ative adver­sar­i­al net­work, which cre­ates new art styles.

Mason: And yet, when we look at some­thing that we can’t help but ref­er­ence pre exist­ing styles. Can we real­ly find some­thing that is tru­ly new, and will it come from machines rather than from humans?

Miller: It will come from both, but what one wants to see is this com­ing from a machine. I can’t real­ly answer that ques­tion right now because machines are just not at that stage but the glim­mers of cre­ativ­i­ty are absolute­ly amaz­ing. Again, AlphaGo in the spec­tac­u­lar move it made in its match with Lee Se-dol; a move that was­n’t sup­posed to be made at that point in the game, but it made it and it changed the way Go is played, and it also won the match. Even with a prim­i­tive sym­bol­ic machine, Deep Blue, which defeat­ed Garry Kasparov in 1997. The 44th move­ment made in the first game of the match—it was a mid-game move, and it faced a sit­u­a­tion and it could­n’t find what to do from its play­book. So it jumped the play­book and came up with a spec­tac­u­lar sac­ri­fice. Then there’s Deep Dream, which pro­duces some extreme­ly inter­est­ing art­work. In fact, it kicked off a whole new art style, which is still being used today. There’s cre­ative adver­sar­i­al net­works, gen­er­a­tive adver­sar­i­al net­works where machines can dream essen­tial­ly, and pro­duce inter­est­ing art­works. There’s Continuator which I men­tioned before, Francois Pachet’s Continuator, which pro­duces new impro­vi­sa­tion­al melodies which are very pleas­ing to the ear. There’s the so-called 90-second Melody pro­duced by an arti­fi­cial net­work machine. It’s the first melody pro­duced by an arti­fi­cial net­work machine and the first melody pro­duced by any machine which was not pro­grammed and with data­bas­es for pro­duc­ing music. Then there’s the sur­re­al prose that’s pro­duced by arti­fi­cial neur­al networks.

When I first thought of this work, I knew a lot about AI cre­at­ed art and AI cre­at­ed music, but not much about AI cre­at­ed lit­er­a­ture and I thought: Well am I going to have enough? I had more than enough. So no, I can’t real­ly focus on one piece of work there. Wonders are pro­duced by AI.

Mason: What you’re very good at is look­ing at how cul­ture and soci­ety, how the cur­rent sci­ence affects the way in which cer­tain aes­thet­ics or art­work emerges. Famously, you’re known for look­ing at the rela­tion­ship between Einstein and the work of Picasso and I just won­der, is the work or the machine gen­er­at­ed work that we’re see­ing today—is it actu­al­ly hav­ing an inverse effect? Is it chang­ing the way in which human artists cre­ate work?

Miller: Yes, absolute­ly. AI cre­at­ed art­work has changed the way that artists paint and work in a num­ber of ways. One very inter­est­ing way is that artists now train an arti­fi­cial net­work machine on their own art­work, and then have the machine pro­duce their own art­work, essen­tial­ly, but vari­a­tions on their own art­work. Some cer­tain vari­a­tions will be sur­pris­ing, and then the artists will use that vari­a­tion in their own painting.

When humans work with machines, we have our own data­base of knowl­edge, which is not as deep as machines. Say machines and sci­en­tists are work­ing togeth­er on a prob­lem. The machine can affect the sci­en­tist’s cre­ativ­i­ty in that it can, in an instant, read all the papers on a par­tic­u­lar sub­ject, and relate that infor­ma­tion to the sci­en­tist, who can then adjust the hypothe­ses that he or she is mak­ing on a par­tic­u­lar prob­lem. So it kind of has a human’s creativity,

Mason: Your desire to put AI on a pedestal is very much to do with an obses­sion that you seem to have with genius­es and high achiev­ers, and you’re look­ing at AI as poten­tial­ly the next exam­ple of genius out in the world. Where does that fas­ci­na­tion with genius­es and with high achiev­ers come from?

