Hume's Problem of Induction and Scientific Knowledge

1. From Experience to the Future: The Central Question

All our factual reasoning that extends beyond immediate perception and memory, Hume argued, is founded on the relation of cause and effect. We believe the sun will rise tomorrow because we assume the same cosmic causes will produce the same effects they did yesterday. As a physicist, you ground your work in laws—like $F=ma$ or Schrödinger's equation—believing they will govern the outcome of your next experiment just as they have in countless past observations.

But what justifies this belief? What rational basis do we have for assuming that the future will resemble the past? This is the central question of the problem of induction. An inductive inference is any inference from observed instances to unobserved instances. Hume's challenge is to find the "chain of reasoning" that validates this leap.

To begin, let's watch a concise overview of Hume's argument. This lecture will re-familiarize you with Hume's Fork and then use it to frame the problem of induction.

2. The Two Horns of Hume's Dilemma

Hume's argument takes the form of a powerful dilemma. He contends that any argument purporting to justify induction must be of one of two types, and neither can succeed. As we established in the last lesson and the video recap, Hume divides all reasoning into two categories:

1. Demonstrative Reasoning: Concerns relations of ideas. The conclusions are certain and their falsehood implies a contradiction. This is the domain of logic and pure mathematics.
2. Probable Reasoning (or "moral reasoning"): Concerns matters of fact. The conclusions are based on experience and are never logically certain. This is the domain of science and everyday life.

Now, let's try to justify our inductive belief that the future will resemble the past using these two forms of reasoning.

Horn 1: Can Demonstrative Reasoning Justify Induction?

No. A demonstrative proof would have to show that the falsehood of induction is a logical contradiction. But it is not. We can, without any contradiction, conceive of a world where nature's laws suddenly change. We can imagine dropping a pen and it hovering in mid-air or flying to the ceiling. The statement "The future will not resemble the past" is perfectly conceivable. Since the contrary is possible, no purely logical, a priori argument can prove that the future must be like the past.

Horn 2: Can Probable Reasoning Justify Induction?

This seems more promising. We could argue: "Induction has worked in the past. Every time we've trusted past experience to predict the future, it has been successful. Therefore, induction will likely work in the future."

But what kind of argument is this? It's an inductive argument! We are justifying induction by using an inductive inference—assuming that because induction worked in the past, it will continue to work. This is blatantly circular. We are, as Hume puts it, "taking that for granted, which is the very point in question."

The following video offers a very clear explanation of this dilemma, using the powerful thought experiment of a "counter-inductivist" to hammer home the circularity problem.

So, we are trapped. We cannot justify induction with logic, and we cannot justify it with experience without begging the question.

3. The Uniformity Principle and a Formal Look

To formalize this a bit, we can say that all inductive inferences implicitly rely on a hidden premise, often called the Principle of the Uniformity of Nature (PUN): the future will resemble the past.

An inductive argument like:

1. Every observed instance of bread has been nourishing. (Premise)
2. Therefore, the next instance of bread will be nourishing. (Conclusion)

...only works if we assume:

1. Every observed instance of bread has been nourishing. (Premise)
2. The course of nature continues always uniformly the same. (Hidden Premise - PUN)
3. Therefore, the next instance of bread will be nourishing. (Conclusion)

Hume's problem can then be rephrased as: what justifies our belief in the PUN? And as we've seen, any argument for the PUN will either fail logically or be circular.

For a rigorous and detailed analysis of Hume's argument, the Stanford Encyclopedia of Philosophy entry is an excellent resource.

As the philosopher Peter Millican has meticulously shown, Hume’s argument is a single, integrated line of reasoning. The following diagram, adapted from Millican's analysis in his paper "Hume's Argument Concerning Induction", provides a clear map of the argument's flow in the Enquiry.

