Elon Musk has a Moonshot vision in AI. Technology takes on all jobs, but “universal high income” means that anyone can access a wealth of theoretical products and services. If Mask's lofty dreams can become a reality, of course, there will be deep existential calculations.
“The question is really one of the meanings,” Musk said at the Vivatechnology Conference in May 2024.
However, according to Nobel Prize winners and “AI Godfather” Jeffrey Hinton, most industry leaders have not asked this question about the end of AI games. When it comes to AI development, Big Tech is not interested in the long-term outcomes of technology, nor is it interested in rapid outcomes.
“For business owners, it's the short-term benefits that drive research,” said Hinton, professor emeritus of computer science at the University of Toronto. luck.
And for the developers behind the technology, Hinton said, he focuses on the work in front of them immediately, not on the final results of the research itself.
“Researchers are interested in solving problems that they are curious about. It's not about how humanity will be, starting with the same goal,” Hinton said.
“We have these little goals. How do we make them? Or how should we make them? How should we make our computers recognize things in images? We can make them produce compelling videos?” he added. “That's what really drives research.”
Hinton has long been warning of the dangers of AI without guardrails or intentional evolution. This estimates the 10% to 20% chance that technology will wipe out humans after the development of super intelligence.
Ten years after he sold neural network company Dnnresearch to Google, Hinton leaves his role with the tech giant, free to speak out about the dangers of technology, fearing that he will “not be able to prevent bad actors from using it.”
Hinton's AI overall
For Hinton, AI risk falls into two categories. Technology itself is the result of the risks pose to humanity's future and the fact that AI is manipulated by people with bad intentions.
“There is a big distinction between the two types of risk,” he said. “There's a risk that bad actors will misuse AI, and it's already here. It's already happening with fake videos or cyber attacks, and it can happen quickly with a virus. That's very different from the risk that AI itself will become a bad actor.”
For example, financial institutions like Singapore's ANT International are wary of the spread of deepfakes that increase the threat of fraud and fraud. Tianyi Zhang, general manager of risk management and cybersecurity at ANT International, luck The company has found that over 70% of new registrations in certain markets are potential deepfake attempts.
“We have identified over 150 different Deepfark attacks,” he said.
Not only is it defending more regulations, but also calling for action to address Hinton's potential malfunctions, he said, is a sudden fight as each technical issue requires a separate solution. He envisions certifications like future video and image origins that counter the spread of deepfakes.
Just like how printers added names to works after the advent of presses hundreds of years ago, media sources need to find a way to add signatures to real works. However, Hinton said the revisions could only be made to date.
“The problem can probably be solved, but the solution to that problem doesn't solve the other,” he said.
For risk AI itself to pose, Hinton believes that tech companies need to fundamentally change their relationship with AI. When AI achieves super intelligence, he said, there is a strong desire to not only outweigh human abilities, but to survive and gain additional control. Therefore, the current framework for AI that allows humans to control technology is no longer relevant.
Hinton assumes that AI models need to infiltrate “maternal instincts” and that it can treat such powerful humans with sympathy, rather than wanting to control them.
Evoking the ideal of traditional femininity, he can quote that something more intelligent falls into the upset of a less intelligent person is the baby who rules the mother.
“And I think it's a better model that we can practice with super intelligent AI,” Hinton said. “They become mothers and we become babies.”
