Microsoft says the new AI is showing signs of human thinking

When computer scientists at Microsoft began experimenting with a new AI system last year, they asked it to solve a puzzle that should have required an intuitive understanding of the physical world.

“Here we have a book, nine eggs, a laptop, a bottle and a nail,” they asked. “Please tell me how to stack them on top of each other in a stable manner.”

The researchers were amazed at the ingenuity of the AI ​​system’s answer. She said put the eggs on the book. Arrange the eggs in three rows, leaving a space between them. Make sure not to break it.

“Put your laptop on top of the eggs, screen facing down and keyboard up,” she wrote. “The notebook will fit snugly into the borders of a book and an egg, and its flat, hard surface will provide a stable platform for the next layer.”

The intelligent suggestion made the researchers wonder if they were witnessing a new kind of intelligence. In March, they published a 155-page paper arguing that the system was a step toward artificial general intelligence, or AGI, which is short for a machine that can do anything a human brain can do. The paper has been published on an online research repository.

Microsoft, the first major tech company to publish a paper making such a bold claim, has sparked one of the tech world’s most tested debates: Is the industry building something resembling human intelligence? Or are some of the smartest minds in the industry letting their imaginations get the best of them?

“I started to be very skeptical — and that developed into a sense of frustration, annoyance, and maybe even fear,” said Peter Lee, who leads research at Microsoft. “Do you think: where does this come from?”

Microsoft’s paper, which provocatively calls “Sparks of Artificial General Intelligence,” goes to the heart of what technologists have been working for — and fearing — for decades. If they build a machine that works like the human brain or even better, it could change the world. But it can also be dangerous.

This could be complete nonsense. Making claims of artificial general intelligence can be a reputation killer for computer scientists. What one researcher thinks is a sign of intelligence can easily be explained by another, and the debate often seems more suited to a philosophy club than a computer lab. Last year, Google fired a researcher who claimed that a similar AI system was sentient, a step further than Microsoft claimed. A conscious system will not just be smart. He will be able to sense or feel what is happening in the world around him.

But some believe that in the past year or so the industry has come a long way from something inexplicable: a new AI system that comes up with human-like answers and thoughts that weren’t programmed into it.

Microsoft has reorganized parts of its research labs to include multiple groups dedicated to exploring the idea. One of them will be run by Sebastien Bobek, who was the lead author of the Microsoft AGI paper.

About five years ago, companies like Google, Microsoft, and OpenAI started building large language models, or LLMs. These systems often spend months analyzing vast amounts of digital text, including books, Wikipedia articles, and chat logs. By identifying patterns in this text, they learned to create their own text, including term papers, poetry, and computer code. They can even hold a conversation.

The technology that the Microsoft researchers were working on, OpenAI’s GPT-4, is the most powerful of these systems. Microsoft is a close partner of OpenAI and has invested $13 billion in the San Francisco company.

Among the researchers was Dr. Bobek, a 38-year-old French expatriate and former professor at Princeton University. One of the first things he and his colleagues did was to ask GPT-4 to write a mathematical proof showing that there are infinite prime numbers and to do so in a consistent way.

The poetic proof of the technology was so impressive – both mathematically and linguistically – that he had difficulty understanding what he was talking about. “At that point, I was like: What’s going on?” he said in March during a symposium at the Massachusetts Institute of Technology.

For months, he and his colleagues documented the complex behavior displayed by the system and believed it demonstrated a “deep and flexible understanding” of human concepts and skills.

Dr. Lee said that when people use GPT-4, they are “amazed at its ability to generate text”. “But it turns out to be a better way of analyzing, synthesizing, evaluating, and judging a text than creating it.”

When they asked the system to draw a rhino using a programming language called TiKZ, it immediately created a program that could draw a rhino. When they removed the code extension that drew a rhinoceros horn and told the system to modify the program so that it would draw a rhinoceros again, it did just that.

They asked him to write a program that took into account a person’s age, sex, weight, height, and blood test results and judged whether they were at risk of developing diabetes. They asked him to write a letter of support for Electron as a US presidential candidate, in the voice of Mahatma Gandhi, addressed to his wife. They asked him to write a Socratic dialogue that explored the misuse and dangers of LLM

She did it all in a way that seemed to demonstrate an understanding of disparate fields like politics, physics, history, computer science, medicine, and philosophy while combining her knowledge.

“All the things you thought he wouldn’t be able to do? He was definitely able to do many of them — if not most of them,” said Dr. Bobik.

Some AI experts saw Microsoft’s paper as an opportunistic attempt to make grand claims about a technology that no one fully understood. The researchers also argue that general intelligence requires familiarity with the physical world, which GPT-4 theoretically does not have.

“‘AI Sparks’ is an example of some of these big companies grasping for the paper format in PR contests,” said Martin Sapp, a researcher and professor at Carnegie Mellon University. “They acknowledge literally in the introduction to their paper that their approach is personal and informal and may not It meets the rigorous standards for scientific evaluation.”

Dr. Bubeck and Dr. Lee said they were unsure how to describe the behavior of the system and eventually settled on “Sparks of AGI” because they thought it would capture the imagination of other researchers.

Because Microsoft researchers were testing an early version of GPT-4 that was not well-tuned to avoid hate speech, misinformation, and other spam, the claims in the paper could not be verified by outside experts. Microsoft says the publicly available system isn’t as robust as the version they tested.

There are times when systems like GPT-4 seem to mimic human thinking, but there are also times when they seem terribly dense. “These behaviors are not always consistent,” said ACE Kamar, a researcher at Microsoft.

Without a doubt, systems like GPT-4 were powerful, but it wasn’t clear if the text generated by these systems was the result, said Alison Gopnik, a psychology professor who is part of the AI ​​research group at the University of California, Berkeley. For something like human reasoning or common sense.

“When we see a system or a complex machine, we embody it; everyone does it — people who work in the field and people who aren’t,” Dr. Gopnik said. “But think of this as a constant comparison between AI and humans — kind of like a game show competition — Just not the right way to think about it.”

#Microsoft #showing #signs #human #thinking

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top