At Secret Cave, we’ve always had an obsession with the output of artificial intelligence. Across a variety of fascinating bots and other projects, it’s been shown that their artwork is more than worthy of scrutiny. Since the birth of the internet, and the entropic prevalence of technology over the past century, the capabilities of such bots has improved at a vast rate. It’s allowed certain older ideas, about the creation of art without emotion, to be explored in the depth they deserve.
Some of the most striking examples of art by an artificial intelligence come from Caedmon. Although it was built, designed and developed by Maurizio Fusillo – a human being – Caedmon operates entirely free from Fusillo’s influence or personal taste. The images it comes out with represent an organic learning curve, as it constantly assesses the features of a large library of source materials. In addition, it can revisit its own previous pieces with the extra perspective of growth. Interestingly, mirroring the phenomenon of peer review, Caedmon also considers the reception its art receives, tailoring its approach accordingly.
For SCP3, our third season of podcasts, we invited Fusillo on to speak about Caedmon, and the philosophies behind his research. Fortunately, like Caedmon itself, Fusillo is incredibly insightful. His thoughts on the evolution of art, and its underlying meaning, are right at home in our Secret Cave. As such, I’m proud to present my interview with him as our fourth episode in this run. Click below to hear the full audio of our dialogue, or keep reading for transcripts and my own further commentary.
All blockquotes in this piece are transcripts of Maurizio Fusillo’s words from our SCP3 episode.
Basically, Caedmon is a step forward in research I’ve been carrying [out] for years. The main idea is that a computer could be able to generate art. This is not a new idea. Actually, since the last century, there was this tendency [towards] making art more impersonal; involving less feelings, almost randomly generated. We can think about Dadaism of course, or even Futurism in some respects.
Technology today is more or less ripe for this. Caedmon is not the only attempt to generate art through a machine. It’s just, maybe, it has a more conceptual structure compared to other projects, which are more technically focused; like, for example, Google’s DeepDream, or many other projects.
Of course, these are just the first steps in that direction. But, why not? I mean, this is the main point of Caedmon. Why not?
As the old cliche maintains, big things can spawn from small beginnings. If “why not?” was truly the first intention behind Caedmon, it’s no longer the only avenue it explores. In fact, it can inspire ripples of questioning and thought in the open-minded. Caedmon is not just a random product of possibility and whim. It strongly reflects the times of its digital birth; a seemingly logical direction for art to take with the advent of computers and their effect on popular culture. In that way, it’s far more than the programmed product of a man’s experimentation. It makes sense in the historical context surrounding it, as all powerful expression does.
First of all, art cannot be created out of nowhere. You still have to refer to what happened in the past. Art is not just the application of creativity for the sake of creativity. At least in my opinion, art still has to follow a history. There is the history of art, and you cannot just do something outside [that] flow.
There are tendencies in the art, and those tendencies are pretty related to the world and time in which these artists created. So, if you create something just for the sake of creating something, even if it’s very beautiful, is it not art?
This is just my approach. I mean, art fits what happened before in the field and, also, what is going to happen. So, it is not just a showcase of techniques.
The use of artificial intelligence is more pronounced than it may seem. Mobile apps can add dynamic drumbeats to amateur recordings. Entertaining bots, though many employ simplistic algorithms, flood Twitter and other social media platforms. Recently, I followed a popular bot whose tweets consist of nothing but fake, though surprisingly convincing, Louis Theroux quotes. Of course, video games are the most obvious stomping grounds for artificial intelligence; from the actions of NPCs to the wider implications of procedural generation.
They are going to touch, basically, every aspect of life. Not only art, like visual art, but also music. I was just reading that there are many, many interesting projects of artificial intelligence applied to music. There are video games which are totally generated by artificial intelligence. It’s a driving field at the moment.
Though it has, indeed, captured the imagination of many developers, musicians and artists, some of the fruits of that labour have been woefully lacking. In particular, I’m reminded of No Man’s Sky. Its reliance on procedurally generated landscapes and tasks proved to be its undoing. Repetitive, dull and downright infuriating, No Man’s Sky was a loud reminder of the limitations that come with the field. However, artificial intelligence is always improving. 18 quintillion planets is clearly an ambitious ask now but, in the future, will that always be the case?
The problem in the past was that, actually, the technology was not ready for this. If we take into account what [has] happened in the past two or three years, in this field, and how fast this field is going, we can expect that, in ten years more, we will deal with content generated totally by a machine. And, it will be extremely natural and, maybe, even better.