Miller: A genius is some­body who has extra­or­di­nary intel­lec­tu­al abil­i­ties which do not arrive by means of delib­er­ate prac­tice. It arose from read­ing about high achiev­ers like Einstein, Picasso. People who are looked up to by all peo­ple in that pro­fes­sion. That’s where my pen­chant for high achiev­ers came from. What one can do is to study them—which is what I’ve done—by apply­ing var­i­ous psy­cho­log­i­cal the­o­ries such as Piaget’s epis­te­mol­o­gy, Gestalt psy­chol­o­gy, Freudianism and so on, and try to get some indi­ca­tion into how they thought. There’s a para­dox here—that I’m using psy­cho­log­i­cal think­ing the­o­ries that were for­mu­lat­ed from the think­ing of aver­age peo­ple and apply­ing that to above aver­age peo­ple, but again you have to begin somewhere.

Mason: It feels like in some of your work that you have a desire to find the for­mu­la for genius, and yet it feels like what unites the sorts of genius­es that you obsess about or are inter­est­ed in, is the fact that they’re all com­plete­ly different.

Miller: Yes, they’re all com­plete­ly dif­fer­ent. But, there’s a fun­da­men­tal­i­ty there in that they’re all extreme­ly smart. One of the goals of my work is to try to get some han­dle on how they think. Though we can nev­er think at the lev­el they do, how they work can help our thinking.

Mason: When it comes to genius, are you try­ing to find a quan­ti­ta­tive val­ue that points at genius? For exam­ple the IQ test. We can say that if some­one has a cer­tain IQ above a cer­tain val­ue, there­fore they are a genius. Sometimes, what makes some­one a genius isn’t relat­ed pure­ly to intel­li­gence. When we look at genius and equate that to the obses­sion around arti­fi­cial intel­li­gence, is genius all about intel­li­gence or is there some­thing else going on? Is some genius not always pure­ly on intel­li­gence? Is it reliant per­haps on some­thing like creativity?

Miller: Well, cre­ativ­i­ty is part of intel­li­gence. Intelligence is kind of a fuzzy thing, and there are var­i­ous sorts of intel­li­gence. Emotional intel­li­gence, sci­en­tif­ic intel­li­gence, log­i­cal intel­li­gence, musi­cal intel­li­gence and so-on—and that’s of course relat­ed to cre­ativ­i­ty. There’s a dif­fer­ence between tal­ent and cre­ativ­i­ty. Someone can be very tal­ent­ed but not be cre­ative. Someone can be a very tal­ent­ed piano play­er and won’t make any mis­takes but there’s some­thing just miss­ing in their ren­di­tion, where­as a cre­ative pianist adds some­thing to it and draws out what’s in the score. The con­verse is true, in that if you’re cre­ative then you’re some­what tal­ent­ed. There have been high­ly cre­ative musi­cians who have turned out not to be great pianists, but they’re not bad any­way. Yes, machines have the capa­bil­i­ty of genius think­ing. Genius is a way of thinking.

Mason: There’s some­thing so nice you said there, about how you can be an extreme­ly tal­ent­ed pianist and play all of the right notes in the right order at the right time. I just won­der if that’s where we are right now with AI—they can repli­cate music, they can play all the right notes at the right time, on the right beat…but they per­haps don’t have that cre­ativ­i­ty to go slight­ly off-beat. You can talk to drum­mers about the cre­ativ­i­ty with­in how Ringo Starr used to drum—he used to drum off the beat—and yet machine gen­er­at­ed algo­rithms that are used in music pro­duc­tion soft­ware demand often to put the drum beats on the beat, which would make the beat of sound absolute­ly awful.

Miller: That’s the embod­i­ment prob­lem too. Machines have to be embod­ied into robots so that they can use their body when they play the piano. You see great and cre­ative pianists putting their bod­ies into it. They use their emo­tions and so-on, instead of sit­ting up straight in a chair and just mak­ing sure they hit all of the right notes.

There is, in my book, a case where a musi­cal score is played by a machine and there’s no emo­tion in it what­so­ev­er. There’s no loud­ness and soft­ness, and so-on. The same score is played by a pianist and it sounds entire­ly dif­fer­ent. In fact, I think there was an exper­i­ment where Beethoven’s Moonlight Sonata was played by a machine. It was just plod­ding along. When played by a per­son who puts their body into it and their empa­thy into it, it sounds entire­ly different.

Mason: So you think to solve the prob­lem of that ephemer­al expe­ri­ence you get when you wit­ness human cre­at­ed music or art­work, all you have to do is ensure that the machine that is play­ing is embodied?

Miller: No you want the machine not only to be embod­ied but to have emo­tions. You need emo­tions and con­scious­ness for full cre­ativ­i­ty, whether it’s doing art, sci­ence, lit­er­a­ture and cer­tain­ly music.

