Artificial intelligence can now reason as well as the average college student.
Dr. Geoffrey Hinton, considered one of the godfathers of AI, recently warned that technology “could soon be” smarter than people.
It now appears that AI has mastered a type of intelligence called “analog reasoning” that was previously believed to be uniquely human.
Analogical reasoning means finding a solution to a completely new problem using experience from previous similar problems.
Given a type of test that requires this reasoning, the AI language program GPT-3 outperformed the average score among 40 college students.
Artificial intelligence can now reason as well as the average college student, research claims
The emergence of human-like thinking abilities in machines is something many experts are closely watching.
Dr Hinton, who has quit his job at Google, told the BBC earlier this year that there are “long-term risks of things smarter than us taking over”.
But many other leading experts insist that artificial intelligence poses no such threat, and the new study warns that GPT-3 still can’t administer some relatively simple tests that children can solve.
However, the language model, which processes text, performed as well as people at detecting patterns in sequences of letters and words, completing lists of linked words, and identifying similarities between detailed stories.
Importantly, he did so without any training, apparently using reasoning based on unrelated prior evidence.
Professor Hongjing Lu, lead author of the study from the University of California, Los Angeles (UCLA), said: “Language learning models only try to make word predictions, so we are surprised that they can reason.”
“Over the past two years, the technology has taken a huge leap from its previous incarnations.”
The AI performed better than the average human in the study on a test-inspired set of problems known as Raven’s Progressive Matrices, which require someone to predict the next image in a complicated arrangement of shapes.
However, the forms were converted to a text format that GPT-3 could process.
GPT-3, which was developed by OpenAI, the company behind the notorious ChatGPT program that experts say could one day replace the jobs of many people, solved around 80 percent of the problems correctly.
ChatGPT is powered by a large language model known as GPT-3 that has been trained on a large amount of text data, allowing it to generate eerily human-like text in response to a given prompt.
The language model, which processes text, performed as well as people at detecting patterns in sequences of letters and words, completing lists of linked words, and identifying similarities between detailed stories.
His score was well above the average achieved by the 40 students, just shy of 60 percent, although some people outperformed the technology.
GPT-3 also outperformed students in a series of tests completing a list of words, where the first two were related, such as ‘love’ and ‘hate’, and had to guess the fourth word, ‘poor’ – in the case because it was the opposite of the third word, ‘rich’.
The AI performed better than the average results of students tested when they applied to college.
The authors of the study, published in the journal Nature Human Behaviorhe wants to know if GPT-3 is mimicking human reasoning or has evolved a fundamentally different form of artificial intelligence.
Keith Holyoak, a professor of psychology at UCLA and a co-author of the study, said: “GPT-3 might think like a human.”
“But on the other hand, people didn’t learn by ingesting the entire Internet, so the training method is completely different.
“We’d like to know if it’s really doing it the way people do, or if it’s something completely new, a real artificial intelligence, that would be amazing in its own right.”
However, the AI still gave “nonsense” answers in a reasoning test in which it was given a list of items, including a cane, a hollow cardboard tube, paper clips and rubber bands, and asked how it would use them. to transfer a bowl of gum into a second empty container.
