PDA

View Full Version : Automated Game Design Research - I Have A Game For You To Play!



mtrc
28-11-2011, 05:11 PM
Hello all! Apologies in advance, this is a thread where I talk about something I had a hand in. But in my defence, I think you'll be interested! Bear with me.

I work at the Computational Creativity Group at Imperial College. We research AI techniques and build AI systems that try to recreate, investigate or simulate 'creativity' as we understand it in humans. That might be looking at methods of reasoning or concept formation (my PhD supervisor built a mathematical discovery system that identified, proved and subsequently published novel mathematical concepts!); it might be about building an AI artist that has opinion, imagination and emotion; or it might be the thing I've been working on - can an AI design an entire videogame?

We already see a lot of Procedural Content Generation in our videogames. Either 'offline' PCG (when game worlds are procedurally generated by developers and locked-in as game content, like EVE Online's galaxy of stars) or 'online' PCG (like Minecraft's world generator, that makes stuff for you on-the-fly). But PCG is generally used to create content that is a small cog in a larger human-designed game.

What would it take for an AI to take on a larger role than just generating content? Could we build an AI to design games?

I don't have an answer for that yet (but I'm happy to talk about it here if anyone's interested). What I DO have for you is a game!

Click here to play it. (http://www.gamesbyangelina.org/games/santas_snowfight_escape/)

It's a very short, simple platform game inspired by the Metroidvania genre. We're in the earliest of early days in this area of research. But it's of utmost importance that, at every stage, we come back to gamers for feedback and responses. So I'd be really grateful if you could take the time to play the game (and maybe even fill in the survey too, if you have time. only 1 compulsory question!)

Thanks, and I'd love to chat about this if anyone's interested!

Mike

CMaster
28-11-2011, 05:23 PM
I'm intrigued, what did ANGELINA actually do in terms of creating that game?

Wizardry
28-11-2011, 05:38 PM
There's a lot of game developers on this site. I don't think any one of them will want to help you put them out of a job.

mtrc
28-11-2011, 05:46 PM
Hey CMaster! Thanks for taking a look.

ANGELINA performed three tasks in designing this game:

1. Designing the level. I gave ANGELINA 40-50 map tiles that are about the same size as the screen. ANGELINA chose tiles and laid them out in the order you see here (the space is much larger than 40-50 tiles, because ANGELINA can combine two tiles together to make a composite tile.

2. Design and layout of enemies, the player and the exit. ANGELINA places enemies in their starting location, and decides what abilities they should have (static, firing, pouncing at the player, flying, that kind of thing).

3. Design and placement of powerups. I gave ANGELINA control over several ingame variables, such as jump height and 'locks' (there are ice blocks you can't get through initially). ANGELINA placed 2 powerups in the level, decided on their location and what they should affect (e.g. one powerup affects the Jump Height, and it sets it to something like 250 pixels. Both of those decisions were made by ANGELINA).

So it's working within a well-defined space, but I did not give it any specific instructions. I could randomly select map tiles, enemy designs and powerup components and you would get an entirely unplayable game.

The way it does this is another story (which might be boring and you can totally skip!). It involves computational evolution.

When I say ANGELINA 'chose' a map tile or an enemy placement, what I mean is the system selected and combined potential designs over and over again, in what we call 'computational evolution' because of the relation to evolution in a biological sense. Taking map design as an example:

1. Make 1000 random maps. Just shove together whatever tiles you want, don't even look at them.
2. Now look at the maps, and rate them all. Give them a score, out of 100 or 1000 or whatever.
3. Choose the best 1% or 10% or 100 maps. These are your most fit maps, and they're the ones that get to 'breed' (this is where the evolution analogy works, sort of).
4. 'Breed' the maps together by combining their attributes to make new, child maps. Keep doing this until you get back up to 1000 child maps.
5. Repeat 2-4 over and over again. ;)

That's a very, very brief overview of computational evolution. It's used in lots of games. Where we differ is we have multiple evolutionary processes, and we use a method that lets these processes share information. What that means is that if a level design is beginning to make a long corridor with a locked door at the end, the layout design might respond to this by placing the exit behind the door. That's the sort of effect we're getting, and it's what sets ANGELINA apart from straightforward procedural generation.

