Chat robots like ChatGPT can write and talk like humans. But that doesn’t mean they’re really intelligent. That’s what Yann LeCun, one of the most powerful men at Facebook’s parent company Meta, says. He is one of the leading scientists in the field of artificial intelligence.
Yann LeCun
AI boss at Meta
Open the people box. Close the people box
Yann LeCun is head of AI at Facebook parent company Meta. He is also a computer science professor at New York University. He is considered one of the leading scientists in the field of artificial intelligence: among other things, he taught AI systems to recognize text. He has received several awards for his achievements in the field of artificial intelligence. Among other things, he received the Turing Award, a kind of Nobel Prize in computer science, and the Global Swiss AI Award from the Mindfire Foundation.
SRF News: How intelligent is generative AI really?
Yann LeCun: Not intelligent at all. Chat robots only appear intelligent because they imitate our language well. But AI is completely helpless in the physical world. AI can pass a bar exam, but can’t learn to load a dishwasher. We humans can do that at the age of 10. We humans learn to drive in about 20 hours. The self-driving cars have been practicing for years and still can’t do it. But AI is changing how we understand human intelligence.
Legend: AI fails in the physical world – this is shown by the example of autonomous driving. imago images/Bihlmayerfotografie
Artificial intelligence can write books and create pictures. Is AI artistically intelligent, in the sense of creative?
Not really. AI writes books and poems and creates pictures. But this is just a new composition of what already exists, oriented backwards. Like a kind of cover of an existing song. That’s not really creative. It takes humans to create something completely new together with AI.
But I am convinced that artificial intelligence will become more intelligent overall and will overtake us humans.
Will machines ever become more intelligent than us humans?
In any case. Today, AI is only more intelligent than humans in very specific areas, for example when playing chess or the game of Go. But I am convinced that artificial intelligence will become more intelligent overall and will overtake us humans. But that will take decades.
How would you like to make machines smarter than people?
By making my research publicly available. At Meta I represent the open source approach. My research is accessible to everyone, including the programming code of our AI. This means that other companies and scientists can build on our work.
With the help of AI, there is a new type of disinformation and fake news.
Can you make an example?
In Switzerland, you might want an artificial intelligence that speaks Swiss German fluently. AI specialists on the West Coast of the USA may not be able to do that. Swiss engineers have to work with them. This is the only way AI can become truly intelligent in the long term. If we assume that artificial intelligence will become the interface to the Internet, then it must know all languages and be able to interpret different cultural values and peculiarities. This can only be achieved if the whole world drives developments forward together.
If everyone can collaborate on AI, how can you guarantee that your research results will not be misused?
This is indeed a problem. With the help of AI, there is a new type of disinformation and fake news. Words are put into politicians’ mouths that they have never actually said. These are so-called deepfakes. We must address this challenge.
Artificial intelligence will become a danger in the super election year of 2024
Open box Close box
The big tech companies like Meta are trying to calm things down. The new type of disinformation that is emerging because of AI can also be recognized with AI. And the disinformation circulating on social media can be moderated and taken down if necessary.
Whether this succeeds, however, is controversial. Because thanks to AI, there is likely to be a new flood of disinformation. Voting is taking place not only in Great Britain this year, but also in the USA and India, where millions of people will be called to the polls. The use of AI exacerbates the problem of deliberately spreading false information.
At the beginning of the year, hundreds of fake videos of British Prime Minister Rishi Sunak circulated on social media. For example, about an alleged financial scandal that never happened.
What are you doing specifically at Meta against this new type of disinformation and deepfakes?
This is nothing new for us. We have been moderating content on social media for years; for example, we remove hate speech or child abuse content from the internet. And creating disinformation, for example using AI, is not the same as spreading disinformation. Disinformation can also be produced without the help of artificial intelligence.
The interview was conducted by Pascal Lago.