General Discussion
In reply to the discussion: OMG!! THIS is what Trump and Musk are up to!! (Forget the sensational distractions!) [View all]Metaphorical
(2,581 posts)This has been the technolibertarian dream for a while now - government exists solely as AI "programs" that bypass all of the messy bureaucracy and simply do what they are trained to do. There are a few HUGE problems with this (AI and high tech is my "beat" as a journalist and I've been following this one for a while now):
1. AI Models are based upon training data - that which has already happened - and are consequently very bad at dealing with the unpredictable, which is very often where having people in the loop makes a big difference. The fact that such models also take quite a while to train mean that they are always going to be at least somewhat obsolete.
2. Bias is introduced by determining the training sets, and once introduced is almost impossible to correct for. Reinforcement learning can make a small dent there, but it also makes the model more unstable.
3. Elon Musk has had terrible success with AI in his business ventures: he made the decision to concentrate on AI with the Teslas rather than on improving the vehicles other obvious faults (both with battery and structure) and if anything they've become less intelligent over time even as they've become increasingly known as death traps in the industry.
4. LLMs (large language models) in general hallucinate - they make up information they don't know, without "realizing" that they don't know it, and are not very good at detecting inconsistencies.
5. There is no legal recourse if an LLM is wrong. They are not transparent, they cannot be double checked or analyzed, and they are almost by definition inconsistent. Proponents treat them as if they are intelligent databases, when they are neither.
6. Moreover, as a society, we have no real legal or ethical framework for LLMs. They exist outside of the law, and consequently they are a huge governance risk.
Finally, Musk has a history of creating half-baked ideas, but he also has the financial means to do so in a way that protects them from real world pressures that hone good companies and kill bad ones. They do eventually collapse: his hypertube concept has already done so, his battery production unit is in the process of collapsing, X is a hot mess, Starlink is increasingly being seen as a security nightmare, and SpaceX has had a few spectacular explosions to date - fortunately, no loss of life (yet), but I wouldn't count on that continuing.