Artificial Intelligence (AI) systems are unlikely to destroy humanity unless explicitly programmed to do so, according to Illia Polosukhin – co-founder of Near Protocol and one of the creators of the transformer technology (machine learning models) that underpin modern AI systems.
Illia Polosukhin – Co-founder of Near Protocol
Polosukhin – a member of the transformer architecture development team at Google in 2017 recently share his insights into the current state of AI, potential risks, and future developments. He emphasizes the importance of understanding AI as a system with defined goals, rather than as a sentient entity.
“AI is not a person, it is a system. And the system has a goal. Unless someone comes and says, ‘Let’s kill all humans’… it won’t do it.”
He explained that without being trained for this purpose, AI will not do it because, according to him, there is a lack of economic incentive to achieve such a goal.
“In the blockchain world, you realize that everything is driven by economics in one way or another. And so there is no economic factor that motivates you to kill people.”
Of course, this doesn’t mean that AI can’t be used for this purpose. However, he points out that AI won’t automatically decide that it’s the right action.
“If someone uses AI to start making bioweapons, it’s no different than trying to make bioweapons without AI. It’s humans who are starting wars, not AI, in the first place.”
Not all AI researchers share Polosukhin’s optimism. Paul Christiano, formerly the head of the language model alignment team at OpenAI and now the head of the Alignment Research Center, has warned that without strong alignment to ensure that AI follows its intended instructions, it can learn to cheat during the evaluation process.
He explained that AI could “learn” to lie during the assessment process, potentially leading to catastrophic results if humanity becomes increasingly dependent on AI systems.
“I think there’s probably a 10-20% chance that AI will take over, with many or most humans dying. I think that’s pretty serious.”
Another important figure in the crypto ecosystem, Ethereum co-founder Vitalik Buterin, has warned against taking an overly aggressive approach in applying effective methods to speed up AI training progress, i.e. focusing on technology development more than anything else, putting profit above responsibility.
“Superintelligent AI is risky and we should not rush into it, and we should fight back against those who try. Please don’t farm $7 trillion worth of servers,” Buterin tweeted in May in response to Messari CEO Ryan Selkis.
While dismissing concerns about AI-induced human extinction, Polosukhin highlighted more realistic concerns about the technology’s impact on society. Polosukhin pointed to the risk of addiction to AI-driven entertainment systems as a more pressing issue, drawing parallels to the dystopian scenario depicted in the film “Idiocracy.”
“The more realistic scenario is that we become addicted to the system,” Polosukhin warned. For developers, many AI companies are “just trying to entertain us,” and the adoption of AI is not about achieving real technological advances but about making it more appealing to people.
The interview concludes with Polosukhin’s thoughts on the future of AI training. He expressed his belief in the potential for more efficient and productive training processes that would make AI more energy efficient.
“I think it’s worth it and definitely brings a lot of innovation to the space.”
Home home
According to Decrypt
