The GenAI-bubble will burst, but don’t give up on AI altogether

Generative AI such as ChatGPT may seem convincing with its beautiful images and eloquent texts. ‘But people keep promoting the belief that generative AI provides universal tools that are capable of much more,’ says Michael Biehl, Professor of Machine Learning. ‘Sooner or later, the genAI bubble will burst,’ he is certain. But that doesn’t mean all of AI should be thrown out with the bathwater.

FSE Science Newsroom | Text Charlotte Vlek

‘What generative AI (genAI) presents us with may look appealing: applications such as ChatGPT or Dall-E generate eloquent texts, beautiful images, and impressive videos,’ Biehl describes. ‘But there are several fundamental problems, such as factually incorrect texts, implausible images, and physically impossible videos. And these are intrinsic features of these systems, not bugs that will disappear with the next version.’

Biehl is concerned about the hype currently surrounding genAI: ‘Exaggerated praise predicts billion-dollar markets, while overwhelming fears range from the loss of countless jobs to the end of humanity.’ There are wild claims that genAI is capable of passing the Turing test (see text box), that it is capable of text comprehension, problem analysis, and logical reasoning.

Ethical problems

‘Sooner or later people will realize that generative AI is an illusion that they have been led to believe by Big Tech,’ Biehl says. ‘Right now, it’s a good thing that people are starting to become aware of the many ethical issues surrounding genAI.’ It is an impressive list, including the enormous waste of resources, the global exploitation of cheap labour to fix AI errors, the unauthorized use of copyrighted materials, the misuse of user data, and the uncritical amplification of biases in the training data. Last but not least, there is growing awareness and fear of possible manipulations by authoritarian regimes or by the big tech companies that release and control these systems.

Biehl: ‘People seem to think that all of artificial intelligence equals this over-hyped generative AI. But there is a longstanding, very broad field of research that  works on very different things besides producing eloquent series of words.’ Researchers at the UG, for example, are working on computer programs that can support doctors in analysing medical data, even when there is insufficient training data, or that can be used as sparring partners in medical diagnosis. The emphasis is often on collaboration between humans and machines, or on computer systems that can support their conclusions with arguments.

When the bubble bursts

‘Eventually, people will realize that even the largest language models are just that: language models,’ Biehl predicts. ‘They put together sentences, shamelessly imitate, or blatantly reproduce and remix pieces of text from their training data. When the genAI bubble bursts — and it will — the media and public opinion will turn against everything called AI.’ That is why Biehl appeals to his colleagues in the field: ‘Don’t fall for the hype, don’t fuel it any further, and don’t try to take advantage of it.’ Biehl even suggests avoiding the term “AI” as much as possible. ‘It will backfire.’