In the 1960s, mathematician IJ Good put forward an idea that has since become the watchword of Silicon Valley. He said that if we create an ultra-intelligent machine, it will be able to design even better machines. Then there will be an explosion of intelligence that will leave human intelligence far behind. In such a situation, the first such machine will prove to be man’s “last invention”. This prediction – which was once the stuff of science-fiction – has today become the objective of the world’s powerful institutions. But even if we assume that future systems can generate new solutions that go beyond today’s models, the theory of the “last invention” will still be in question. We must remember that innovation is not a seamless race from idea to impact. Instead, the discovery process is like a chain, in which there may be a weak link. These weak links determine most of human progress. In 1986, the Space Shuttle Challenger disintegrated just 73 seconds after its launch. Not because its world-class engines or software had malfunctioned, but because a small rubber seal had failed when exposed to cold atmospheric temperatures. It was called O-ring. He has become a metaphor for the critical bottlenecks that can cause even the most sophisticated systems to fail. Artificial general intelligence (AGI) may dramatically accelerate early-stage medical research, but if it cannot advance clinical trials, be manufactured, or receive regulatory approvals, its so-called break-through will never become an invention that improves our lives. Even when the initial stages of discovery are automated, the role of humans does not go away; She simply shifts towards the remaining obstacles. And these are the challenges where judgment, underlying knowledge and practical skills matter most. AGI will not only have to outperform humans; Rather, it will have to do so using AGI. If the “last-inventor” theory is to be true, we humans would also have to become unnecessary as companions or supervisors of AI. This is still a long way off. In 2016, Google DeepMind’s superiority seemed certain after AlphaGo defeated top professional player Lee Sedol 4-1. But in 2023, researchers showed that the top engines failed in unusual conditions outside their training. Even a human with modest computing skills can defeat them. The “ultimate-invention” theory also assumes that all relevant information can be codified, but this is usually not the case. The information that runs complex systems is often dispersed, local, and tacit. Knowledge is not a portable thing. In such a situation, calling AGI the last invention of humanity is an exaggerated claim. (@ProjectSyndicate)
Source link
Carl Benedict Frey’s column: AGI for us humans "’The last invention’ will not prove