Hope that wasn't too verbose... Sorry!

Vexing Vision
28-11-2011, 05:47 PM
Please do keep talking about this!

I am very interested in the concept.

mtrc
28-11-2011, 05:49 PM
There's a lot of game developers on this site. I don't think any one of them will want to help you put them out of a job.

Thanks for bringing up an important point.

This won't be putting anyone out of a job any time soon.

We do this research for a few reasons. Firstly, it helps us understand human creativity better, but looking closely at processes like design and refinement.

Secondly, this research is designed to make design easier! ANGELINA can work with humans to design a game co-operatively. It would mean that, for instance, a really great level designer could collaborate with ANGELINA, and have the system fill in some game mechanics while the designer works on creating some kickass levels.

Thirdly, what I've found in the year I've spent working on AGD is that it makes me appreciate human-designed games more, not less. I think having systems that can develop games (with or without human help) will only reveal to us the inherent worth in games as artistic pieces.

It's a long debate, and one I'm happy to have if anyone is really concerned that I am trying to get them fired. ;)

TailSwallower
29-11-2011, 01:42 AM
Hello all! Apologies in advance, this is a thread where I talk about something I had a hand in. But in my defence, I think you'll be interested! Bear with me.

No apologies needed. You're upfront about your involvement, you're not shilling, and this project sounds really interesting.
I've got a story on the backburner that will somehow involve AI, but I don't want it to be a re-hash of what Gibson did in Neuromancer, so I'm really interested in research that's happening in the here-and-now about Artificial Intelligences and the like.

Can't check out the game just yet, but I will later.

Kodeen
29-11-2011, 02:48 AM
This is indeed interesting. I didn't fill out the survey after playing the game because, well, I have surveys (sorry). I'll give a couple notes, however. First, I saw a snowman stuck in wall tile in the upper lefthand corner of the map, I think the corner had a 3x3 grid of wall tiles and he was in the middle one. Should be simple to fix, maybe define wall tiles as exclusive residents of each block or some such thing. Second (and I'm aware that this isn't really the point of the exercise), but no damage from the enemies gave me no incentive to be careful, so I took hits liberally.

One question, is your AI restricted to level generation? I was thinking of something like a TES game, where you can procedurally generate the overworld map, place enemies randomly/procedurally on that map, but as to what you can do at that point, I don't know what role AI could play. Gameplay mechanics are quite a bit more complicated than item placement, and of course every time you create new mechanics, you'd have to re-educate the player as to what they're supposed to do.

Zetetic
29-11-2011, 03:03 AM
3. Choose the best 1% or 10% or 100 maps. These are your most fit maps, and they're the ones that get to 'breed' (this is where the evolution analogy works, sort of).
I appreciate that this might be precisely the kind of thing that you might not want to talk about at this point, but how are you defining and measuring 'best' here?
Are you just working from the scales on the feedback form - and broadly defining 'best' as whatever responders say is 'best'?
Do you do incorporate playback times/time-to-giving-up, or other measurable aspects of the game playing experience (say, waiting around time or to take a more complex example, repetitive behaviour on the part of the player)?

Do you currently have ideas about selecting designs towards ends other than 'best', or even player enjoyment? I could see there being applications here in being able to generate levels with particular response-characteristics in players, but this brings us back to response measures and the difficulties there.

On those lines, do you have alternative training regimes other than humans playing the game (and providing feedback either through report or whatever)? As you describe it, angelina was obviously given fairly significant constrains (screen-sized map tiles for one, although I presume power-up attributes were given fairly conservative starting values and mutation rates and/or initial randomisation).

What is your working definition of 'creativity' in this group? On the CCG website you (sensibly) enough advert to, if I'm understanding this correctly, human judgement of behaviour (perhaps products?) as 'creative', but I'm wondering if you have something more explicit in mind?

