Gongol.com Archives: June 2023
June 5, 2023
Where can we find the emergency brake?
New technologies almost invariably bring about both good and bad uses; going all the way back to the first tools invented by our prehistoric ancestors, we find that the same knife that can be used to hunt game to feed a family can also be used to commit murder. With very few exceptions, technology is value-neutral; its good or bad use is in the hands of the people using it. ■ The human element, which not only involves operational supervision but also judgment about its use, is what makes the explosive arrival of large-scale artificial intelligence tools such an imminent cause for concern. On one hand, we have this widely-signed statement of the Center for AI Safety: "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war." Heady stuff. On the other hand, we have AI-topians sharing "hacks" like "Here's how to get a professional therapist for free [by training ChatGPT]". ■ As researcher Neil Renic notes of that latter group, "These people are rightly ridiculed for their exaggerated faith in AI. But worse is their miserable lack of faith in humans and meaningful connection between humans." It's telling, isn't it? Instead of discussing how the technology could be used to train more and better human therapists (which is not only a conceivable goal, but a highly worthwhile one!), certain of the AI-topians want to fervently believe that a conscience-less technology can take a human's place within a deeply intimate relationship. ■ There are right and wrong answers when it comes to new technology. We don't know all of them yet, but it's imperative to get to work on defining some workable heuristics as swiftly as possible. A good starting heuristic might go like this: Are we using AI to replace human judgment? Danger! Use extreme caution. Are we using AI to enhance human judgment? Proceed carefully, putting safeguards in place to slow down the unintended consequences and installing kill-switch options wherever possible. ■ Warren Buffett cautioned an audience about AI last month: "When I'm told something can do all things, I get worried because you can't uninvent it. We invented the atom bomb out of necessity during World War II, but was it really a good thing for the next 200 years? AI can change a lot, but it can't change how people think." What makes artificial intelligence unique in this regard is that its overwhelming use advantage is to act faster than human judgment can supervise. ■ Becoming too enamored with that power puts us at risk of omitting vital checks along the way: The tracks of a railroad certainly automate away a huge number of individual decisions that would otherwise need to be made by a conductor steering a load down a road, but we still give train engineers emergency brakes to help avoid or mitigate disasters. Sooner or later, we'll uncover the need for similar "emergency brakes" on AI technologies, to ensure that human judgment still plays a dispositive role. Let's hope it's sooner.
