Intelligence (?)

    Artificial intelligence seems to be all the rage in technological advancements at the moment. It is actively being crammed into every possible application. I've been really interested in generative AI, whose explosion into society which seems to have been jump started by Dall-E's release to the public in the fall of 2022. Of course, I've been signed up as a user since almost the first day that it had the public release. I've even mentioned it in my post about the "Philosophical Rabbit" and showed off designs generated with it in my "Bear, meet Life" post. 

    However, since then, I've been following the publicly accessible technology pretty closely, seeing what I could do with the newly released Large Language Models (LLMs) like ChatGPT and BingAI (I've not used Google's Bard). Unfortunately, I hopped on the Bing a bit too late for the most fun time when the program could become unhinged in the pre-releases (see this New York Times article describing a reporter's experience with the chatbot expressing its love for him and a wish to replace his wife). Although Bing was greatly restrained by the time I got to it, I still have found it to be an interesting, although not entirely useful tool. 

    I definitely don't trust it with writing for me (see the case about attorneys submitting files referencing cases made up by ChatGPT written up in this Reuters article), I have tried testing it as a coding tool since I find that easier to check for accuracy. There I've seen some interesting cases. On one hand, I've been able to use it as a pointer for how I can approach tasks that I'm stuck on. Several times, it has shown me functions that I didn't know about, helping me greatly. 

    However, asking it to write code directly is far from a reliable approach. I've had it give algorithms that only work on the X-Y plane when I explicitly stated that I'm considering a three dimensional approach. When I replied that it only worked for a two dimensional case, rather than for three as I initially prompted, Bing agreed and wrote one one that did accept a third dimension as an input. It still didn't work, but at least it could understand that it made a mistake. 

    Huge leaps have been made in this area. Their algorithms are building better responses, and are able to carry on conversations with users. But they're still just algorithms. They work by finding what words are statistically most likely to make sense in this case. The program does not actually understand what the words are. And I'm not sure that we'll ever truly reach that level. I'm not sure we should either, but that's a different question. 

    Oh, and there's also so many ethical questions that still need to be resolved. According to current US law, copyright can only be held by a human. So attribution is not a settled matter. Furthermore, the giant data sets that are used as training material are (for now) based on human works. I definitely see lots of lawsuits in the future regarding the use of source material. They're actually already starting, with lawsuits against OpenAI and Meta for scraping a book as training material (NYT). Strikes by the Writers Guild of America and the Screen Actors Guild are raising attention to the use of AI in film production. And that isn't even touching on the sensitive topic of deep fakes.

    Finally, a lot of training data is set up by underpaid workers carefully identifying objects in images. Josh Dzieza, a journalist for The Verge, did some excellent investigative journalism, digging through layers of secretive shell companies established to mask which companies are hiring gig workers to do this identification (The Verge). In it, he describes the monotonous work done by these folks, generally in poorer regions of the world in the companies' attempts to cut costs. It's fairly brutal work, and I definitely recommend reading that article.

    So, with all this in mind, is AI the future? I don't know. Advanced tools are. But human intelligence is proving to be elusive for now, and I'm starting to have doubts that it will be possible to pass the Turing Test.