Godfather of AI: Artificial intelligence will surpass human – 08/05/2023 – Tech

Generative artificial intelligence (AI) models like ChatGPT have basic knowledge about any subject and therefore are 1,000 times smarter than the average human being. The assessment is from the computer scientist known as the “godfather of AI”, Geoffrey Hinton, who asked to step away from Google this Monday (1st).

For Hinton, these models will be able to overcome collective human intelligence in the coming years. He discussed, on Tuesday (2), the future of artificial intelligence at the EmTech Digital event, organized by the MIT Technology Review magazine. “The risk of AI taking over society is real.”

This argument from the godfather of AI is in line with the letter to stop the development of artificial intelligence signed by intellectuals and technology influencers such as writer Yuval Harari and Apple co-founder Steve Wozniak. Hinton himself, still at Google at the time, did not endorse the document.

He says he left Google because he wanted more freedom to talk about his concerns about AI risks.

Harari wrote in an article in the New York Times that a plane with a 10% chance of crashing would never take off. For Hinton, the chances of humanity losing control over this new form of intelligence represented by the great language models, such as the GPT, is around 40%. “I’m not pessimistic, not optimistic, but mildly depressed.”

Hinton developed in the 1980s the algorithm behind the current neural networks, the backpropagation (reverse propagation, in English). In short, this algorithm allows machines to learn from examples, in what is called training. The scientist tried to simulate the functioning of the neurons and connections of the human brain to elaborate the technique.

For having developed reverse propagation, Hinton received, alongside Yann LeCun and Yoshua Bengio, the 2019 Turing Prize, known as the Nobel Prize in mathematics.

Today, Hinton assesses that the human brain does not work with reverse propagation. “The algorithm we created in the 1980s is more efficient than the brain. Today’s great language models can know more than a person, with far fewer connections.”

Current AIs like GPT-4 available in the paid version of ChatGPT have less than 1 trillion calls. The human brain has 10 trillion, out of 100 billion neurons. “It’s a new form of intelligence, without the energy limits and inherent vices of biological evolution.”

Questioned by Professor Luís Lamb, from UFRGS (Federal University of Rio Grande do Sul), Hinton stated that generative AIs will be able to innovate, just like human beings. “Current models have not yet been trained to develop complex logical reasoning. This is possible if these technologies gain objectives.”

This would be one of the risks, according to the scientist. “Artificial intelligence will evolve further when it learns to break down processes. From there, technology can learn that taking responsibility speeds up task completion. This can lead to a control bias.”

He added that GPT-4 does not function as a statistical parrot, which repeats most likely answers from available text-based knowledge on the internet. “I told ChatGPT that I had a house with white, blue and yellow walls. The yellow ones faded and turned white after one year. So I asked what I needed to do to have all the walls white in two years.”

GPT-4 replied that the scientist should paint the blue walls yellow. “This is not simple reasoning. The robot needs to understand the concept of fading and time. And the AI ​​didn’t give the trivial answer, which would be to paint all the walls white.”

Hinton says that when he started researching the subject, in the 1970s and 1980s, there was no way to predict the dimensions that this problem would have, compared by him to climate change. “So I don’t blame myself for doing science.”

The godfather of AI also praises Google. The company developed the first transformer, a model that gave rise to the current language models. “Google realized the risks, created security commitments, and moved forward with caution.”

For Hinton, the new platforms launched by OpenAI, ChatGPT, and the Dall-E image generator, changed the scenario. “The Microsoft-sponsored startup brought these products to market, and Google had to respond. That’s how competition works.”

Though concerned, the godfather of AI is skeptical of a pause in AI development. “If the United States stops, how can we guarantee that China would also end its research in the area or even create weapons with it?”, He asks. “Disputing for leadership is something inherent in capitalist society”, he adds.