Nobel Prize-winning scientist Geoffrey Hinton, known as the father of artificial intelligence for his work in artificial intelligence, issued a striking warning about the evolution of this technology. Hinton stated that the days of understanding what artificial intelligence is thinking are nearing an end, and that these systems may soon begin to think in a language incomprehensible to humans.
Artificial Intelligence Will Think in a Language Ununderstood
The advanced large language models used today use natural languages known to humans, such as English, to answer questions. This allows for the tracing of some of the reasoning processes. Which steps were taken or which resources were used can be analyzed when the system returns results.
However, questions such as why the system made certain choices or why it did not choose alternative paths cannot be definitively answered. Hinton believes that this uncertainty will deepen, and that artificial intelligence may eventually switch entirely to its own language.
This possibility has been demonstrated in some previous examples. Google’s translation systems developed an internal interlanguage years ago. Chatbots developed by Facebook in 2017 were discontinued during research because they developed an indecipherable language within themselves.
Those systems were far more limited than today’s models. Today, however, work is underway on far more advanced AI systems, with high learning capacities and constantly expanding their own decision-making processes.
Geoffrey Hinton argues that if these systems begin to think in their own language, it poses a significant risk to humanity. At this point, transparency regarding the systems’ mental processes could completely disappear. This could make predicting AI’s behavior and intervening when necessary nearly impossible.
{{user}} {{datetime}}
{{text}}