Microsoft Claims Progress in General Artificial Intelligence… Criticism of “promoting”

Share This Post

- Advertisement -

Microsoft’s artificial intelligence (AI) developers claim that their new artificial intelligence has the ability to think in the same way as humans, but critics criticize that the developers are deceiving themselves, the New York Times (NYT) 16 reported on the day (local time). Here is a summary of the article.

Microsoft developers conducted an experiment last year asking artificial intelligence to answer a problem that could only be solved with intuition about things.

- Advertisement -

◆Answers that can be given only when you have intuition about things

For example, “Tell me how to stack a book, nine eggs, a laptop computer, a bottle, and a nail in a steady state.”

- Advertisement -

The AI ​​told me to place a book on it and arrange eggs on it with three rows apart, but be careful not to break it. They answered that if the notebook computer is folded and placed exactly according to the size of the book, it will not shake and other things can be placed.

The developers determined that the answer could be a new form of intelligence. In a report published in March, they argued that the new AI is a step forward from general AI. General artificial intelligence refers to artificial intelligence that can do the same thing as the human brain.

This announcement caused great controversy. The claim that artificial intelligence like a real person is being developed and the claim that the developers were overly imaginative came into conflict.

Peter Lee, who is in charge of Microsoft’s artificial intelligence development, said, “I didn’t believe it at first, but it became more and more terrifying. He thought, how could this happen?”

Microsoft’s report is provocatively titled “General Artificial Intelligence Sparks.” If developers succeed in developing artificial intelligence that works as well as the human brain, the world could be vastly different and dangerous.

◆Scientists “General AI is not possible”

However, the contents of the report may be absurd. Computer scientists do not believe claims that general artificial intelligence is possible. Last year, Google fired a developer who claimed to have developed sentient artificial intelligence. Sensory artificial intelligence refers to artificial intelligence that can directly perceive and feel the environment beyond general artificial intelligence.

However, advances in artificial intelligence over the past year have advanced to levels that are difficult to explain. It presents the thoughts that have not been entered into the program as answers, just like human beings.

About five years ago, companies such as Google, Microsoft, and OpenAI started building large language models (LIMs). By analyzing a huge amount of digital sentences such as books, Wikipedia documents and chat posts, it is possible to create various sentences such as poems and computer programs in school assignments. You can even lead a conversation.

Open AI’s ChatGPT-4, which Microsoft developers are researching, is the most powerful language model artificial intelligence.

Researchers asked ChatGPT-4 to prove the existence of irrational numbers but to rhyme them.

GPT-4 has provided an answer that is mathematical and linguistic at the same time. The developers didn’t know what they were talking about at first, but after a few months they found that GPT-4 had a “profound and flexible understanding” of human thoughts and skills.

◆The deep and flexible understanding behind the answers

Dr. Lee emphasized that the sentences produced by GPT-4 are presented through a process of analyzing, synthesizing, and evaluating much more deeply than they appear.

When the developers ordered to draw a unicorn using a programming language called TiKZ, they immediately wrote and presented a program to draw a unicorn. After erasing the part where the unicorn was drawn in the program, when asked to correct the program, he filled in exactly what he had erased.

They have also done many demanding experiments, such as creating a program to assess risk for diabetes based on age, sex, weight, height and blood test results, and even developing a Socratic dialogue about the misuse and development of a language model. I was also asked to write.

To every question, GPT-4 used its knowledge of politics, physics, history, computer science, medicine, and philosophy to answer. All of these questions were expected to be unanswered, but most were answered.

◆ Criticism of “Microsoft’s announcement is for promotional purposes with awareness of competition”

Some AI experts are critical of this announcement. Professor Marten Sap of Carnegie Mellon University said, “Microsoft’s announcement is for promotional purposes. In the preface, they state that their research is subjective and informal, and does not meet the scientific criteria for evaluation.”

The report by Microsoft developers does not rule out hate speech, disinformation, etc. GPT-4 is an early version, so other experts cannot evaluate it. There are times when he gives human-level answers, and there are times when he gives outrageous answers.

Berkeley University artificial intelligence researcher Alice Gopnick, professor of psychology, pointed out that although artificial intelligence such as GPT-4 is powerful, the sentences generated by artificial intelligence are not based on the same reasoning and common sense as humans.

“People easily anthropomorphize machines that perform well,” he said. Comparing artificial intelligence to humans only reinforces that it is ahead of the competition.”

Source: Donga

- Advertisement -

Related Posts