Why IQ is a weak test for AI

Particle The last press viewOpenai General Director Sam Altman said that the EU IQ was observed in the last few years.

“It’s very rough, it looks like it – this is not scientifically accurate, this is just a vibe or moral response – we move a standard deviation of IQ every year,” he said.

Altman, as a benchmark for the AI ​​progress, is the first to use the IQ of a person’s intelligence. Those who influenced AI IQ tests on social media have given the models and gave results.

However, many experts say that the IQ model is a weak size – and is wrong.

“It can be very tempting to use the same measures we use for people, to describe opportunities or improvements, it is like to compare apples with orange with orange,” Sandra Wachter, technological and regulatory researcher at TechCrunch.

Altman equated Altman IQ with intelligence in his statements in Presser. Again IQ tests are relative – not objective – Measures definite Types of intelligence. Have somewhat consensus This IQ is a reasonable test of logic and abstract substantiation. But it does not measure practical Intelligence – know how things work and this is the best momentary image.

“IQ is a tool to measure human opportunities – based on human exploration,” Wachter noted. “But you can’t use the same size to describe the AI. A car is better than people. However, this does not mean that the car or submarine does not exceed human intelligence. This is another aspect of performance with human intelligence complicated. “

Superiority to the origin of the test in an IQ test Some historians Eugenika returns, a large-scientific theory that people can be improved by selective training, a test buyer must be a test buyer Strong business memory and knowledge of Western cultural norms. This, of course, invites the opportunity for the bias, so A psychologist called IQ tests “Ideological perverse mechanical models” of intelligence.

An IQ test that a model can in good way in an IQ test is more than the performance of the model, OS Keyes, the University of Eticatura, an ethical AI reads.

“(These) tests are quite easy for the game, if the tests are practically infinite memory and patience,” he said. “IQ tests are a very limited way to measure anything we know since the invention of the digital computer since then, cognition, confidence and intelligence.”

AI is most likely to have an unfair advantage in IQ tests, as well as taking into account the massive amount of memory and internal knowledge of the models. Often, models are taught to public web information and are full of questions from the Internet IQ tests.

“Tests inclined to repeat very similar samples – a very foolish way to raise the IQ, to carry out the IQ tests, the King’s Colle specialized in the AI, as a result of the AI, 1 million, unlike AI times I don’t get the pipeline into my brain with perfectly clearly and I can’t work without a noise or signal loss. “

As a result, IQ tests – are designed to evaluate the ability of cooked cooked, baked, baked, baked, as they are. They are not in line with a technology that solves the problems, in a very different way of people.

“A crow can use a tool to restore a dish from a box, but it does not mean that it can be written on Harvard.” “When I solve a math problem, my brain is very cold to read the words on the page, or in the room is very cold in the room. In other words, human brains are fighting more – any problem, any a problem, IQ tests or other way – and less assistant (more than AI) “

All this Need better AI testsCurrently, the EU’s EU Head AI scientist Heydid Khllaaf said in TechCrunch.

“In the calculation history, we have not clearly comparisons, because the calculation system can always complete the tasks outside the human ability,” he said. “This idea that we compared to direct systems against human abilities, this idea is a recent event, and covers the arguments of expanding and moving to evaluate the AI ​​systems.”