The probability of humanity being exterminated by artificial intelligence is known: what scientists say
There is a very low probability of a scenario in which the development of artificial intelligence will eventually lead to the extinction of humanity. Scientists believe that the probability of such events is no higher than 5%.
This is evidenced by a study based on a survey of 2,700 artificial intelligence researchers. According to NewScientist, scientists believe that there are wide disagreements and uncertainty about the risks of AI for humanity.
The study is the largest survey of its kind to date. Participants were asked to share their opinions on the possible timing of future technological milestones of AI, as well as on the positive or negative consequences of these achievements for society.
As it turned out, almost 58% of respondents believe that there is a 5% probability that the development of AI will lead to the extinction of humanity or other extremely bad consequences.
"This is an important signal that the majority of AI researchers do not think it is unbelievable that advanced AI will destroy humanity," said Katja Grace, the author of the study from the Machine Intelligence Research Institute in California.
At the same time, Émile Torres from Case Western Reserve University in Ohio warns that expert surveys on AI have already proved to be wrong. Thus, he recalled that in 2012, a study showed that in the long run, the predictions of AI experts were no more accurate than the opinions of ordinary people.
The authors of the article also note that AI researchers, despite their status and understanding of the situation in the industry, are not experts in the question of how AI will develop.
However, a similar survey in 2022 showed that many researchers still predicted the development of the industry relatively accurately. In particular, they predicted how the situation with the debut of ChatGPT and other chatbots based on large language models would develop.
Thus, the researchers in the new survey stated that there is a 50% (or even higher) probability that AI systems will successfully solve most of the 39 test tasks, including writing new songs or coding an entire payment processing website from scratch. However, they will still have some trouble with tasks such as physically installing electrical wiring in a new home or solving long-standing math puzzles.
Researchers also give a 50% chance that by 2047, AI will be able to outperform humans in any task. They also believe that by 2116, there is a 50% chance that humanoid robots will become fully autonomous.
At the same time, Torres suggests that such estimates may be too optimistic and does not rule out that "the AI field will go through another winter," referring to the exhaustion of funding and corporate interest similar to those observed in the 1970s and 80s.
The researchers also believe that there is a 70% probability that AI will be used to create dipfakes, as well as to manipulate public opinion, develop weapons, introduce authoritarian control over the population and deepen economic inequality.
Earlier, OBOZ.UA reported that artificial intelligence was taught to predict the death of people. It turned out to be right in 78% of cases.