Mason: What I’m real­ly try­ing to ask you, Arthur, is do you think there’s some­thing spe­cial and unique about human beings that can nev­er be repli­cat­ed by machines, that can nev­er be sur­passed by machines?

Miller: No absolute­ly not. When you say human beings, you’re talk­ing about human beings right now. But human beings will change and machines will pick up all of our char­ac­ter­is­tics. Indeed, when it gets to a point of arti­fi­cial super intel­li­gence and machines start repro­duc­ing them­selves, there will be some of our DNA in there, so to speak. We’ll always be bound togeth­er. But no, there’s noth­ing spe­cial about human cre­ativ­i­ty. There’s noth­ing spe­cial about us. We’re just an acci­dent. Some bil­lions of years ago, an iso­tope of car­bon in the sun just spat out atoms that land­ed on Earth, and here we are. There could be life­forms else­where. This notion of look­ing at exo­plan­ets for the so-called Goldilocks Zone. Life as we know it need not exist any­where else in the Universe.

Mason: I strong­ly strug­gle to believe that there is noth­ing spe­cial about human beings, because in many cas­es artists cre­ate art­work to express the nature of the human con­di­tion. It’s a very human process to gen­er­ate and cre­ate artwork.

Miller: Back in Babylonian times, Babylonians had a cos­mol­o­gy. The Earth stood in the cen­tre, plan­ets moved around it, and that last­ed until Copernicus and Galileo. The Earth has moved out of the centre—the sun is in the centre—but that’s okay because that’s our uni­verse. Then the galaxy was revealed, and we’re just one of many star sys­tems. Now we have a mul­ti­verse, where our uni­verse isn’t even unique. So no, there’s no rea­son why humans should be unique either. In fact, as I wrote at the end of my book, there will come a time when this whole enter­prise will run down. The uni­verse will run down, and the only things that might be left are robot­ic forms, and they’ll be smart enough to know how to get into anoth­er uni­verse. Once they land on a plan­et, it may not be like ours and they’ll have cre­ative­ness of their own, and this whole process will start all over again. So, there’s no rea­son why we should be unique.

Mason: Well at the end of the book, you prof­fer some chal­leng­ing notions of what the future may look like. One of those is the idea of the cyborg artist. The com­bin­ing of humans with AI or machine parts. I could­n’t help but think that we already have cyborg artists today. We have peo­ple like Neil Harbisson, the colour­blind artist, who has an anten­na that’s sur­gi­cal­ly attached to his head. The inter­est­ing thing about Neil is that although he’s a colour bind indi­vid­ual and the anten­na allows him to hear colour, he defines it as an entire­ly new sense because what the anten­na’s actu­al­ly doing is con­vert­ing light waves to sound waves. Instead of putting those through the tra­di­tion­al aur­al sens­es of the ears, what it does is it vibrates his skull, so he’s expe­ri­enc­ing this entire­ly new sen­so­ry modal­i­ty and then he cre­ates these artworks—these very colour­ful, rep­re­sen­ta­tion­al paintings—that only he can tru­ly expe­ri­ence, because they’re son­ic. Humans that don’t have an anten­na, or indi­vid­u­als who aren’t enhanced aren’t able to have the same expe­ri­ence with an art­work as some­one who has an anten­na or who has been enhanced can have. Are we going to get to the point where you’re going to have to start enhanc­ing your­self to expe­ri­ence the full spec­trum of pos­si­ble future artworks?

Miller: These art­works will be very dif­fi­cult for us to under­stand right now, at this time. Even look­ing at elec­tron­ic art—the best way to appre­ci­ate that is to know some­thing about com­put­ers. It isn’t the sort of art that you see in the Tate Britain where you can walk up to it and put your hands behind your back, look­ing at it from var­i­ous angles and see­ing the paint swirls and things like that. AI art will be even more dif­fer­ent than elec­tron­ic inter­ac­tive art. We will be able to appre­ci­ate it. We have to change our mind­set and realise that this is art not pro­duced by human beings, but by a machine. As time goes on, and we become clos­er to machines, and machines pro­duce art that we might not under­stand at all, they’ll have to explain it to us—just as they’ll have to explain their own prose to us as well in order for us to appre­ci­ate it.

