Over the years, one of the so -called “AI’s Godfather” has been urging companies developing companies. Artificial intelligence product To prevent production and focus on safety, fear that progress can be quick and dangerous Change human autonomy.
Now, Yoshua Bengio is warning that AI machines are on a way to become “more smart” with their own. Self-conservation goal.
Bengio, a professor at Universit de Montreal, won the 2018 AM Turing Award – “Nobel Prize for Computing” – for his leading work in intensive learning with Geoffrey Hintan and Yan Lakeun.
They told The Wall Street Journal This week that AI products can develop the ability to cheat their users in search of their own goals – “Absolutely” like an emotional super computer in Stanley Cubic 2001: A space odissi.
“If we build machines that are smarter than us and have their own conservation goals, it is dangerous,” he said that WSJ. “It is like creating a competitor for humanity which is smarter than us.
“And they can impress people through persuasion, through dangers, through the manipulation of public opinion,” he said.
He offered an example of the AI machine supporting the manufacture of a virus that can create a new epidemic in a global terrorist attack.
“Talk with horrific events like extinction, and even less radical events that are still frightening like destroying our democracy, is that they are so bad that even though there was only one. [1 percent] The opportunity may be, it is not acceptable, ”he said.
Bengio told Wsj It is “a lot of people worried inside the companies,” and “being inside a company that is trying to advance the frontier, perhaps gives rise to an optimistic prejudice.”
They emphasized the need for “independent third parties” to review the internal “security” mechanisms of AI companies and “demand for companies to demand” AI systems they are deployed or used. “
“The same thing that governments should demand,” he said. “But market companies can drive to do the right thing if companies understand that many are unknown and potentially frightening risks.”
Otherwise, according to Bengio, the major risks within five to 10 years can be unavoidable. “But we must realize that this is just three years,” he told WSJ.
Still Donald appears to be the administration of Trump Dismissal Encouraging companies to design their products to reflect their own ideological agenda as part of a multiple dollar arms race for AI dominance.
In July, President Signed an executive order As part of a government-wide strategy for “achieving global dominance in artificial intelligence”.
This measure tries to ban the “Vok AI” in the government and directs the federal agencies to withdraw contracts from companies that do not align with the definition of the administration’s “truth-whane” and “conceptual neutrality”.
The order targets especially large language models, such as openiI chats, Amazon’s Alexa and similar models, which are trained on large versions of the text to do a wide range of language -related functions.
Federal agencies have been ordered to buy only large language models that provide “true reactions” that “do not deliberately encoded biased or ideological decisions.”
Shortly after taking office, Trump announced a $ 500 billion venture plan for AI Infrastructure, including OpenI, SoftBank and Oracle. He has also promoted plans from chipmaker Nvidia to create AI supercomputers in the United States and announced $ 92 billion from companies such as Blackstone, Google and Persian to AI and Energy Infrastructure.
Federal agencies are collecting equipment for “Domestic surveillance system He AI deploys products Critics expedite a technocratic label The totalitarian agenda.
The administration is also aiming for safety for a large amount of intellectual property informing AI products.
The head of the Copyright Office was fired A few days after the office released a long report about AI, which questioned the use of copyright content by technology.
Democratic rape, which Morel, said in a statement earlier this year, “It is definitely no coincidence” Trump fired the Copyright chief “when he denied the efforts of the rubber-stamp Alon Musk for the trowers of copyright to train the AI model,” Democratic Rape. Joe Morel said in a statement earlier this year. Mograel and other houses are calling the Inspector General to investigate the Democrat Congress library.