Many people know that I am not a big fan of AI, or perhaps of the discourse around AI. I worked for many years as a scientist on AI research including on areas that eventually were included in the Large Language Models (LLMs) that now drive much of modern developments. I understand what many people, hypnotized by the hype around ChatGPT and other similar tools, seem to miss, that these are not, strictly speaking, “reasoning tools”, they do not “think” in any way commensurate with what humans do. They do a good job of mimicking our thinking processes because they handle language well. But their ability to generate “new thoughts” is actually very limited; it is basically a kind of copy of things humans have thought, with perhaps the addition of the ability to combine disparate thoughts, although that also can be questioned. All AI tools perform pattern recognition, LLMs do so on text but other AI tools work off visual or auditory patterns, or indeed, any kind of pattern to be found within data. Humans, of course, do pattern recognition, but our reasoning and creative abilities extend far beyond such things. Indeed, no AI currently in operation does anything like embodied experience. As was the case with earlier developments in AI such as the development of Grand Master level chess players, AI has taught us to rethink what humans do, what it means to be humans.
OK. That’s my reality check. That said, LLMs do certain things well. Because they do pattern recognition on large quantities of data about humans, they are useful for certain kinds of analysis and research. The company I work with uses AI extensively to write grant proposals. What used to take weeks can now be done in a few hours. Their limitations is that that have no real ability to judge “truth”, and so can mislead in remarkable ways if one is not attentive and does not double check the source materials used. They are likewise poor “counsellors” as a result of this inability to judge. And don’t get me onto the issue of writing. LLMs are designed, in a sense, to write. For research reports and that kind of thing they can be highly useful. For fiction, not so much. Fiction, as it turns out, requires the ability to judge truth, as well as embodied experience, two areas where LLMs fail utterly. Novelty is also a problem for LLMs; they can put together disparate units of information to give an appearance of novelty, but are unable to filter these to select ones that are useful to humans. We have to do that work for them.
What I am trying to say is that AI tools can be useful as tools, under human oversight. We have seen some of the problems with AI writing emerge as things have been tried of late. I am largely opposed to the use of AI in the writing of fiction or poetry, partly because I am really only interested in reading material that is grounded in embodied life experience. I have no interest in reading any simulation of that work. And among our Metapsychosis editors we have decided not to accept texts that were developed even in part by AI algorithms, for good reason. A sizeable percentage of recent submissions have the AI “look and feel”, and the quality is lower overall as a result. But we need to work out a general policy for the co-op that both recognizes their utility and acknowledges their limitations. AI is here to stay. The landscape is still changing, but we don’t have to wait to develop our own policies.
Some of the questions I have are: To what extent should we require declarations of AI use in material posted to Infinite Conversations, Metapsychosis and even Untimely Books? And how should this be policed, if at all? What kinds of limits should be imposed? How much space should we give to discussions about AI use? Should there be any limits, in any of our spaces? Personally, I tire quickly of the endless debates around AI, but I do not feel strongly enough about this to suggest any kind of censorship. Are we open to any kind of AI/human interaction in creative content? What about AI-driven research on a topic? I am sure there could be other questions brought to bear. We need a lively discussion on this area to develop our policies as a community.
