Hey Hi

Artificial Intelligence (AI) has been a big media topic over the past year or so. I cannot say I have a fantastic grasp of exactly what it is or what distinguishes AI from other kinds of computer activity. Even the question of exactly what intelligence could be is beyond me. I even think of a choose your own adventure book as a simulation of intelligence in paper form.

The key distinction to draw however is that the writer of such a book or program understands exactly how it works and what it can do. The latest generation of AI is different - even its own programers cannot understand exactly how it works or what results it will produce. Another factor that has made it controversial is that it can engage in seemingly creative acts that till recently only humans could do. Now machines can do much more than repetitive manual work or the purely logical processing of information. The Internet is a big factor here and AI can draw on a massive body of human-generated creativity in its own complex act of regurgitation.

My concern over this is eclipsed by an older worry over how mobile Internet has been altering human interactions. And my own work as an integration aide in adult eduation is intensely human-facing so will still be a job for some time to come. But I may as well get a passing familiarity with some of the latest AI tools. To that end I played with two particular applications.

* * * * *

I chatted with something now called Gemini - a generative AI chatbot that can hold a conversation and produce written content on command. I acted as if it were a person and complimented it on its polite manner. I also asked it what it would do if someone were rude to it. It told me it would ultimately end such a conversation. I was impressed. Evidentally its programmers at Google have put various parameteres on its behaviour. It results in some overly repetitive language as it apologetically stipulates what it can and cannot do. I can relate. I asked it for feedback on my own writing and it advised that I could be more succinct and use more active language. This is true. Next I gave it some more substantive tasks.

I asked for a description of a medieval fantasy village. It did pretty well but some aspects of the setting felt more Midsomer Murders than The Forgotten Realms. I asked it to describe a mid-sized faster-than-light spaceship that uses centrifugal force to simulate gravity. Once more it did well but completely neglected the bit involving spinning circular forms. It can definitely produce flowing and grammatical word-count that a human can then tidy and personalize (assuming that they bother).

Next I asked it to make some judgements on matters non-fictional. What characteristics both unite and distinguish classical liberalism, democratic socialism and traditional conservatism? It did well in drawing on academic definitions but there were some nuances I felt needed adjusting. Hardly surprising given my own qualifications. I asked it if my statement "funk is to soul as metal is to rock" made sense and it gave grounds for saying that it did. However it cannot recommend a flowing selection of tracks to save its life.

One thing I neglected to do was ask it about patently dodgy topics - it would be interesting to see how it responded to a conspiracy theory or bogus conception of the world. But I'm wary of even entertaining such notions online. I prefer these tools to be for fun.



I recently resumed GMing some role-play games and hunted online for free artwork to illustrate characters or locales. I noticed something was suddenly different. There was a lot of content from a handful of new websites. It was superficially decent but the closer I looked the more flaws I noticed. The overly idealized yet creepy faces. The deformed hands. The accidentally Escheresque architecture. I realized what I was seeing was a host of collaborations between AI and users describing what they were imagining. And then I decided, if you cannot beat them, join them, and started experimenting at a website called Night Café.

The resulting pictures were fine for my purposes of printing them as small grey images to flash at players as I described non-player characters. But then I moved onto something I hoped would work with the flaws of this new method. Toys are caricatures of reality. They are mass-produced and simplistic. They depend on commonly understood concepts and imagery. Hence I created The Sub-Culture Kids Action Figure Selection.

These are images of retro action figures depicting youthful members of well-known sub-cultures from a variety of backgrounds. They were produced by wrangling text over several fiddly and frustrating iterations. Nonetheless this method of producing images with colour and perspective was far quicker than anything I could do manually. And some of them are better than others.

Raving Rina is the best - even the characteristic AI mistake of blending her headphones with her hairstyle somehow works. Skating Suzana comes complete with an accessory. Hugo Hearts Hip-Hop and Grungy Gus are decent variations on each other but the latter needs more long and lank hair that I never managed to coax from the AI. Feral Faris looks like a hippy but is supposed to be of the more recent iteration of feral. Finally Gothic Greta seems like she comes from another toy line altogether - her proportions are too mature but older drafts of her looked far too childish.

It seems like I'm blaming my tools. I jumped right in rather than looking for elaborate instructions or getting advice from other Night Café users. But the whole attraction of this stuff is in time-saving. If I want I can use such imagery as drafts from which to hand-draw things that I can more exactly control. For now I have some sense of what the latest applications of AI can and cannot do.