Although various forms of artificial intelligence have had behind-the-scenes applications across numerous industries for quite some time now, up until fairly recently, a layman might be most familiar with AI through sci-fi depictions of the technology. In works like “2001: A Space Odyssey”, “The Terminator”, and “The Matrix”, AI is often a villain leveraging its greater-than-human intelligence to bring about our doom. Unfortunately, at least one figure heavily involved in the invention of AI as we know it today believes there’s a genuine possibility we could face serious consequences if an AI’s intelligence ever exceeds that of a human.
Geoffrey Hinton knows a thing or two about this tech. A Nobel laureate who received his PhD in Artificial Intelligence from Edinburgh way back in 1978, Hinton’s been working in the field for decades, and has even earned the unofficial moniker “the godfather of AI.”
Speaking to an audience at the Legislative Summit of the National Conference of State Legislatures (NCSL) in Chicago, Hinton addressed the fact that “a lot of the public are now worried about [AI].” He might not have helped ease such concerns when he added “they’re correct to be worried about it.”
Hinton made it clear that it’s impossible to tell what will happen as AI continues to develop. Speaking to attendees, he explained “nobody knows what the future’s going to be like.” However, he also referenced the fact that AI development’s pace has been faster than expected, causing him to muse about the price humanity might pay for unchecked AI expansion.
AI experts discusses how AI can undermine humans to achieve goals
Discussing the potential downsides of rapid AI development, Hinton cited a common concern: loss of jobs. He also commented on the fact that AI can disrupt general trust in the validity of information people find themselves presented with. When anyone can use AI to generate images, videos, and audio, it’s difficult to know what is and isn’t real. Naturally, bad actors can manipulate this distrust to further their own aims.
Perhaps more frighteningly, Hinton pointed to actual examples of AI planning to take action against its human operators when the AI determines that doing so is necessary to sustain its own existence. Hinton specifically described an experiment in which an AI was trained on a fictional company’s internal emails. Through these fake emails, the AI “learned” that one of the company’s engineers was involved in a romantic affair with another employee.
The team behind this informal experiment then showed the AI a message indicating the engineer having the affair was going to replace the current AI with another AI product. Thus, without any human prompting it to do so, the AI outlined a blackmail plan in which it threatened to expose the engineer’s affair if they moved forward with replacing it.
This demonstrates how an AI may “go rogue” to ensure self-preservation. It isn’t necessarily that the AI wants to live the same way a human does. The AI is simply designed to complete certain tasks. In situations like the one described above, an AI may take significant measures to ensure it stays “alive,” otherwise it can’t complete the tasks it’s programmed to complete.
The Godfather of AI on the ‘existential threat’: ‘Nobody knows how to solve it’
Hinton’s remarks at the NCSL’s Legislative Summit don’t represent the first time he’s sounded the alarm on AI’s threats. Speaking to the Senate of Canadawhen asked about the “existential threat” that AI could pose to humanity, Hinton conceded that no one currently knows what solution will be effective. Hinton pointed out that humans are still technically the ones building AI. As such, he stated that we may be able to program it with “maternal instincts” so that it will “love us” instead of viewing us as a threat.
Hinton made it clear how dire the situation is. He flatly stated “we’re toast” if humanity fails to solve the existential threat problem before AI’s intelligence exceeds human intelligence.
Those who claim Hinton is exaggerating the risks ignore his credentials. For what it’s worth, there have already been instances when popular AIs have admitted they would kill humans for the purposes of self-preservation. Although today’s common AI models may not have achieved sentience yet, that doesn’t mean they never can.
Why one of the key figures behind AI thinks AI regulation is essential
Hinton’s statements aren’t all doom-and-gloom. For example, he pointed out that AI is already proving to be more effective at diagnosing injuries and illnesses than human doctors. If these capabilities continue to expand, it’s easier to imagine how AI could significantly benefit the human race.
However, if AI becomes superintelligent and decides humans are its enemy, any potential advantages of the tech will pale in comparison to the threat it poses. That’s why Hinton has emphasized the need for greater regulation of AI development, as well as more research into ways to manage the various risk factors.
Hinton admitted that tech companies aren’t fans of regulation. He stated they often see innovation as being similar to a car’s accelerator, with regulation being more like the brakes. Hinton has advocated for a different perspective, in which “regulation is the steering wheel,” ensuring that innovation allows “progress to happen in the right direction.”
For what it’s worth, Hinton isn’t the only one voicing concerns. Workers across multiple industries have pointed out that AI can be a dangerous tool. Luckily, experts like Hinton at least seem to believe it’s not too late (yet) to prevent the danger from growing beyond the bounds of our control.
