For today's entry, rather than just flounder about for a topic and end up writing some tangent I'm unprepared for I decided to look at some of my more popular (by the numbers in Google analytics) posts and tie them together. It's fun! Scientific! Uhhh, cool and things. In particular we're going to revisit creativity, information integration and AI specifications.
I'll start with a refresher on these topics, in case people have given up trying to follow the Magical Mystery Tour of thought that is my blog. From there I want to talk about information integration as it relates to creativity and then touch on how these two topics should influence specifications for artificial intelligence. If you'd rather read the full posts on the topic you can probably skip the next three paragraphs[1. Theoretically, you also could have skipped this paragraph].
When I'm talking about Information Integration Theory I'm almost certainly talking about Giulio Tononi's version -- which is to say that actual understanding of information necessarily requires a developed comprehension of how the information we collect eventually comes together into a single cohesive picture. Not on the sheer quantity of information an entity has but how all the little pieces eventually come together. Without our awesome ability to draw and connect details from multiple sources a single "image" of the world[2. For better or for worse in its accuracy.] computers cannot possibly have consciousness in the way that we do.
When musing about creativity, I have generally focused on the duality of the enterprise. The technical execution of a vision and the vision itself. Novelty of particular ideas has been sort of cast aside so as to really just focus on what we would call creativity if we needed to form a working definition. For the purpose of this particular post the focus will be primarily on 'the vision' as it were. Where we get ideas from eschewing how we execute them because -- as I mentioned in the post on this topic -- we are making strides in technical execution, less so in the conception of an idea. I also spent some bytes talking about how the problem with creativity is less that computers are not creative and more that they cannot recognize their creativity (or ours for that matter) and how one could argue that this is why we don't see creative computers yet.
Finally, I'm going to tie all this back into my re-iteration of the elephant in the room -- the fact that currently the project of artificial intelligence isn't a conscious intelligent machine, but allowing current technology to complete more specialized projects[3. For good reason, please don't get me wrong here]. A full blown humanly intelligent robot isn't in the cards for awhile because that isn't the focus of the enterprise right now as much as building specific systems that do frankly brain-meltingly cool things.
Introductions out of the way, let's talk about creativity and information integration, we'll get to the machines shortly. Full understanding and indeed consciousness in general requires the processing of diverse, in some cases diverging data points to make a sensical picture that we can then act upon within reality, creativity can sit like a layer on top of this, not only are we required to have a conceptual space within which we act, but we should then be able to build on this conceptual map with our existent minds. In some ways we can think of creativity[4. Remember this is the conception, not execution stage] as simply this step, extending perceptions to something we are not consciously aware of. (Maybe not, I'm not selling myself on this, but I'll let the reader toy around with this theory too). Creativity seemed to me to most closely resemble an otherwise confusing bundle of information that can be extended into a concrete idea. As I've said before stringing random words together probably doesn't Shakespeare make[5. Unless they are specifically chosen random nouns], so how do we parse out the art?
And more to my ends how do we then train a computer to parse out the words that we like to hear together -- there isn't really a direct algorithm and I think this is where we can look back to our revised goals for AI for a reason why. If we view information integration as a sort of base upon which we build not only consciousness but creativity also it stands to reason that we would need a full model of human understanding and consciousness in order to construct a truly creative machine. Which was where Minsky and the impression that the focus for AI needs so badly to be diverted back to a general intelligence approach instead of specialized systems.
I could maybe go on, but I have a headache and this post is pretty impressive as-is, however it's safe to say I'll be coming back to this topic in due time. This semester has me all kinds of burnt out with pretty labour intensive courses, hopefully in January I'll be able to have ideas and not just outline things that I muse about. Cool?