If you believe Google’s AI researchers, rapid leaps in AI capabilities aren’t necessarily ushering in a horrific Terminator-style future… but they might be.
What happened: Google’s DeepMind released a new paper claiming that it’s “plausible” for high-level artificial general intelligence (AGI) — AI systems that can do most nonphysical tasks as well as or better than nearly all humans — to arrive by 2030.
- The researchers suggest that AI systems could soon begin to automate the development of more powerful versions of themselves, kicking off an accelerating feedback loop.
Why it matters: If Google’s researchers are correct, we’ll need to figure out how to keep highly capable, self-improving AI systems from running amok or being weaponized by malicious users — and do so in the next couple of years.
- Even if doomsday predictions never leave the pages of sci-fi, the emergence of AI that could do most white-collar jobs as well as humans at a fraction of the price would be an economic and social earthquake.
Zoom out: In another recently published piece, a former OpenAI researcher (with a track record for eerily accurate predictions about AI developments) posits that an AI arms race between the U.S. and China could threaten human civilization by 2028.
Yes, but: Not everyone agrees that these forecasts are accurate — one researcher at the University of Alberta argued that predictions of imminent AGI depend on AI systems being able to improve themselves, which has yet to happen.—TS