July 31, 2023

expert reaction to study looking at GPT-3 large language model and ability to reason by analogy

A study published in Nature Human Behaviour looks at emergent analogical reasoning in large language models.

Dr Andrew Rogoyski, Director of Innovation and Partnerships at the Surrey Institute for People-Centred AI, University of Surrey, said:

“This paper is fascinating and adds to the small but growing body of research that hints that there’s perhaps something going on in LLMs that we didn’t expect – the whole is being shown to be greater than the sum of the parts.

“Emergent behaviours of AI models is perhaps one of the most interesting developments in the field, essentially the system is able to solve problems that it wasn’t specifically designed to do.  This could be one possible path (there may be many) to AGI.

“If scale turns out to be the most important factor in achieving truly intelligent machines, this raises questions about how we control such systems, who has access to them, who should have access to them, who has the resources to build them, and so on.  The answers will define how AI impacts our future societies.

“If emergence in LLMs is becoming evident, we need to be cautious about building bigger AIs until we know what is actually happening and what the consequences might be.

“If this research really shows that we’re on the path to AGI, then there is going to be even further pressure on companies and governments to get there first.  Such technology would be transformational, helping us solve problems of climate change, disease and food production.  It could also be very dangerous.”

‘Emergent analogical reasoning in large language models’ by Taylor Webb et al. was published in Nature Human Behaviour at 16:00 UK time on Monday 31 July 2023.

DOI: 10.1038/s41562-023-01659-w

Declared interests

No reply to our request for DOIs was received.