Collusion AI labs: creations and discoveries!


I’ve just had an amazing week in Cambridge as one of eight artist residents selected for the Collusion Lab for Artifical Intelligence. We were based at the Leverhulme Centre for the Future of Intelligence – a tucked-away building in a courtyard near, er… that bit of the river where the Anchor pub and the cows are (as a former Cantabrigian, my mental map is colourful, but patchy). The plan was to immerse ourselves in talks and discussions about AI from a range of experts for three days, before gathering our thoughts, and perhaps emerging with the beginning of a new project.

There was a good range of interests, backgrounds and age between us, but somehow (I think) we all found overlaps with everyone else in the group. It seemed to me, anyway, that everyone got on with everyone else really well.

Collusion AI Labs June 2017

Of course, in three days and for a general audience, it was never going to be highly technical. But it sure was inspiring and fun. We had talks from some of Cambridge’s top A.I. thinkers including AI anthropologist Beth Singler and philosopher Stephen Cave (who I’d interviewed way back, for Hack Circus issue 5, about his book on immortality). We met Murray Shanahan, AI consultant for the movie Ex Machina. We heard from wonderful artists working with new tech, music and video like Vicki Bennett and Ronald Fraser Munroe, and saw how puppetry, data and invisible intelligence could come together, in the gorgeous work of Jo Lawrence. We thought about Amazon Alexa and ARM’s medical potential and data privacy, and training AIs on 80s video games, and the ‘watercooler’ chats in the break room took it all even further.

And there was VR tech on hand! We had a go on a Hololens, and tried 3D drawing with Tilt Brush.

We were also treated to private tours of the excellent Makespace in the same building, and the nearby archaeology & anthropology museum.

There was a real lightness and joy to the week that I think informed the work we ended up doing. The weather helped, with seminars out on the grass and plenty of authentic Cambridge pub experiences. And honestly, I learned as much through chats with the others in the group (who ranged from theatre directors and performers to those working with public space, VR and slime mould) as I did from the talks.

The Mystical AI Tarot 


The format was roughly 3 days learning, 1 day making, 1 day presenting/discussing. By Thursday, on a lot of caffeine and very little sleep, I had loads of questions and thoughts jotted down. Eventually I managed to find a way of integrating some of them into a rough draft of a Mystical AI Tarot.

I was thinking about the absurdity of training computers by using games, the way we experience games very differently to machines –  but the similarities in spatial and categorising work we do with card games and how they generate ‘algorithms’ of sorts, for us. Also things like punchcards, where reason meets hopes/fears, and ghosts in machines.

I wondered what would be the most human game possible, involving the most intuition and the fuzziest rules. What would be the game most difficult for an AI to play? So I came up with the AI tarot idea. A lot of talk about AI is ritualistic and speculative and the union of the two worlds felt right.

I’d had an interesting chat with Collusion’s Simon about readymades, and what makes something art. In my constant desperation to simplify the world, I wondered whether art can be reduced to a few ‘ways’ – a few classes of experience which could indeed be ‘experienced’ by a machine in an identical manner to humans.

Not only that, but does the means of AI learning not echo the artistic experience? Is it even the case that algorithms are already being trained to ‘experience art’ – that it could be, in fact, all they can do?

In my final presentation I called upon Dominic Wilcox’s wonderful art exhibition for dogs as an example of the value of art for non-humans. AI are an ‘other’ – a someone totally unlike us. It seems like a very good moment, right now, to consider valuing and even honouring perspectives we have no way of understanding.

NB, if anyone would actually like an AI Tarot pack, get in touch and I’ll happily do you one for a very small fee!

On our (long!) development day, I also made this training video for humans, from the perspective of an AI (with algorithmic music):

Anyway, here’s where I’ve got to with my thinking about AI and art:

  1. While Collusion supported anything going in any direction, there is still an institutional bias controlling the relationship between the arts and sciences that it’ll take some force to shift. In the final questions just before we all left for our trains, Huw Price said something about arts acting as a communicator for the science discoveries. I was extremely tired and didn’t pipe up, but I’ll say it here: I think that’s about the worst scenario imaginable… worse even than the evil paperclip AI! We’ve forgotten, I think, that the arts are artful artifice. There must be freedom to be an unreliable narrator, to pursue one’s own agenda, to chase an emotion – however scientifically inaccurate or absurd. There must be space for a take on an approach that doesn’t back that approach. There must also the possibility for open interpretation and accidents, and a validity in individual readings. I’m not pushing wilful contrarianism; great art return us to some kind of truth – it always does. But artists need not bend to another’s worldview or agenda. Science represents a kind of truth and a particular journey into the unknown, but there are others.
  2. My fascination with contemporary economics continues apace. We have the lowest unemployment since 1971 but as wages are dropping off sharply at the same time, it’s difficult to read this as a good thing. Basically, something REALLY weird is happening RIGHT NOW with humans and jobs. Various things are holding the economy to ransom: interest rates are low and wages are low and the cost of the things we buy is increasing so the debt bubble expands. People are dispensable and powerless. Why expensively automate your entire workforce when you can get a load of humans in for a pittance on zero hours contracts? I put it to you, gov’nr, that – just as we once feared robots and UFOs so we didn’t have to fear communists – we now fear AI so we don’t have to fear the very real dehumanisation of the workforce right in front of us.
  3. Artists are metaphorical, emotional and social. What artists are not, necessarily, is particularly illustrative (for that stuff, you want a designer.) It’s easier to make an AI who can design, than one who can drift away from the literal, and dredge up a deep, uncomfortable truth. AIs can drift away and dredge up nonsense we may or may not find meaningful, but they have no intent and we may as well be looking for faces in clouds. It seems to me that’s where we are right now.
  4. Part of the problem is surely that AIs don’t know what touches individual humans – but it strikes me that programs may be qualified to affect other programs with whom they share values. In this case, perhaps AIs can design art for one another? Art by machines, which touches, persuades and somehow changes other machines, and which we mere humans have no understanding of at all. Art which reminds us of the most important thing – that we are not the only ones making and receiving ideas, there are others working in ways we will never understand. Art that we don’t understand, that makes us smaller and more humble. A double blind art experiment, by AIs, about human tolerance of difference.
  5. For now, philosophically, AIs may as well be aliens, but at least they’re aliens we can program.




Leave a Reply

Your email address will not be published. Required fields are marked *