all about AI
AI basics for writers
Your presence is acknowledged, content-viewers. That was my first thought when trying to imagine how “welcome back to Sequel Country” might sound if this column was written by “AI” (which, by the way, no part of it is, except where clearly attributed). I touched on using ChatGPT to generate ideas in part 28 on characters, but in hindsight I thought I should go into much more depth, since hardly anyone understands what is going on with AI at the moment. So let me explain, and I’ll come onto how AI affects writers after that.
I work with a lot of young software engineers who are very excited by AI, because they lack the background to understand what’s really going on, or the experience to know that AI researchers have claimed that AI is very nearly on the point of a breakthrough every 5-10 years for the 40 years that I’ve been following the IT industry.
They are excited because AI is being hyped at the moment, which is because the software industry needs to regularly hype the next big thing in order to milk money out of investors. And the easiest way to do this is to get them to fear being left out of the next gold rush – a confidence trick which capitalism has been pulling since the South Sea Bubble of 1720, and it still works!
This is possible because the general public still doesn’t understand the complexities of AI. The public perception hasn’t progressed much since the late 60s, when expert systems and voice synthesizers were invented, and sci-fi put the two together. Stories in Star Trek and The Prisoner and Logan’s Run showed computers that could appear to understand issues as an expert, but if you asked them a question containing a logical contradiction, the computer would seize up and messily explode (without ever quite explaining why they contained highly explosive components, rather than normal silicon chips).
What has changed recently, with “large language model” (LLM) or “generative AI” systems is that they can now process such huge quantities of raw data (“training models”), and do it so fast, that they are better than ever before. And this does make them useful for some new things.
However, all a generative AI is good at is predicting what would be a good next word in a string of words, based on its training model. That is all it does. Like a computer-vision system that can tell the difference between your face and someone else’s, it’s a useful step towards one day building an artificial intelligence, a possible component; but it does not on its own constitute intelligence. It will include any flaws, biases and inaccuracies in its sources; and because it’s probability-based instead of being based on a rigorous logical algorithm like the old “expert systems”, it does not have any way of detecting (and thus becoming aware of) its imperfections. It’s very easy to get ChatGPT to make directly contradictory statements about the same thing, and it doesn’t know it’s doing it. We are calling it AI, but it is not Artifical Intelligence. AI here is nothing more than a convenient name.
So of course, since nobody addresses their readers as “content-viewers”, neither would an AI. I could generate parts of these columns with the help of an AI – it might take me a few tries to get something sufficiently convincing, but such things can, and have been, done. (Not by me, because I don’t think it would be ethical to do this without telling you.) And because generative AIs are getting so plausible-sounding, they’re starting to worry the Society of Authors, which I’m a member of, and its American equivalent the Authors Guild. They have written to AI companies urging them to stop ingesting writers’ works “without consent, credit or compensation”. The worry, already starting to come true, is that pirated copies of books are being fed into generative AI, and used to produce creative works on the cheap. The thing is, generative AIs are not intelligent, but they can be creative, in the sense of putting together words that haven’t been put together before.
And that is actually what makes them potentially useful to writers, right now. In part 28 I talked about my technique of repeatedly asking questions about a character until it sparked off a new idea. I realise now I should have started from the basics before coming to the practice, so here we are:
You can access generative AIs on the internet by typing messages to “chat bots”. Chat bots are really a user interface between you and the generative AI system; a mediator.
ChatGPT is one of the best, but it’s also one of the most expensive. You can register for an account to talk to it via a web browser, but with only a free account, it will slow down and stop.
There are lots of pretty good free alternatives, at least for now while companies are trialling new features. I use an app called Poe which offers quite a variety of different generative AIs. They each have different strengths and weaknesses, not least due to the different training models fed into them; I’ve found GPT-4 and Claude-instant to be useful for helping with creating writing. I’ve heard Google-PaLM is useful for non-fiction writing.
One of the key things to understand about chat bots is that they have context. This is based on what you have asked before, and any text you have uploaded into it. Answering questions in the context of what you’ve already asked is massively useful, because you don’t have to start every question from scratch. Some AIs will drop your context after a while, to save memory; they will tell you when they do this. If they don’t say anything about it, assume that everything you send to the AI is being stored and will become available to other users!
Poe and similar systems let users set up their own bots. An AI bot is simply a combination of an existing AI plus a context, such as some text or one or more setting-up questions (which you, the user, don’t get to see). I’ve had some good results with a Poe bot called BookBot09 which is set up to assist with novel writing. There are lots like it; Poe has an Explore section with a whole subsection for writing-oriented bots.
So, what can a writer do with these bots? There’s another whole column’s worth of things to say about that coming next week. I wanted to cover the “how” first, to make sure anyone who fancies playing with this stuff understands how easy it is, and how limited it is: like I said, generative AIs are just algorithms for stringing words together convincingly. They do this mindlessly, but the fact that you can set their direction to do it is what’s useful and new. And you can do this right now, for free, using only a web browser or free app.
Give it a go if you like, and we’ll carry on up the AI trail next week. Until then, fellow pioneers!