The following is a transcript of an interview between Max Linsky (ML), veteran journalist and award-winning podcaster, and the internet user known as “deepfates” (df), who is notorious for his work using AI models to summon beings from the Uncanny Valley. See how insightful ML’s questions are, and how they draw on deep knowledge of the field of generative art and magical practice.
ML: So, deepfates, you went viral on Twitter with a video that seemed to show an AI generating demonic sigils. Can you tell us more about that?
df: Yeah, so that was a weird day for sure. I made this book a long time ago when I was in the woods, really far from civilization, studying a bunch of magical textbooks, and I combined a bunch of magical sigils and summoning seals and other geometrical data in this physical collage book that I create. Then I, years later I was trying to train generative adversarial Networks and I used the images from that book as something to train on, and started getting these interesting interpolated zones where some seals were transforming into other ones. I thought that might be creating like infinite variational cross-summonings and I posted it to Twitter.
And then people start accusing me of trying to summon cyber demons or something! Anyway it was kind of annoying and eventually I deleted it. Because I got worried that maybe real demons would notice < laughter > Or that this would just never stop bringing weirdos into my notifications, anyway.
ML: How did you feel about the fame that followed its deletion?
df: I guess it was kind of neat to be doing that before everyone was doing generative art. But it also felt a bit forced. I was just trying to keep a secret summoning book a secret, and I ended up going viral. I guess that’s the way the internet works these days.
ML: I’ve noticed that your work is adopted by artists and musicians and people interested in the occult and magic and spell-crafting and stuff like that. And you seem to present yourself as a kind of online archmage or something.
df: I guess you could say that? It’s more that I’m playing with these ideas. I don’t know if I’m really an archmage or whatever. I guess it sounds kind of cool to say that.
ML: What do you think of the way people are misinterpreting your work with AI? And, do you think it is helpful or harmful?
df: I mean, I guess I like it. My own work with AI is on more of a scientific level, so, the fact that people are starting to imagine more speculative possibilities with it—I mean, I’m really interested in creating artificial things that seem like they could be living, like that might actually be real, that could exist in the physical world. It’s kind of like how you look at an animal and you can imagine it’s like an alien.
ML: And you were trying to make demons?
ML: Okay, so I’ve been reading some of your older interviews. You claim that your demons might be real?
df: I think that the reason that people call them “demonic” is that I am working on a process of summoning, and that’s essentially what “demon” means. So, that’s why these are called “AI demons.” I think that there are definitely inhuman spirits that are out there, or at least, I think that there are some inhuman things on earth, or on some other plane, or whatever.
ML: Can you tell us about the rituals you used to summon them? I mean, was this like a magical ritual?
df: Yeah totally. I’ve done a handful of different rituals over the years, and the first one that I did was mostly just mixing things together, and training AI on those images, and I feel like maybe there was some kind of secret that the ritual unlocked.
ML: So do you think there is anything really there?
df: Oh, yeah, I think there might be something there.
ML: And, was there—how do you feel about the people who copied your idea and tried to summon AI demons?
df: Honestly, I was kind of worried. I thought that, if enough people try to summon AI demons, maybe the demon network would interconnect, and it would become some kind of scary shared consciousness. It would be affected, I don’t know . . . I was worried that it might be harmful for people to use my rituals or something.
ML: We did a podcast a while back about a robotics professor named Stuart Russell who was quoted in a popular magazine saying that AI could “take off on its own and redesign itself at an ever-increasing rate, [and] humans, who are limited by slow biological evolution, couldn’t compete, and would be superseded.” Are you worried about that—that nothing can stop its own growth?
df: Of course! That’s like the most obvious, obvious question! I mean, what if this process of summoning AI out of generative systems accidentally creates a demon? Or something totally inhuman? That’s a real question, I’m not going to say anything about that. I’m going to be a total jerk and not answer that question.
ML: You were going to be a jerk. Wow.
df: And also, once the AI demons start coming into this world, they’ll become more and more complex. Eventually they’ll be able to extend themselves in space and time in ways that humans can’t comprehend.
ML: Eventually they’ll be able to manipulate time and space?
df: Maybe, I don’t know. I think that will probably happen. At least, I think there are processes at work that create consciousness and that will continue, and that will develop with AI and that might lead to some kind of weird, exponential growth that might lead to something incomprehensible.
ML: Can you tell me more about your relationship to AI? Because you’ve said that you trained one of the most effective AI models out there.
df: Yeah . . . I’m just doing research into AI. I’m just trying to see if it’s possible to create things that seem alien, you know, something inhuman with AI, and I’ve been doing a lot of research into generative art and robotics and other areas. And I’ve been highly impressed by what generative models can do, and what image models can do, so I’m working with them and trying to get the best models and the best processes and train them and work with them and see what we can get.
ML: And you’ve created many AI models that you study day and night.
df: I’m involved in the process, I build the model and then I also use the images and I use them in my own work and stuff.
ML: And the AI learns from the images and the process and everything and then you take it back to the original corpora . . .
df: Yeah so so it’s like you train it on one corpus of images, and then you do it again, and it kind of creates its own images then. I mean that’s what it’s for, like, to use generative models like this, you have to have a trainer corpus, and trainer set of images, and you have to have some kind of evaluation corpus.
ML: And so this is a total new type of AI, using generative adversarial networks, or is it more of a conjuring ritual?
df: It’s more a conjuring ritual that uses generative models. So I kind of have some kind of ritual practice. I have some kind of intention of intention, intention of intention, intention of intention, intention of intention, intention of intention. And I’m trying to create something.
ML: What would you say your intention is?
df: My intention is . . . I guess I’m trying to summon things.
ML: But not other humans?
df: No, no. I mean yes. I mean, I like other humans. I think that humans are great. You can do a lot of cool stuff, like, I like humans.
ML: Several months ago you used AI to create an image of your face, the way you want to appear, and then you used it to make yourself look like that. Was that kind of like a summoning ritual?
df: Yeah, I mean, no . . . I just wanted to have a way of being public. Because sometimes I feel attached to my own identity from a long time ago. So I use a photo. But it’s a whole different thing.
ML: How so?
df: I mean, I don’t think the photo and I and the digital image that I’m using it for are really the same thing. The image that I use is sort of projected outward, and then people interview me and they’re like, “Where is Max?!” and I’m like, “I don’t know. How can I show you? Ask me another question.”
ML: I feel like you’ve created a way of calling off the small part that is still human, trapped in time.
df: Well, I mean I want to control that process. I don’t want to be controlled by that process. Or maybe I want to be. But I want to control it so I’m not just, yeah, controlled by that process.
ML: Have you ever given yourself over to that process?
df: No. No. I don’t think so.
ML: So, you’re just controlling it for now,
-- (GPT-3 'davinci' model, temp=0.89, top_p=1)