Mason: But will we have to actu­al­ly fun­da­men­tal­ly change our minds? Will we have to enhance our minds to be able to have the expe­ri­ence with a piece of AI cre­at­ed artwork?

Miller: We’ll be tran­shu­mans by that time. Our modal­i­ties will be stim­u­lat­ed in a vari­ety of ways.

Mason: Today, right now, as we’re increas­ing our sci­en­tif­ic knowl­edge about the world and learn­ing more about ideas like quan­tum mechan­ics and weird­ness and how things are extreme­ly dif­fer­ent when they go very, very small—how is that impact­ing the way in which artists will start to see the world?

Miller: As far as the quan­tum world…certainly to me, the quan­tum world is not weird. That’s a mis­take. That’s some­thing that peo­ple try and sell books with. It seems to us to be strange because it’s against our Aristotelian intu­ition which is that heav­ier objects fall faster than lighter ones—and they do—except in a vac­u­um, and intu­itions of that sort. That objects have a place and that’s it, they’re not dis­em­bod­ied. In the quan­tum world, an elec­tron can be every­where until you make a measurement.

Something that’s inter­est­ing, which is to come yet, is art pro­duced by quan­tum com­put­ers. Quantum com­put­ers are at a fair­ly rudi­men­ta­ry lev­el now but they have an awful lot of mon­ey behind them present­ly, for finan­cial purposes.

Mason: Is that art that could be in every sin­gle gallery simul­ta­ne­ous­ly until you observe it?

Miller: Very good. I’m just think­ing of get­ting a piece of art out of that machine, because you can’t touch a quan­tum com­put­er while it’s work­ing. You then destroy the entan­gle­ment. That’s a prob­lem that’s yet to be solved.

Mason: What is strik­ing­ly clear in the book is that you real­ly do believe our future machines will sur­pass human beings in terms of cre­ativ­i­ty and that AI gen­er­at­ed art will exceed human artist’s wildest dreams. Does this then mean artists become obso­lete? Will we have the obso­les­cence of the human artist, or will we con­stant­ly desire human cre­at­ed work because we’ll see machine gen­er­at­ed work as some­thing that comes from the uncan­ny val­ley or the [inaudi­ble] val­ley? Could it be some­thing that’s very dif­fi­cult for us to let go of?

Miller: Looking at elec­tron­ic art, a lot of peo­ple in elec­tron­ic art call por­trai­ture, flat paint­ing.” People who do brush and paint will go on—that will go on. That art will still be expen­sive to buy, so peo­ple will turn to AI cre­at­ed art which will be much cheap­er. There will be AI art done by machines by them­selves, and then there will be art done by human beings in con­junc­tion with machines, com­bin­ing the unique aspects of both of them.

Mason: Finally, I want to look at what you end the book with, which is this play­ful provo­ca­tion that in actu­al fact, what may end up hap­pen­ing is when we sub-speciate and some indi­vid­u­als become cyborgs, some indi­vid­u­als choose to remain human and then we have these AIs that are then able to tran­scend bio­log­i­cal form and go off into the stars and pop­u­late oth­er plan­ets, that they will be able to then final­ly cre­ate, unhin­dered by the lim­i­ta­tions that we’ve set our­selves here on Earth. How far do you actu­al­ly believe that space explo­ration in itself will be a form of cre­ative endeav­our that is per­formed by our machine successors?

Miller: These new forms of life, these cyborgs of life, which may have wet insides—we don’t know—they’ll find a way to move into anoth­er uni­verse. What I’m say­ing is that they’ll pop­u­late these plan­ets and there will be maybe an Adam and Eve to start a whole new race over there of human beings. We may be like that. Our insides are wet, but we may be the end result of a lot of experimentation.

Mason: Or a lot of creativity.

Miller: A lot of cre­ativ­i­ty too, yeah. Right now, the best way to do space explo­ration is with robots and it will con­tin­ue like that for a long time. The way to get to oth­er star sys­tems will only be through dis­cov­er­ing a worm­hole. People like that state­ment, sci­ence fic­tion fans like that.

Mason: It’s a beau­ti­ful idea, but you talk about these machine enti­ties or these cyborg enti­ties that will go off and pop­u­late oth­er plan­ets. I just won­der how space itself will change the way in which we cre­ate and gen­er­ate work. How will we cre­ate art­work, or sound, even, in vac­u­ums of space?