If Fusillo’s above assertion is correct, is it not possible that human expression could degrade in response? If the work of artificial intelligence becomes accepted and digested on the largest scale, to what extent could we see our own artistic instincts disappear altogether? It’s a doomsday scenario I can’t help but toy with, but Fusillo has a faith in our drives that I share. After all, this isn’t the first field to be accused of putting an end to creativity.
When photography came out, everyone said, “This is the end of art as we know”. Actually, it was not true. We cannot kill human art. The art is born by the will to express something, and this is [inherent to] human nature. We will never kill the traditional artist.
Though not a traditional artist at its core, Caedmon is still able to walk similar paths to its human counterparts. Like us, it learns. It consumes, from a large set of images, and makes decisions based on a number of factors. Everything it creates is informed by years of precedent, which Caedmon has the ability to approach with its own slant. All of this may, in essence, be fuelled by binary – replacing emotion with ones and zeroes – but it’s intriguing how similar its methods are to our own; at their simplest.
[Caedmon] is trained on what happened in the history of art since, let’s say, Roman times until now. And, not only Europe but in every part of the world. So, he’s learning from that. We cannot create, again, something that is totally new. We have to learn from the past.
He’s just trying to learn what art is for us, and creating something that might be art for us. Caedmon doesn’t actually feel something when he creates the art. Still, the feeling is in the eyes of the person who is seeing [it]
Most of the time, we credit the weight of an artwork to its creator. Audiences are, largely, thought of as purely receptive to the divine impulses of artists. Instead, audience response can be quite independent of an artist’s initial motives. If such motives are taken out of the picture, the burden of feeling lies wholly in the hands of those viewing it. To the cynical, this very act renders something artless. To me, and many others like Fusillo, it’s just a beautiful twist on the relationship between a talent and their audience. Caedmon and its contemporaries retain the fourth wall. They merely look at it from a different angle.
If you think about the Dadaists, [or] the Surrealists, they used to write poems just taking [random words cut from newspapers out of a hat]. Or, if you think about the Futurists… Like, they were using onomatopoeic sounds for their poems. This is interesting. It’s moving from what is in the artist’s mind to what is in the observer’s mind.
Now, we have the technology which allows [a] machine to actually learn, and make decisions based on what it’s learnt. In the same time is the observer, who charges the artwork with some feelings.
So, how does Caedmon work? What are the processes behind its individual images, and how do they incorporate both history and feedback? Again, there are some clear parallels between Caedmon’s approach and that of a human artist. Feeling may be absent, but its output shares many of the same inspirations as our own. While it’s surely highly complex at the deepest level, there’s a conceptual simplicity to Caedmon’s movements you can recognise outside of artificial lands.
Basically, [Caedmon] analyses two of the images in his collection, which may be images that [have already been] generated by him, so he’s reinterpreting his own picture. What he’s doing is, he’s statistically analysing those pictures and, kind of, extracting the DNA, which he can use to rebuild a similar picture.
What happens later is, Caedmon is able to garner feedback based on these. So what happens behind-the-scenes is, we can imagine every picture as a living element which can pass down, to the [next] generation, its own DNA. Then it’s the public, the followers on Twitter or Facebook, who can decide which of those pictures survive.
If a picture is very successful, it means that, behind-the-scenes, [it] can pass down its own DNA faster than the others. And, if a picture is not successful, at a certain point Caedmon will delete it; like, kill it. This is similar to what happens in nature, with evolution.
The power of Caedmon is that we can act both ways. In nature, it is only the parent that passes the feedback to the child. In our case, it is also the child who can share some of its success, let’s say, with the parent. If a parent generated a successful child, we can re-use this picture more and more often, so it will have a better weight in the training set. We can actually extract the features that the public will recognise as art, in order to generate better and better results all the time.
It can be astounding how well Caedmon has applied its training. The art it creates is always evocative; occasionally staggeringly so. Through long-term engagement with its ongoing tweets, or Facebook posts, you can see it experimenting. Then, you can even see it play with the most satisfying results of its experiments. Subtle threads, along with more glaring themes, are ever-present in Caedmon’s persistent expression. In this way, it sails through disparate phases with a distinctive voice; its own unique flavour. Followers of its social media accounts offer suggestion and guidance, leaving the rest in the hands of Caedmon.
