What is intelligence anyway?

Is AI really intelligence at all?

We’re obsessed with intelligence. We have game shows like ‘Who Wants to be a Millionaire’ and ‘The Weakest Link’ that reward people for answering questions. We have societies like Mensa for people who score well on certain tests. We even design puzzles for animals whether intentionally in laboratories or unintentionally as in the case of this squirrel outsmarting a device designed to keep it from accessing a bird feeder. The latest frontier in this debate is the discussion of artificial intelligence and human intelligence.

To say perspectives vary on this topic would be putting it mildly. Writing in TechTarget, Michael Bennett talks about the distinction between artificial intelligence and human intelligence. Bennett argues that human intelligence is superior for multiple reasons such as the ability of humans to learn from very few examples compared to the vast amounts of data required by AI. Science fiction writer Ted Chiang goes further by stating that many applications of AI could better be described as applied statistics. Chiang argues that while AI can do very impressive things its behaviour relies on observing patterns in data and replicating these patterns.

There’s a broader question here. Intelligence is only one of the many human capabilities AI creators are trying to replicate. Consider for example an app called Xiaolce that acts as an AI companion for people who are struggling and need emotional support. A France24 article gives a story of a woman who had recently experienced a breakup and was using the app for emotional support. Apps like these add multiple dimensions to the debate on the capabilities of AI versus humans. Debates on intelligence tend to whether AI can learn as quickly as humans, whether it’s as flexible in performing tasks etc. The discussion generally considers how humans and AI behave in isolation. Debates on interpersonal traits like empathy focus on the way in which humans form relationships and whether AI can ever meet the criteria. This discussion is inherently relational. To determine whether an AI produces a reasonable approximation of empathy, emotional understanding etc. we must talk about how it interacts with humans and whether its responses meet standards that we would expect from other humans.

The term ‘applied statistics’ may become more and more relevant as we try to build some machines that are intelligent, others that are empathetic and so on. Rather than claiming to create an artificial equivalent of a human trait we might more accurately say we are applying statistics to replicate examples of that trait. Although I’ll admit that sounds a lot less appealing.?