A study by the University of Oxford shows disturbing results on the subject of artificial intelligence. It is quite likely that the AI will be directed against the people.
Oxford – Engineering and technology are constantly changing and evolving. The most exciting is certainly the research on the subject of artificial intelligence (AI). In some areas, such as search engines or facial recognition in smartphones, AI is already being used successfully. The electric car manufacturer Tesla also wants to use the technology to enable autonomous driving in the future.
However, there is a certain degree of uncertainty when using artificial intelligence. People wonder if intelligent technology will not eventually turn against its creators. The eminent scientist Stephen Hawking said some time ago that AI is promising and offers great opportunities. At the same time, however, it also harbors great dangers. A study by the University of Oxford came to a similar conclusion.
Artificial intelligence could become a risk for humanity – according to the study
Stephen Hawking said at the time: “I fear that AI could completely replace humans.” A study by researchers from the University of Oxford has now come to the conclusion: “an existential catastrophe is not only possible, but also probable”. The scientists came up with this by using reward models as a basis, as reported by t3n.de. “Advanced artificial agents are interfering with the delivery of rewards,” summarizes Michael K. Cohen, the study’s lead author, summarizing the study’s gist. If that is the case, then it will have very bad consequences.
According to the scientists, an advanced AI could also see the successful prediction as a reward for a simple task, such as predicting the next number in a series of numbers. Now, when different models predict different rewards, they also identify different features of the world that might determine them, the authors explain.
It is feared that this could allow the AI to strategize to receive rewards in a more efficient way. Should she also act with the outside world, there are “infinite possibilities” for manipulation. According to the study, it is conceivable that she might trick people into secretly helping her, or that the AI installs countless, unnoticed helpers who are beyond human control. Last but not least, there is also the danger that there is an incentive to “eliminate” human ability or even to control or destroy the computer on which the main instance is running. The more success the AI had with its scams, the more rewards it would reap – and arouse its ambition.
Google Deepmind employees also worked on the study on artificial intelligence
Marcus Hutter, senior researcher at Google Deepmind, also collaborated on the Oxford University study. The Google company has been driving the development and programming of artificial intelligence for years. However, Google announced that Hutter only took part in the study as part of his position at the Austrian National University. Deepmind was not involved, but also invests great effort to help protect against malicious applications.