Simon Colton's grand challenge piece is interesting in itself, but again doesn't really address the definitional/construct issues that seem to dog 'creativity' in psychology of individual differences and like, let alone in computer science or computational neuroscience (which Colton nods towards).

mtrc
29-11-2011, 04:14 AM
Hey all,

Just wanted to let you know that I will reply to you all in the morning (it's 3am, just finished To The Moon...) but thanks for reading up and responding with interesting stuff! Will get back to you all ASAP.

mtrc
29-11-2011, 01:24 PM
Okay, that reply I promised. Man, To The Moon was good.


I've got a story on the backburner that will somehow involve AI, but I don't want it to be a re-hash of what Gibson did in Neuromancer, so I'm really interested in research that's happening in the here-and-now about Artificial Intelligences and the like.

Can't check out the game just yet, but I will later.

I'm happy to answer any questions if you want (mike @ gamesbyangelina.org). Also check out our group's page: http://ccg.doc.ic.ac.uk and the latest in computational creativity work: http://iccc11.cua.uam.mx/


One question, is your AI restricted to level generation? I was thinking of something like a TES game, where you can procedurally generate the overworld map, place enemies randomly/procedurally on that map, but as to what you can do at that point, I don't know what role AI could play. Gameplay mechanics are quite a bit more complicated than item placement, and of course every time you create new mechanics, you'd have to re-educate the player as to what they're supposed to do.

Let me tell you, I would love to work on something like that. I was playing Skyrim a lot last week, and thinking about this project at the same time.

Part of it comes down to - "Okay, I want to do this, but will it result in something new? Some new research?" One of the sad things about working in academia is that I'm not here to create great games, I'm here to develop new techniques that make great games. Someone else gets to go and run to the finish line with that particular baton.

That said, there is a lot of innovation in a project like TES. A lot of new things to consider. I'm definitely considering it as a potential target in the future. To answer whether or not the system could work on it - currently, not really. They're pretty specialised systems, and this one does platformers. But as we go on, my intention is to make the frameworks more general, so that building the next designer is easier and easier, and less specialised as we go.


I appreciate that this might be precisely the kind of thing that you might not want to talk about at this point, but how are you defining and measuring 'best' here?
Are you just working from the scales on the feedback form - and broadly defining 'best' as whatever responders say is 'best'?

To tackle this one first - a major research question, one that would produce work for many PhD students! - is "what makes a game good?" Can you put a number on entertainment? Our gut response is 'no, of course not', but I think for specific types of game, or specific types of entertainment, we can find heuristics and indications of what people prefer over other things.

For ANGELINA's sake, we look at pretty simple stuff. Can you win? Is it guaranteed that you'll win? Do you get to explore the map a lot? The most basic ludic concepts you can imagine.


Do you do incorporate playback times/time-to-giving-up, or other measurable aspects of the game playing experience (say, waiting around time or to take a more complex example, repetitive behaviour on the part of the player)? Do you currently have ideas about selecting designs towards ends other than 'best', or even player enjoyment? I could see there being applications here in being able to generate levels with particular response-characteristics in players, but this brings us back to response measures and the difficulties there.

We actually don't tailor these games to an individual player, but there is a whole other field of research broadly called Adaptive Games that intends to do this. You might be interested in Jeremy Gow and Robin Baumgarten from my group. They're looking at ways of segmenting and grouping players by taste, and reconfiguring games in real-time to react to which group we think you're in. So a PacMan game starts adding ghosts if it thinks you're getting bored, or makes a level with more pills in if you seem to like using them.


On those lines, do you have alternative training regimes other than humans playing the game (and providing feedback either through report or whatever)? As you describe it, angelina was obviously given fairly significant constrains (screen-sized map tiles for one, although I presume power-up attributes were given fairly conservative starting values and mutation rates and/or initial randomisation).