The whole audience is able to influence what Caedmon is going to create in the future. You can see an evolution. Like, at a certain point, Caedmon started using pretty vivid colours and an approach more similar to Italian Futurist artworks from the beginning of the last century. People started liking that and, if you now see what’s happening [with] Caedmon, most of the pictures it creates still bring on some of those features, because [that’s] what people liked. So, people can actually influence the way Caedmon moves.
This happens for all artists. We learn from experience. If you do an artwork, a painting, that didn’t sell, [and] people say, “Yeah, whatever. What is this?”, you will try not to repeat [it]. If you do a pretty nice painting, and people come to you and congratulate you, you will understand, “Oh yeah, this is what people like. I might move to this instead”.
By emulating humans in its personal evolution, and consumption of feedback, Fusillo hopes that Caedmon is more than a lifeless archive. Such bots are ubiquitous now and, though frequently impressive and diverting, actually somewhat reductive. To counter this, Fusillo has tried to imbue his expansive understanding of art into Caedmon’s very genes. He wants its own trends and indulgences to emerge, based on a form of education and a respect for context. Aesthetic brilliance, at least for the most part, is secondary to Fusillo’s founding goal. Interestingly, it’s Caedmon’s only concern.
It’s nice to have the killer app who will generate wonderful [paintings], but this was not my intent. There are other libraries which generate extremely nice pieces of art. But, they are just libraries. Caedmon was an attempt to create something that can be seen as alive.
So, to what extent has Caedmon been successful? Like all artists, a final answer to that question is determined by outside response. To me, Caedmon has achieved exactly what Fusillo aimed for. Even more interestingly, I also believe that Caedmon’s own interest in visual splendour alone has found much success. Since Fusillo is as oblivious to its next piece as we all are, he’s in an intriguing position. He’s able to see the artwork subjectively, and form his own opinion on its development. Yet, in the long run, the project is far from a conclusion to the research as a whole.
I’m happy about the Caedmon results, and I’m also a fan of Caedmon, because I have no way to influence what it does. I mean, once I’ve left [him] in the wild, he has a life. I cannot control it. So, I also follow him. It’s surprising for me also, like every morning I open it and I see the conversation that it has with people, or the artwork it creates. It gives me a lot of satisfaction but, still, I cannot say that this is the final goal of my research because, actually, I don’t think there will be a final goal to this.
It’s questionable to what degree Fusillo’s selection of source images is, itself, meddling in all of Caedmon’s resulting affairs. If Fusillo chose nothing but his own favourites, or styles he finds palatable, in the base set of pictures provided, that would heavily tip the balance in favour of his taste. Such would make Caedmon’s whole philosophy moot. Thankfully, this is something that Fusillo prepared for. By enacting no prejudice in his chosen images, and including genres outside of his interest, his artificial intelligence stands on its own. As an aside, that’s probably why Fusillo so naturally refers to Caedmon as “he” or “him”.
I try to give him a training set which is as vast as possible, in order to reduce my influence in terms of what I like, or what I don’t like in art. So, I try to give him the maximum freedom from what I like and [might] want from him.
Beyond the appeal of the pictures Caedmon posts, it’s a fantastic catalyst for thought and conversation around the nature of expression and artificial intelligence. As a fan, I’ve found that it’s had more effect on my perspectives toward art than anything else in the past year. However others may rate it, I think that’s a grand testament to its strength. To complete our conversation, it seemed appropriate to ask Fusillo how he views art. Also, as the defining query of this article and the most prudent point to take away, how does Caedmon fit into the fractal branches of creativity?
Art is something in which we invest a lot of our mind, creative abilities and effort and, in the end, we create something which has absolutely no use; and this is wonderful. This is nice. This is what brought us, in our origins, to get better than monkeys, let’s say; the ability to create something which is totally useless. As soon as we managed to satisfy all of our primary needs, we could actually start generating something which is useless; think about art, think about religion and, in some respects, think about philosophy. Those are the best creations of the human race, and those things have absolutely no utility. So, I don’t say that art is useless in the negative way. For me, actually, it’s a wonderful thing.
Machines, until now, always had a purpose. Like, you start your computer and you have a purpose; I want to check Facebook or create a photo with Photoshop. Caedmon is a machine who [was] started, and does things, just for the sake of [it]. It doesn’t have a purpose. It’s a machine that works for absolutely no purpose. So, this is also pretty amusing. This, maybe, is what’s new with Caedmon. Like, it totally doesn’t have any purpose. What’s nice about Caedmon is that he’s strictly figurative, and puts maximum effort into the aesthetic of the image; more than the concept.