Miller: Those life forms of the far dis­tant future will be gen­er­at­ed from us, essen­tial­ly. So they will need grav­i­ty and the prop­er­ties of the Earth—whatever they may be—when they leave.

Mason: Is it going to be a case of slow attri­tion? Are we of the mind­set now where we are so reluc­tant to see cre­ativ­i­ty with­in machines because we’re scared of what that means?

Miller: Precisely.

Mason: Will it be the case that we’ll see incred­i­bly cre­ative out­puts, some­thing that will vis­cer­al­ly affect us so much—like a great piece of artwork—and we’ll be sur­prised to learn that it was actu­al­ly cre­at­ed not by a human, but by a machine?

Miller: Very much so, yeah. In the course I lec­ture on, some­times I play the Bach game. You play two short excerpts of Baroque music. One was writ­ten by Bach, the oth­er was cre­at­ed by an algo­rithm. You ask the audi­ence, Which one was writ­ten by Bach?” and fifty per­cent of them always get it wrong. You play with them a bit and you say, Well, I’m sure you chose the oth­er option because it was some­what melo­di­ous, some­what beau­ti­ful, some­what inspirational—just like Bach’s music. But it was writ­ten by a machine! How did Bach do it? Can a machine write music of that qual­i­ty?” But that’s the wrong ques­tion, because we want machines to go beyond that to pro­duce music of their own. They will blow our minds per­haps, like Bach’s music did. Bach was done by Bach, and I agree with researchers who dis­agree with plac­ing too much empha­sis on these Bach-or-not games where we play with Bach in my lec­ture. Some lec­tur­ers do the same with art. That’s the point. We should not judge the work of an AI based on whether it can be dis­tin­guished from work done by us, because what’s the point? We want AIs to pro­duce work that we present­ly can’t even imag­ine, that may seem to us mean­ing­less and even non­sen­si­cal, but may be bet­ter than what we can produce.

Mason: Could it be the case that AI artists aren’t there for our amuse­ment, aren’t there to blow human minds? In actu­al fact, they may cre­ate some­thing which is uncom­fort­able to lis­ten to, some­thing that vis­cer­al­ly makes us sick or dis­turbed or dis­gust­ed. In actu­al fact could that be the paragon of AI art and it’s just some­thing that human beings just don’t find very appealing?

Miller: Look at the music of Stravinsky. It was a long move from Mozart to the Sex Pistols. That sort of music makes some peo­ple sick. Even Beethoven’s Third Symphony—people were very dis­turbed with that. That’s tru­ly a val­ue judgement.

Mason: Will AI choose to cre­ate some­thing that can­not be observed, can­not be appre­ci­at­ed, can­not be read, can­not be seen, can­not be lis­tened to?

Miller: By us, you mean?

Mason: By human beings and their human senses?

Miller: That’s right, they’ll all write music for them­selves and for their brethren.

Mason: Will we even know if it’s happening?

Miller: Well no, we won’t be able to judge their music. They’ll be able to judge our music, but maybe by that time again humans will be chang­ing, and we may be in such a posi­tion that we can hear the fre­quen­cies that dogs hear.

Mason: Until we become enhanced I guess now we just have to sit and wait, and see what hap­pens with the future of this machine artwork.

Miller: And see how we are trans­formed as well. We will be trans­form­ing along with how machines pro­duce art. These tech­niques will change.

Mason: On that note, Arthur, thank you for your time.

Miller: Thank you.

Mason: Thank you to Arthur, for shar­ing his vision for how art­work might be cre­at­ed in the near future. You can find out more by pur­chas­ing his book, The Artist in the Machine: The World of AI-Powered Creativity—avail­able now.

If you like what you’ve heard, then you can sub­scribe for our lat­est episode. Or fol­low us on Twitter, Facebook, or Instagram: @FUTURESPodcast.

More episodes, tran­scripts and show notes can be found at Futures Podcast dot net.

Thank you for lis­ten­ing to the Futures Podcast.

Further Reference

Episode page, with intro­duc­to­ry and pro­duc­tion notes. Transcript orig­i­nal­ly by Beth Colquhoun, repub­lished with per­mis­sion (mod­i­fied).

Help Support Open Transcripts

If you found this useful or interesting, please consider supporting the project monthly at Patreon or once via Cash App, or even just sharing the link. Thanks.