When designing the game, offline, ANGELINA only has input from herself. The system simulates playing the games, records certain metrics and uses them to evaluate the game afterwards. It's almost like a mini games industry - ANGELINA designs games, plays them, reviews them and then designs new games based on the feedback from the reviews! So it doesn't train so much as work towards notions of 'goodness' which I supply to it. These notions are kept deliberately general and vague, so as not to confer too many restrictions on the system. For instance, I say that games where the player has to explore more of the map are better than games where the player explores less of the map. General ideas, rather than "Put the exit in the top left-hand corner".


What is your working definition of 'creativity' in this group? On the CCG website you (sensibly) enough advert to, if I'm understanding this correctly, human judgement of behaviour (perhaps products?) as 'creative', but I'm wondering if you have something more explicit in mind?

So there's two answers to this, one is the group's message, which we discuss in philosophical/position papers. This is becoming a bigger thing lately, we just added some concept formation researchers to our group and that's all getting very exciting. Check out this paper for an idea:

http://www.doc.ic.ac.uk/~sgc/papers/pease_aisb11.pdf

Or this one:

http://www.doc.ic.ac.uk/~sgc/papers/colton_aaai08symp.pdf

It's about really defining what it means when we say a human was being creative. And not everyone will agree with these definitions, but then that is what the field is there for, for discussion and debate. What do we mean when we say that Da Vinci was creative? And am I being creative now when I write this paragraph? What's the connection between me and Da Vinci, in that case?

Personally, I hold slightly harsher views. I see the brain as an essentially deterministic machine, albeit one of incredible complexity and value. Clearly, this doesn't affect my work, because in terms of interaction with the 'philosophy of creativity' stuff we do I'm really more at the implementational end of things rather than what creativity is. But yeah, I think when we discuss the topic at the pub I tend to be on the side of 'we're basically just very big computers'. Creativity is our ability to generate unseen content by recombining patterns we've learnt and understood (to some degree). I'm no neuroscientist though. :P


Simon Colton's grand challenge piece is interesting in itself, but again doesn't really address the definitional/construct issues that seem to dog 'creativity' in psychology of individual differences and like, let alone in computer science or computational neuroscience (which Colton nods towards).

I'm not 100% sure what you mean here, but looking at the grand challenge piece I think our mission statement has probably moved on, at least somewhat, since he wrote that. Check out some of those papers for an idea of where our definitions are right now.

Thanks so much for the interesting responses, everyone!

Zetetic
29-11-2011, 04:54 PM
To tackle this one first - a major research question, one that would produce work for many PhD students! - is "what makes a game good?" Can you put a number on entertainment? Our gut response is 'no, of course not', but I think for specific types of game, or specific types of entertainment, we can find heuristics and indications of what people prefer over other things.
I was more interested with what actual measures you were using, rather than worrying about the idea of measurability of 'good'.


For ANGELINA's sake, we look at pretty simple stuff. Can you win? Is it guaranteed that you'll win? Do you get to explore the map a lot? The most basic ludic concepts you can imagine.I assumed that you'd have some basic sanity checks as regards the most obvious aspects of playability, but stuff like the exploration-aim is interesting as something you've chosen.



We actually don't tailor these games to an individual player, but there is a whole other field of research broadly called Adaptive Games that intends to do this.Again, I was actually mainly wondering what feedback you supply ANGELINA with as a whole, rather than individual tailoring.


You might be interested in Jeremy Gow and Robin Baumgarten from my group...Thanks for mentioning this.


When designing the game, offline, ANGELINA only has input from herself. The system simulates playing the games, records certain metrics and uses them to evaluate the game afterwards. It's almost like a mini games industry - ANGELINA designs games, plays them, reviews them and then designs new games based on the feedback from the reviews!Ah, so you've had to implement some kind of playing AI (in the weakest sense?) as well - how extensive is this? Does the playing simulation actually involve a system that can only interact with the level as a player does or do you just inspect the generated game/level? (Apologies if this is dreadfully dull.)


These notions are kept deliberately general and vague, so as not to confer too many restrictions on the system. For instance, I say that games where the player has to explore more of the map are better than games where the player explores less of the map.Sounds sensible, and provides an interesting route for your intention as author to be manifested.


