Referred to as the “Godfather” of AI in the English-speaking media, Hinton has been working on neural networks, the basic framework behind ChatGPT, the Midjourney image generator and many others for almost half a century. He worked as an expert in the field at Google for around ten years before announcing his retirement on Monday. He now wants to “talk about the dangers of AI,” he wrote on Twitter.
In the “New York Times” (“NYT”), he saw the misuse of AI as an unstoppable problem: “It’s hard to imagine how you can prevent bad actors from using it for bad things,” Hinton told the paper . He fears a flood of disinformation: images, videos and texts that the average person “can no longer distinguish from the truth”.
Criticism of competition between Google and Microsoft
Hinton also criticized the race of the IT giants. Microsoft invested billions in the ChatGPT developer OpenAI and built the technology into the Bing search engine. His previous employer Google recently followed suit and wants to counteract this with the AI Bard. The corporations are in a race that may be unstoppable, Hinton said in the NYT.
Hinton suspects that the race between Google, Microsoft and others will escalate into a global race. Like many of his colleagues, he sees global regulation as the answer, even if it’s difficult: you don’t know whether companies or states are working on the technology in secret. “I don’t think this technology should be expanded further until you understand if you can control it,” he said.
Fear of upheaval on the labor market
Hinton also sees the job market in danger or at least in the process of restructuring. Chatbots like ChatGPT are currently complementary to human labor, but in the longer term, such tools could replace labor, whether it’s translation or paralegal. “They do the tedious work for us,” he said. “You could take away more than that,” Hinton continues.
“The idea that this stuff could actually get smarter than humans — only a few people believed that,” he told NYT. “But most people thought that was completely absurd. And I too thought it was completely absurd. I thought that was 30 to 50 years or more away. Now I obviously don’t think that anymore.”
Warning of independent “sub-goals”
He also outlined apocalyptic scenarios to the BBC, but at the same time pointed out that these were “worst-case scenarios”. For example, he warned against giving robots the opportunity to “set their own sub-goals”. The scientist said this could eventually lead to “sub-goals like ‘I need to get more power.'”
And he pointed to essential differences to human intelligence. “We are biological systems, and these are digital systems. The big difference is that in digital systems there are many copies of the same weights, of the same model of the world.” All of these copies could learn separately from each other, “but immediately pass on their knowledge. It’s like having 10,000 people, and whenever one person has learned something, everyone automatically knows. And so these chatbots can know a lot more than a single human.”
Next AI warning
Hinton is another prominent representative from the industry who warns of the dangers of AI. In open letters, leading scientists warned of similar scenarios – Hinton did not join the criticism at the time. During his employment with Google, he did not want to publicly criticize his own and other companies.
On Twitter, however, he made it clear that his criticism did not refer to his former employer anyway: “Actually, I resigned so that I could talk about the dangers of AI without considering how it would affect Google. Google has acted very responsibly.”
Speaking to the NYT, he said part of him now regrets the work he’s done over the decades. “If I hadn’t done it, someone else would have done it,” Hinton said. He is said to have liked to paraphrase the “father of the atomic bomb”, Robert Oppenheimer: “If you see something that is technically good, then you do it.” “He no longer says that today,” the US daily newspaper now writes about him researcher.
