Slime mold, Trump, or ChatGPT: who is truly intelligent?

What is human intelligence and how does it relate to artificial intelligence (AI)? Associate Professor Sebastiaan Mathôt’s keynote speech focused precisely on this question. He delivered his talk during the first edition of Tammes Talks at House of Connections.

It is a well-known experiment: a slime mold can find its way through a maze to reach food. The mold appears to possess goal-oriented intelligence, but does this have anything to do with human intelligence? Mathôt posed this question at the beginning of his talk. He then showed a photo of an NVIDIA computer chip used for AI and one of Donald Trump. Do we consider AI intelligent? And what about the American president?

All three, the slime mold, the president, and the video chip, exhibit some form of intelligence, he continued. The use of the term "intelligence" is highly diffuse, which the keynote speaker emphasized with these examples. To clarify, Mathôt focused his lecture on how the field of psychology approaches the term. The central question was whether theories about human intelligence can help us better understand artificial intelligence. The associate professor also wanted to address whether AI influences human intelligence.

Olympic Games

In psychology, intelligence describes how people perform cognitive tasks. Mathôt emphasized that general intelligence predicts performance across a broad range of cognitive tasks. People who perform well on one cognitive task often excel in others as well. The associate professor compared this to the various events at the Olympic Games: people with very good physical condition will generally perform better in almost all events compared to those who are not fit. 

The correlation between cognitive tasks is clearly absent in slime molds but likely present in Donald Trump. But what about AI? Various studies have compared the performance of humans and AI on cognitive tasks. Compared to humans, AI has a much better vocabulary and factual knowledge, but it (so far) performs less well on certain visual pattern and reasoning tasks. What may be more important, however, is that research shows that AI models that perform well on one cognitive task often also excel in others. So, there is a correlation between performance on cognitive tasks in AI as well. Mathôt called this a non-human intelligence profile.

Outsourcing intelligence 

Finally, the associate professor reflected on the development of intelligence in humans and the potential influence of AI on this. The average IQ test score showed a "bizarre rise" worldwide in the last century, likely due to better nutrition and greater equality of opportunity. But in the past twenty years, the trend has reversed somewhat: scores have been gradually declining. Possible explanations include the influence of social media, COVID-19, or helicopter parenting, Mathôt suggested.

AI could also influence the development of our intelligence, as people increasingly rely on AI to perform cognitive tasks for them. There is already some research on the effects of AI use in education, showing that children who use artificial intelligence for studying perform equally or even worse on exams. If we let AI do our thinking for us, we are essentially outsourcing our intelligence, and this will undoubtedly have consequences for human intelligence, the associate professor concluded.

Tammes Talks was a new initiative at the JTS, which brought together communities to discuss digital trends and challenges. The programme included sessions on Digital Literacy & Inclusion, Digital Sovereignty & Data Autonomy, Autonomous Systems, and Applied AI.