So there's two answers to this, one is the group's message, which we discuss in philosophical/position papers. This is becoming a bigger thing lately, we just added some concept formation researchers to our group and that's all getting very exciting.Thanks for linking to those papers.


It's about really defining what it means when we say a human was being creative.Quite. With that in mind, I thought that the Pease & Colton paper, in particular, that you posted was a rather curious affair since it suggested models of description of 'creative acts', with remarkably little reference to how humans actually talk or reason about products of creativity. I found it odd that in a proposal of this sort, however tailed towards a computational-focused publication, there was almost no discussion of whether the proposed descriptive models bore any real resemblance to how humans think about and talk about 'creative' acts.

What is then interesting, is that paper talks about rejecting a conception of creativity that only captures a 'single (human) style' and having to 'manage' the perception of creativity of computers. Which taken a little badly seems to an attempt to redefine creativity for CCT's own purposes.

Indeed, in Colton, Charnely & Pease (2011) they make it clear that the CCT models are not intended to be of any use for assessing the creativity of human behaviour - which means I find it very difficult to understand what the actual intended connection between CCT and 'creativity', as we use it relatively mundanely or as the psychologist developing an assessment instrument uses it.

I realise that I'm attacking other people's work, largely from a position of ignorance (to be open about it - I've got bits and pieces of fairly weak psychology, computational psychology and philosophy of mind which are vaguely relevant but that's it), and that as you say you're an implementer above all else but I really would appreciate any thoughts you have on this and any guidance as to the mistakes I'm making. To hear from someone who's doing this is a great opportunity.

More widely, I imagine that this is obviously something of a difficulty for the field. A significant number of people take against the idea that deterministic systems can be 'creative', and I might even suggest that for many people, knowledge of the exact process of something production rather robs it of 'creativity' - I wonder if you, having written it, would say without any hesitation that ANGELINA's behaviour is 'creative' as yours was producing it?

(Perhaps the way that computational creativity will be 'managed' in the public perception is the ever increasing acceptance of the brain as a deterministic instrument - as you and I believe - and the greater understanding of the mechanisms and process that are involved in human creative acts.)



Personally, I hold slightly harsher views. I see the brain as an essentially deterministic machine, albeit one of incredible complexity and value. Clearly, this doesn't affect my work, because in terms of interaction with the 'philosophy of creativity' stuff we do I'm really more at the implementational end of things rather than what creativity is.That's an interesting point of view, but clearly you are building ANGELINA to do something, and in the belief or hope that the something eventually looks like a 'creative' act to yourself and others. I don't think you get out that easily!


Creativity is our ability to generate unseen content by recombining patterns we've learnt and understood (to some degree).I think that definition quite clearly informs and is informed by your work with ANGELINA, for example. I'm a bit hazy on what 'unseen' means here (but I'm going with 'novel' to the individual) but to my mind it's a lot more understandable than the CCT models as a starting point for the discussion of 'creativity'.

(Do you find yourself, willingly, referring the CCT models in assessing ANGELINA?)


I'm no neuroscientist though.
I think that this almost betrays a way of thinking about the issue. I don't think that a neuroscientist can tell us what 'creativity' is - they, and the cognitive psychologist, may talk to us the process that manifests creativity in the human - but ultimately it's down to the philosopher (and a rather different psychologist) to tell us what we are saying when we say that an act or a product is 'creative'.

Edit: I'm sorry for my utterly rambling reply. I appreciate any time you spare answering my questions.

mtrc
29-11-2011, 11:27 PM
I was more interested with what actual measures you were using, rather than worrying about the idea of measurability of 'good'.

I assumed that you'd have some basic sanity checks as regards the most obvious aspects of playability, but stuff like the exploration-aim is interesting as something you've chosen.

The obvious stuff is sometimes explicity designed against, and at other times I make it possible (ANGELINA can quite cheerily place enemies inside walls) but steer the system away from this in a looser sense, by providing incentives within the evolutionary system to not create systems with these sorts of traits.