The Structure of Hume's Argument
(Adapted from P. Millican's structural diagram)

1. Foundation of Factual Reasoning: All reasoning about unobserved matters of fact is founded on cause and effect.
2. Foundation of Causal Reasoning: All reasoning about cause and effect is founded on experience.
3. Conclusion of Part 1: Therefore, all reasoning about unobserved matters of fact is founded on experience.
4. The Supposition of Experience: All arguments from experience proceed on the supposition that the future will be conformable to the past (The Uniformity Principle, or UP).
5. The Central Challenge: This supposition (the UP) must itself be justified by reasoning. Hume's argument now becomes a dilemma to show this is impossible.
   • Horn A (Demonstrative Reasoning):
     • The contrary of the UP (that the course of nature might change) is distinctly conceivable and implies no contradiction.
     • Therefore, there can be no demonstrative argument for the UP.
   • Horn B (Probable Reasoning):
     • All probable arguments are themselves founded on experience (from step 3).
     • All arguments from experience presuppose the UP (from step 4).
     • Therefore, any probable argument for the UP would be circular.
6. Final Conclusion: Since neither demonstrative nor probable reasoning can justify the UP, our belief in it is not founded on reasoning at all.

4. Hume's Solution and the Significance for Science

If reason does not compel us to believe in induction, what does? Hume's answer, which he calls his "skeptical solution," is custom or habit. After repeated observations of a conjunction, the mind develops a non-rational instinct to expect the same conjunction in the future.

This conclusion has staggering implications for science. If the principle of induction—the engine of scientific generalization and prediction—is merely an animal instinct, what separates science from superstition? Both seem to involve forming expectations based on past patterns.

Hume's answer is that while both are based on instinct, science applies this instinct consistently and systematically. The "wise man," he says, proportions his belief to the evidence. But the fundamental problem remains: the entire edifice of science appears to rest on a foundation that cannot be rationally secured.

Your own work in physics and computational modeling provides a sharp illustration. When you create a simulation of barchan dunes, you implement a set of rules (e.g., a cellular automaton or a Lattice Boltzmann model) and assume they remain constant throughout the simulation's "future." This assumption is an instance of the Uniformity Principle. You cannot prove the rules won't change at the next time step; you rely on the inductive assumption that they are stable.

The "No-Free-Lunch theorems" in machine learning are a modern, formal proof of a similar idea: no learning algorithm can be universally superior to any other across all possible problems. An algorithm's success depends on an implicit "inductive bias"—an assumption about the structure of the problem it's trying to solve. This is Hume's problem expressed in the language of computer science.

Let's explore some of these more advanced connections. The following reading discusses probabilistic approaches like Bayesianism, the No-Free-Lunch theorems, and how we might live with Hume's skeptical conclusion.

Conclusion

Hume's problem of induction presents a stark challenge to the rational credentials of empirical science. By analyzing the logic of our inferences about the world, he arrives at a powerful dilemma:

Key Takeaways:

• The Dilemma: The principle of induction (that the future will resemble the past) cannot be justified by demonstrative reasoning (as its contrary is conceivable) nor by probable reasoning (as this would be circular).
• Hume's "Skeptical Solution": Our belief in induction is not a product of reason but of custom or habit—a non-rational, psychological instinct.
• Significance for Science: Hume's argument calls into question the rational foundation of scientific laws and predictions. It raises the "demarcation problem" of what distinguishes science from superstition if both are based on non-rational pattern recognition.
• Modern Echoes: The problem persists today and is formalized in concepts like the No-Free-Lunch theorems in machine learning, which show that any successful learning algorithm must have an "inductive bias" or make assumptions about the data it will encounter.

Hume's argument was so powerful that it famously woke Immanuel Kant from his "dogmatic slumber" and inspired him to write his Critique of Pure Reason.

Preview of the Next Lesson:
We will turn to Immanuel Kant, who took Hume's challenge with the utmost seriousness. Kant's "Copernican Revolution" in philosophy is a monumental attempt to synthesize empiricism and rationalism and, in doing so, to save the possibility of scientific knowledge from Hume's devastating critique. We will explore how Kant tries to reground knowledge on a new and secure foundation.