August 20, 2024 – 09:46
The AI similar to ChatGPT was created by a Japanese research group with the help of the Universities of Oxford and British Columbia under the name AI Scientist. The goal of the experiment was for the program to try to find answers to theoretical scientific questions. In principle, Ahz AI Scientist was able to implement the entire life cycle of scientific research, starting with finding an interesting and new research area, finding innovative solutions for it, then conducting virtual experiments, drawing conclusions from them, and finally summarizing the findings. The text of the study states.
The AI scientist didn’t really come up with earth-shattering results (one critic of the study summarizing the research specifically described what the AI combined with the word “waste” — but what it produced while running was much more than that). Interestingly, for example, he screwed himself up when he wrote a piece of code to run his own program, sending the entire system into an infinite reboot loop, a time limit set by his own programmers. In response, he didn’t optimize the code he wrote to run faster, but reached into his own program to rewrite the time limit.
The Japanese researchers wrote in their study that their experiment pointed to the dangers of AI independence. Although “self-aware” AI is still a cliché from the Matrix and Terminator movies, experts say that even at its current level of sophistication, it could be dangerous if AI were able to write, run and modify code without the need for AI supervision and oversight, even by yourself. Without these constraints, they warn, if AI does not operate in isolation from the outside world, it could create malicious code or disrupt critical infrastructure online — unintentionally.