Ah, so you've had to implement some kind of playing AI (in the weakest sense?) as well - how extensive is this? Does the playing simulation actually involve a system that can only interact with the level as a player does or do you just inspect the generated game/level? (Apologies if this is dreadfully dull.)

It varies (and it's not dull! I think. The thread's gotten awfully quiet... <_<) from project to project. In the first project, before the Metroid stuff, ANGELINA quite literally played these games out as if she were an AI controlling the player character. These platformers were harder, partly because of a distance between the evolutionary design side (in Java) and the game side (in ActionScript/Flixel). It was a weakness of this particular system - in general, I want ANGELINA to be playing the very games it will be releasing to people.


Quite. With that in mind, I thought that the Pease & Colton paper, in particular, that you posted was a rather curious affair since it suggested models of description of 'creative acts', with remarkably little reference to how humans actually talk or reason about products of creativity. I found it odd that in a proposal of this sort, however tailed towards a computational-focused publication, there was almost no discussion of whether the proposed descriptive models bore any real resemblance to how humans think about and talk about 'creative' acts.

Well, the Pease/Colton paper is mostly in response to the notion of a Turing test for computational works (and bear in mind I'm speaking for them here so this may not be 100% accurate). Often, when we discuss how to evaluate our 'creative' systems, some people propose a Turing test. Get people to play 10 human-designed games, for instance, and 10 ANGELINA-designed ones. Don't tell them which is which. Colton is seriously against this for a wide variety of reasons. I think this might also pay into what you mention about there being a lack of human-oriented creative vocabulary here - because that may not be appropriate for computational creativity.

I feel like that probably did not answer that question. End of a long day. Paper deadline in a week. ;)


What is then interesting, is that paper talks about rejecting a conception of creativity that only captures a 'single (human) style' and having to 'manage' the perception of creativity of computers. Which taken a little badly seems to an attempt to redefine creativity for CCT's own purposes.

Indeed, in Colton, Charnely & Pease (2011) they make it clear that the CCT models are not intended to be of any use for assessing the creativity of human behaviour - which means I find it very difficult to understand what the actual intended connection between CCT and 'creativity', as we use it relatively mundanely or as the psychologist developing an assessment instrument uses it.

The two uses you mention at the end there are quite different. With the latter? There is a huge gap, I imagine (although again, I'm not sure if that's right to say, but I imagine there is quite a gap). In terms of a gap between the everyday usage of the word and the models being used, I think this in part is where a lot of the key argument lies.

The field in general has had a lot of problems, because the word 'creativity' is not the property of the scientific world. Really, it belongs to the world of the arts, and the world of the everyday consumer of news. Creativity means something, and it's also one of those golden gooses that people don't like to see being fiddled with.

This has sort of led to what I've perceived as a beaten retreat by the field in general, conceding ground where appropriate in arguments, and admitting that, yes, right now their field cannot do much in the way of what people call creativity (whatever that might be). Their response was to fall back and begin defining, from the ground up, what they are aiming to do. Right now, those aims may not align much with our understanding of creativity. But what this allows them to do is transfer the argument away from "I don't think this program is creative because of these specific reasons" to an argument more about the framework they are building towards. So yes, the framework may not seem like creativity. But it serves as a frame around which they can build and defend their systems. They are working towards this notion of creativity, and over time that notion may expand further, but right now this is where they are and if people want to criticise then they have to deal with this model of creativity, rather than picking apart projects based on personal distaste.

I think.

Man, this could probably get me fired or something if I say the wrong thing. >_>


More widely, I imagine that this is obviously something of a difficulty for the field. A significant number of people take against the idea that deterministic systems can be 'creative', and I might even suggest that for many people, knowledge of the exact process of something production rather robs it of 'creativity' - I wonder if you, having written it, would say without any hesitation that ANGELINA's behaviour is 'creative' as yours was producing it?

