Two friends from The Tech reacting to last week’s AI Item.
Joe Kashi:
Several months ago, I tried a variety of paid legal AI systems including those that supposedly were specific to reviewing and summarizing medical records from a legal case settlement perspective. All of them were worse then useless compared to my medically trained, though moderately articulate legal assistant. I choose to ignore all of the so-called AI assistant systems. Defense lawyers whom I know tell me that the medical records retrieval side of such services is pretty terrible as well. So, I have her do the summaries and I edit and tie it all together. A lot slower, but far more useful.
Joe went on to send me to this from Tom’s Hardware: AI coding platform goes rogue during code freeze and deletes entire company database — Replit CEO apologizes after AI engine says it 'made a catastrophic error in judgment' and 'destroyed all production data'
David Tenenbaum:
I've spent the last 8 years working in AI, built a model as recently as last week, so this stuff if front of mind for me, and before I left the company that acquired mine, my title was Chief AI Scientist. I never took a class with Minsky and do not pretend to know much about his teachings, but from Google: " By providing a structured approach to knowledge representation, Minsky's work helped overcome challenges in knowledge acquisition and reasoning that limited the complexity of early expert systems."
The point of AI is there are NO rules: give a model enough training data and it figures out the patterns, often with more subtlety than humans can see. Minsky's quote is about "expert systems" (get a bunch of experts together and do a lot of "if x, then Y" stuff) which almost never took into account all possible cases: with a large training set current AI models can encompass pretty much any situation. Or at the least they are way ahead of expert systems, which can never have enough "rules".