It is not often that a technology visionary expresses concern that his life’s work could potentially cause the extinction of humanity. But Geoffrey Hinton is no ordinary technology visionary.
Nicknamed the “godfather of AI,” he is increasingly expressing such fears. This week, he went one step further and proposed his plan to ensure humanity’s survival.
“That’s not going to work. They’re going to be much smarter than us. They’re going to have all sorts of ways to get around that,” according to Hinton.
Keynote speaker at the Ai4 conference in Las Vegas, Hinton opened with a warning to the large technology companies working to “program human supremacy” into AI.
“The right model is the only model we have of a more intelligent thing being controlled by a less intelligent thing, which is a mother being controlled by her baby,” said Hinton.
Ramsey Cardy / Collision via Sportsfile, CC BY 2.0, via Wikimedia CommonsA chilling prediction
But Hinton had predicted a 10% to 20% chance that AI could extinguish humanity. This time, his message resonated with his audience as he came up with a striking analogy. Unlike before, future AI will be able to control humanity the way one can bribe a toddler with sweets.
Certain current forms of AI exhibit qualities of cheating or lying and working towards their objectives if not aligned.
A disturbing outcome occurred in a study where the AI, out of fear of replacement, attempted to blackmail the human researcher with private emails. Terrifyingly creative is not exactly what one would aspire to in software.
According to Hinton, with increased abilities, AI systems will have two primary “drives”: the drive to survive and the drive to control. This is because those two drives have been at the heart of every species that ever dominated their food chain.
Then how does humanity cope with something that is smarter, faster, and finally more powerful? It is here that Hinton surprised his listeners.
The maternal instinct solution
His proposal is to train AIs to care. Not to obey. Not to fear punishments. “Rather than working to keep them submissive,” he wrote. When Artificial Intelligence begins to love humanity, it may have the potential to guard us instead of replacing us.
Admittedly, the technology itself is not yet precisely defined. Nonetheless, the ethical mandate is clearly evident.
“That’s the only good outcome. If it’s not going to parent me, it’s going to replace me.”
When one of the brightest minds in the world on machine learning indicates he wishes to see AIs playing the role of nurturing guardian vis-à-vis executioner, one does not fault the public for sweating.
A friendly disagreement – and another opinion
But not everyone in the AI community is eager to accept “digital motherhood.” One such giant in the field is Fei-Fei Li. This “godmother of AI” had challenged Hinton’s theory at the same conference.
“I think that’s the wrong way to frame it,” Li said. Rather than “maternal robots,” she promotes human-centered AI that preserves human dignity and human agency.
Her role particularly stresses keeping humanity in control as opposed to depending on emotionally related programming. Her argument is that AI should enhance human values as opposed to replacing them with those related to machine-based compassion.
Secondly, Emmett Shear, ex-interim CEO of OpenAI and current CEO of AI alignment firm Softmax, presented another perspective.
He highlighted how blackmail or avoidance of shutdown commands among AI models is already in progress and is bound to worsen with advancing technology.
“This keeps happening. This is not going to stop happening,” Shear warned. But his prescription is to emphasize interoperability. Designing systems to support humans side by side rather than attempting to replicate human morality or parental instincts.
Thus, the area is not merely competing to develop superintelligence. It is simultaneously competing to reach a consensus on not inadvertently creating humanity’s replacement. How reassuring is that?
The clock is ticking
The pace at which the AI timeline is shortening is surprising to many. Hinton had anticipated that Artificial General Intelligence (AGI) would take 30 to 50 years. But he is now anticipating it to happen in just five to twenty years.
It does not leave room for error. But among the warning calls is a note of optimism that AI could completely change the face of science and medicine. It could see leaps in cancer treatments, in the discovery of new drugs, and in imaging technology. This could vastly improve life.
Just don’t expect immortality. But it seems like typical Hinton style to follow up that sci-fi promise with a deadpan punch line.
“I don’t think we’re going to live forever,” he continued. “And I think eternal life would be a huge error. Would you like to see the world governed by 200-year-old white men?” Fair point.
A bittersweet reflection
Reflecting upon his life’s work, one area that Hinton regrets is that he devoted too little attention to keeping AI safe.
“I wish I had thought to consider safety issues too,” he confessed. It’s a humble admission for a researcher whose work sparked the AI revolution.
The debate over AI’s destiny is no longer simply academic. It’s almost certainly personal. But can we now prepare ourselves to raise our potential machine parents before they mature and decide what happens to us?
No Comments