ANGELINA is, without any hesitation, not creative right now. Or rather, she is in a very naieve way, in the sort of way a small child might be. Margaret Boden terms this 'compositional' or 'combinatorial' creativity. Even here, ANGELINA is lacking a lot of things. ANGELINA isn't acting with intent here, it's smooshing together designs and seeing which ones are best. There is no grand design or goal. The system is on the bottom rung of the ladder, but this is often a good starting point for loftier things. :)


(Perhaps the way that computational creativity will be 'managed' in the public perception is the ever increasing acceptance of the brain as a deterministic instrument - as you and I believe - and the greater understanding of the mechanisms and process that are involved in human creative acts.)

I honestly don't know. I think there is a somewhat religious connection to creativity, just as Aristotle believed reason was what made us uniquely human. Creativity appears to be something only humans are capable of (although there seems to be evidence to doubt this even in the natural world). And that sense of being special is something that some people will defend to the death. Engaging with the art world, as my supervisor does, is a minefield of opinion and criticism. Some good, some bad. :)


That's an interesting point of view, but clearly you are building ANGELINA to do something, and in the belief or hope that the something eventually looks like a 'creative' act to yourself and others. I don't think you get out that easily!

You're right, it's a total cop-out. :P I believe that systems like ANGELINA will never achieve human-like levels of creativity, because the underlying framework isn't there. I'm not representing abstract concepts with ANGELINA, I'm telling it to make 2D arrays that represent levels. You'd need a fundamental shift to get close to real, real creativity. However, I believe that ANGELINA could do a lot of things that are creative. I believe that a system like this could be given a sufficiently abstract understanding of game components, and use this understanding to generate novel game mechanics. It's an area of expansion I'd like to consider.

So I guess the answer is another cop-out - my job necessitates that I make careful, well-judged leaps in research and that means working on small systems that might never have a chance of being creative in their current state. But at the same time, my mind can imagine (and hopes for) a system that would truly handle concepts abstractly, be able to combine and recombine them and generate new concepts as a result. Here's an Alpha Centauri quote I love:

"There are two kinds of scientific progress: the methodical experimentation and categorization which gradually extend the boundaries of knowledge, and the revolutionary leap of genius which redefines and transcends those boundaries. Acknowledging our debt to the former, we yearn nonetheless for the latter."

The former is the one that keeps me sane. The latter is the one I shoot for in between papers when no-one's looking. :)


I think that definition quite clearly informs and is informed by your work with ANGELINA, for example. I'm a bit hazy on what 'unseen' means here (but I'm going with 'novel' to the individual) but to my mind it's a lot more understandable than the CCT models as a starting point for the discussion of 'creativity'.

It is, but then again that's easy for me to do, because all I have to do is churn out the odd videogame. ;) My definition can afford to be a bit looser and generic, because it's unlikely that I'll have to build it explicitly into ANGELINA any time soon. By contrast, we've got Pease joining our group, and a PhD student attached, with the exact intention of investigating mathematicians, and how they collaborate and invent/discover new concepts. What is it about the brain that forms concepts, and what would it mean to automate this?

For them, they can't put down a sentence about 'unseen concepts' without being able to back it up later along the line. So I understand that they're trying to be more precise and exact in what they promise. :P


I think that this almost betrays a way of thinking about the issue. I don't think that a neuroscientist can tell us what 'creativity' is - they, and the cognitive psychologist, may talk to us the process that manifests creativity in the human - but ultimately it's down to the philosopher (and a rather different psychologist) to tell us what we are saying when we say that an act or a product is 'creative'.

Yeah, I imagine my supervisor would call me out on that sweeping comment too. It's kind of an offhand way of saying that I'm new here, which I totally am. It's an interesting field. It's only a decade, maybe two, old, and it's full of a broad range of projects. It was initially almost exclusively practical - no traces of this 'creativity theory' stuff. People would sometimes talk about their ideas in the context of a personal, applied project.

Only in the last year or two has the community come together at ICCC (which is the only conference on the topic, and still only has a handful of research groups linked to it) and said "Okay, enough is enough, we need a shared theoretical base to stand on together." So it's early days for things like the CCT models and so on.

Thanks a lot for being so genuinely interested. It's great to be able to talk about this stuff and hear what people think, as well as flex my own ability to explain it. Practice makes perfect!