On 8 February 2023, around USD 100 billion was wiped off the share price of Google's parent company Alphabet. That's because Google's AI chatbot 'Bard' made a mistake during its inaugural presentation – it attributed the first photo taken of a planet outside the solar system to the wrong telescope. The share price has long since recovered and has in fact more than doubled – that's partly because this is not a Google-specific problem. Similar, sometimes dreadful, errors can occur in all large language models (LLMs): "LLMs are prone to hallucinations – fabricating content and sometimes even sources," reveals Carl Benedikt Frey, Professor of AI & Work at the University of Oxford.
Frey believes this is likely to drastically reduce the opportunities for using AI in knowledge-based jobs. If efforts to rid AI of errors prove unsuccessful, it can only be used "where mistakes can be tolerated". An incorrect text can be deleted – but a wrong decision made by a self-driving car cannot be reversed.
At the GDI's 21st European Trend Day on 12th March 2025, Frey will explore the impact of automation and generative AI on the employment market. He does not expect AI to achieve a universal breakthrough: "AI – in its current form – is less likely to be deployed for higher-stakes applications, like driving, than in lower-risk activities, such as customer service or warehouses. Widespread use in the real world will require more robust and dependable AI solutions."
Carl Benedikt Frey
Professor of AI & Work
Oxford University
Great Britain
Website
Top performers and newcomers set to benefit
The potential applications in the digital world are far greater – because almost any job that can be done online can, in theory, also be performed by AI. One major exception is the creation of completely new pieces of work: "Generative AI systems are good at remixing and rehashing existing music or text. While we don’t know how Marcel Duchamp came up with the idea of making his ready-made 'Fountain' from a porcelain urinal, it was certainly not by analysing a dataset of impressionist paintings."
While AI can certainly create variations, sequels and iterations of existing material, it can't come up with anything new without a model on which to base the work – that's a feat only a few creative human minds can achieve. In his view, this may mean highly talented people within the creative industries could become irreplaceable superstars, while those in the middle of the pack could come under pressure from AI.
Mediocre performers face a threat from below too. Carl Benedikt Frey uses the example of ChatGPT to illustrate the point: "ChatGPT has been shown to improve the productivity of writers, particularly less skilled ones, in completing tasks. ChatGPT will not replace journalists, but it is making these tasks easier for industry newcomers, creating fiercer competition."
This can put pressure on wage levels. Frey sees a parallel here to the situation in the taxi industry following the emergence of Uber: a study carried out by his Oxford Internet Institute and Lund University found that after the arrival of Uber in a city, hourly taxi wages there fell by around 10%. However, the quality of the service did not suffer, as GPS technology also guided less experienced drivers to their destinations without any problems. Frey believes we may see a similar development in writing with the help of ChatGPT: "Might LLMs prove to be a GPS for language?"
Join us at the 21st European Trend Day as Carl Benedikt Frey explores how generative AI is reshaping winners and losers in the job market and highlights the opportunities and risks that come with it.