Rumors that machines will soon take over the world’s most powerful positions are grossly exaggerated. In order to show that power still emanates from Washington and the people who rule there, Joe Biden spoke to the conscience of the masters of artificial intelligence: “What you are doing has enormous potential and enormous danger,” said the US President on Wednesday white house. A video shows Vice President Kamala Harris sitting in front of him with the most powerful entrepreneurs in the AI world: Sundar Pichai, head of Google parent Alphabet, Microsoft boss Satya Nadella, and Sam Altman of Open AI, the company behind the chatbot Chat GPT. Dario Amodei from the start-up Anthropic also came. Harris had summoned the managers to the meeting. According to the White House, leading AI developers from Open AI and Nvidia to Stability AI have agreedto have their AI systems publicly audited.
It’s about the new wave of artificial intelligence, so-called large language models, which can communicate amazingly well with humans. They learn from large data sets what statistical probabilities follow human language or art styles, and can therefore be creative in their own way, creating images and texts. At the moment, the marketing and Internet sectors in particular are relying on the new speech generation software. Even Biden, the White House said, had already tried out chat GPT.
However, because it is unclear what exactly the programs derive from their learning data sets from images and texts, the technology also triggers fears. Entrepreneurs like Elon Musk, AI pioneers like Geoffrey Hinton and progressive researchers like former Google employee Timnit Gebru warn. Some are concerned with scenarios in which AI gets out of control and subdues humanity, others with the fact that certain groups could be increasingly disadvantaged in everyday life.
The US government is now trying to draft rules for AI at an early stage. This time, politics should remain in control, not only wake up after headline-grabbing incidents, such as in the debate about political manipulation on Facebook after the 2016 election or the data protection incidents at Facebook parent Meta – whose boss Mark Zuckerberg should be happy that he this time did not belong to the group of those invited.
The White House wants to use the hacker scene to test the systems. At the big hacker conference Defcon, which takes place in Las Vegas in August like every year, the AI companies are supposed to release their software for launch: Experts are supposed to subject it to tough tests and find weak points. They are supposed to practically provoke the models: If they react with racism, calls for murder, plans to build bombs or even world domination, they would have to be improved. Thousands of AI experts will examine the models and determine whether they meet the government’s requirements for trustworthy AI. In addition, seven new national AI research centers are to be created with 140 million euros in tax money, which are intended to ensure that AI serves the common good.
Google and Microsoft are always at the table
From Germany, Christoph Schuhmann and his association Laion are building an alternative to the commercial programs of US corporations. He is skeptical about the hacker tests: “In the end, these experts will probably only have limited access again, so they can evaluate the outputs of the systems. But then the model behind it would remain a complete black box. Really transparent evaluation can only take place if both the weights of the model as well as the training data would be completely transparent and openly available to all of these thousands of experts.” The so-called weights are the settings in the programs that determine how precisely they learn.
After the meeting, Open AI’s Sam Altman said he and the other AI bosses shared the US government’s perspective on rules – “surprisingly”. However, the meeting also shows that the most powerful AI applications are developing structures that could be suspected of being a cartel. Open AI is considered the most advanced company – but Microsoft was also there, the main investor in Open AI. Alphabet was also represented twice: Not only did CEO Pichai come, Alphabet also put hundreds of millions of dollars into Anthropic, whose boss Amodei was also invited to the White House. Whatever the future of AI, Google and Microsoft are always at the table.