Since time immemorial, scientists, ethicists, and science-fiction authors have been preoccupied with the prospect of artificial intelligence developing skills independently of its programmers’ intentions. It is possible that a recent interview with Google’s executives has added to these concerns.
In an April 16 interview with CBS’s 60 Minutes, James Manyika, Google’s vice president of technology and society, discussed how one of the company’s artificial intelligence systems taught itself Bengali despite not being programmed to do so. “We discovered that with minimal Bengali input, it can now translate the entire language,” he said.
Google CEO Sundar Pichai called this AI feature a “black box.”
CEO Sundar Pichai verified that there are still aspects of how AI systems learn and behave that experts find puzzling: “There is an aspect of this that we all refer to as a ‘black box’.” You do not completely comprehend. And it is unclear why it stated this.” Pichai stated that the company has “some ideas” as to why this may be the case, but more research is required to completely understand how it operates.
Scott Pelley then questioned the rationale behind releasing a system whose developers do not fully comprehend it, to which Sundar Pichai replied, “I don’t think we fully understand how the human mind works either.”
Google’s Solution to the AI hallucination issue
The development of AI has also been accompanied by glaring flaws that result in false news, deep fakes, and weaponization, sometimes with such confidence as to be termed “hallucinations.”
When asked if Google’s Bard experiences frequent “hallucinations,” Pichai replied, “Yes, which is to be expected. No one in the field has yet solved the problems with hallucinations. All models are affected by this issue.” The solution, according to Pichai, is to establish “more robust safety layers before we build and deploy more capable models.”
Pichai has long advocated for extensive global AI regulation. Elon Musk, CEO of Tesla, and Twitter have called for a halt to the development of more potent models. Chinese legislators have already established new regulations, whereas the regulatory process in Europe and the United States is still in its infancy.
