David Hume, Skepticism, and AI

Detail of David Hume's eyes and a bit of what I assume is a turban. The original painting is by Allan Ramsay and was painted in 1758.

People on LinkedIn enjoyed my post on AI for quiz generation (Building Reading Quizzes with AI). Here’s another installment in a different direction but still comparing output between OpenAI‘s ChatGPT and Anthropic‘s Claude. The results underscore the value of getting answers from more than one AI. I think they also are a good example of AI-as-tutor, which is one of the better roles AI can play in education.

So, here’s the setup. I’ve been reading David Hume‘s “An Enquiry Concerning Human Understanding.” I like Hume, but his extreme skepticism about causation was making my head hurt. I kept wanting to mount a defense against it, but I wasn’t gaining any ground. AI chatbots can be good tutors, so I formulated this prompt:

Was David Hume, in “An Enquiry into Human Understanding” wrong about causation? With modern instruments and a modern understanding of physics, isn’t the level of skepticism about, to take his example, the predictability of one billiard ball transferring motion and a predictable trajectory to another unwarranted?

The prompt I used with Claude and ChatGPT

I gave Claude first crack at it. Here was the reply:

You raise some thoughtful points about David Hume’s views on causation. Here are a few perspectives to consider:

  • Hume was writing in the 18th century before modern scientific instruments and understanding of physics. His skepticism about our ability to definitively prove causal connections between events was partly influenced by the more limited empirical knowledge of his time.
  • However, Hume’s main argument was that causation itself cannot be directly observed – we only see sequences of events, not the actual “necessity” that connects cause and effect. Even with today’s advanced science, we cannot perceive causation directly. We can only measure correlations and regularities between events.
  • So from a philosophical perspective, Hume’s basic critique about the limits of human knowledge on causality still holds some validity. No amount of scientific progress allows us to directly perceive necessary connections between events. We can only build probabilistic models of causation based on observation and inference.
  • However, modern science does give us very high confidence in many causal relationships. While philosophers may still debate the absolute provability of causation, practical science and engineering relies heavily on causal models that are extremely predictively reliable.
  • So in everyday practical terms, Hume’s skepticism goes too far – we can be highly confident in causal relationships like billiard balls transmitting motion. But at a philosophical level, Hume’s basic argument retains some merit – causation itself remains a metaphysical assumption rather than something we can directly prove.

In summary, I would argue Hume underestimated the ability of science to establish robust causal models, but his core philosophical doubts about the limits of human knowledge remain meaningful even today. The billiard ball example illustrates how science gives us high confidence in causality, but not absolute proof.

Claude’s reply to the prompt (October 13, 2023)

Naturally, I wanted to see what ChatGPT had to say in response to the same prompt:

David Hume’s position in “An Enquiry Concerning Human Understanding” is rooted in his skepticism towards the notion of necessary connections in causation. He argues that our belief in the uniformity of nature, which underlies our ability to make predictions based on past experiences, is not grounded in reason but rather in custom or habit.

Hume argues that when we see one billiard ball strike another and observe the second ball move, we do not have any rational basis for claiming that the first ball caused the motion of the second. Instead, we have a habitual expectation based on our repeated observations of similar events in the past.

According to Hume, our minds naturally associate events that regularly occur together, but this association is not a result of reason or logical necessity. Therefore, he contends that we cannot use reason alone to predict future events based on past experiences.

Many philosophers have engaged with Hume’s ideas, and there are various interpretations and responses to his arguments. Some may argue that Hume’s skepticism about causation is too extreme and that there are rational grounds for making predictions based on observed regularities in nature. Others may agree with Hume that there are limits to what reason alone can tell us about the natural world.

Ultimately, whether one believes Hume is wrong or not depends on their philosophical stance and interpretation of his arguments. There is ongoing debate among philosophers about the nature of causation and the limits of human understanding, and different perspectives continue to be discussed and refined.

ChatGPT’s reply to the prompt (October 13, 2023)


I found Claude’s reply more useful, thought provoking, and nuanced. In five bullets, Claude does a good job of showing why Hume was right while also giving some real consideration to why a person might think his argument goes too far. ChatGPT starts off strong and gives a very nice summary of the relevant areas of concern with remarkable brevity, but, in the final two paragraphs, descends into it’s typical sort of wishy-washy different-people-think-different-things conclusion.


If you’re an educator or an instructional designer, the differences in the responses create an opportunity for assignments where students must use their own writing and critical thinking skills. You might, for example, require students to write an brief argumentative essay or discussion board post where they stake out a position defending one response over the other, with direct quotations and proper citations. You might have them–in the context of that assignment or in a different discussion board post–find three strengths and three weaknesses with each AI-generated response. And, of course, it doesn’t have to be Hume. The same comparison assignment can work with any text. This sort of assignment gets students using AI but also teaches them to be critical about the texts that AIs generate.


The featured image of this post is a detail from a painting of Hume by Allan Ramsay (1713-1784). You can learn more about it at National Galleries of Scotland.