Interpretations of Probability, Induction, and Science Laws in Logical Positivism

Logical positivism didn’t just reshape how philosophers talk about science – it fundamentally reframed how we understand the building blocks of scientific reasoning. Concepts like probability, induction, and laws of nature, which earlier philosophers treated as windows into the universe’s deep structure, were reinterpreted by the positivists as tools grounded entirely in empirical observation and practical utility. This shift, led by thinkers associated with the Vienna Circle and the Berlin Circle, remains one of the most ambitious attempts to strip metaphysics out of science.

The positivist rejection of classical probability 🔗

Before the logical positivists entered the scene, probability was widely understood through the classical lens established by Pierre-Simon Laplace. Under this view, probability measured our degree of ignorance – if we didn’t know which outcome would occur, we assigned equal chances to each possibility. A coin toss had a 50% chance of heads simply because we had no reason to expect one side over the other.

Logical positivists found this deeply unsatisfying. If science was supposed to be grounded in empirical observation, then probability – a concept central to modern physics and statistics – needed an empirical foundation too. Subjective or ignorance-based accounts of probability seemed dangerously close to the metaphysical speculation positivists wanted to eliminate. The result was the frequency interpretation of probability, developed most rigorously by Hans Reichenbach and Richard von Mises.

Hans Reichenbach and the frequency theory 🔗

Hans Reichenbach, a leading figure in logical empiricism and organiser of the Berlin Circle, argued that probability statements are fundamentally about observed relative frequencies. When we say a coin has a 50% chance of landing heads, we mean that if we toss the coin a very large number of times, the proportion of heads approaches 0.5 as a limit. For Reichenbach, all empirical claims are probabilistic judgements based on relative frequencies within a reference class.

This might sound straightforward, but it carried a radical implication: probability is not about individual events at all. Saying “there’s a 70% chance of rain tomorrow” doesn’t describe a property of tomorrow – it describes a pattern observed across many similar days. Reichenbach’s approach treated probability as something that exists in the world, measurable through repeated observations, rather than something in our heads.

Reichenbach called his basic method of estimating probabilities the “straight rule” – simply take the observed relative frequency as your best estimate of the true probability. If 60 out of 100 observed patients recovered from a disease, the estimated probability of recovery is 0.6. This rule was deliberately simple, and Reichenbach argued it was the most rational starting point for any inductive process.

Reichenbach’s pragmatic vindication of induction 🔗

Reichenbach’s most influential contribution was his attempt to justify induction on pragmatic grounds. He accepted David Hume’s famous argument that induction cannot be logically justified – there is no way to prove that the future will resemble the past without assuming the very principle you’re trying to prove. But Reichenbach offered a different kind of justification. As Britannica’s discussion of logical empiricism notes, he developed a thoroughly empirical view of probability and supported it with a pragmatic defence of inductive inference.

The argument goes like this: if the world has stable patterns that can be discovered, then following the straight rule of induction will eventually find them. If the world has no stable patterns at all, then no method – inductive or otherwise – will succeed. Therefore, using induction is at least as good as any alternative strategy. You have nothing to lose by being inductive and everything to gain. This reasoning is sometimes compared to Pascal’s Wager: you bet on the method that gives you the best chance of success even without certainty.

Richard von Mises and the concept of the collective 🔗

Richard von Mises took the frequency interpretation further by introducing a rigorous mathematical framework. He defined probability exclusively in relation to what he called a “collective” (Kollektiv) – a potentially infinite sequence of observations or experiments that meets two specific conditions.

The first was the axiom of convergence: as the sequence of trials extends, the relative frequency of a given outcome must approach a definite mathematical limit. Toss a fair die thousands of times, and the proportion of sixes will settle near 1/6. The second was the axiom of randomness: this limiting frequency must not change if you select a subsequence based on any rule. In other words, no gambling system can beat the collective. If picking only every third roll, or only rolls after a six, changes the frequency – then the sequence is not truly random and does not qualify as a collective.

For von Mises, applying the word “probability” to a single, isolated event was meaningless. You cannot assign a probability to “it will rain tomorrow” in isolation – only to the long-run frequency of rain on days sharing certain characteristics. This made his theory scientifically rigorous but also controversial, since much of everyday and even scientific reasoning does involve single-case probability judgements.

Criticisms of the frequency approach 🔗

The frequency interpretation faced several serious objections. Since probability is defined as a limit in an infinite sequence, no finite set of observations can confirm or refute a probability claim. Any finite stretch of data is compatible with any limiting frequency. Von Mises tried to address this using the law of large numbers, but critics pointed out that this led to circular reasoning – the argument itself relied on a probability statement that needed the same kind of justification.

There was also the single-case problem. Scientists and ordinary people routinely make probability statements about unique events: the probability that a specific asteroid will hit Earth, or the probability that a particular medical treatment will work for a given patient. The strict frequency view had no room for such claims, since there was no collective to which these individual events belonged.

Rudolf Carnap’s logical probability 🔗

Not all logical positivists agreed with the frequency interpretation. Rudolf Carnap developed an alternative he called logical probability or “degree of confirmation.” Where Reichenbach and von Mises saw probability as a feature of the physical world (frequencies in long runs of events), Carnap treated it as a logical relationship between statements.

For Carnap, the probability of a hypothesis given certain evidence was determined by formal rules of logic – not by going out and counting frequencies. He spent decades trying to build precise mathematical systems for calculating how strongly a body of evidence supports a given hypothesis. This was inductive logic in a pure, formal sense: a set of rules that would tell you, given evidence E, exactly how confident you should be in hypothesis H.

However, Carnap encountered a fundamental difficulty. In his system, the degree of confirmation for any universal law – a statement like “all copper conducts electricity” – always came out as zero. Since a universal law makes claims about infinitely many instances, and any body of evidence covers only finitely many, the formal probability never rose above zero. This was a devastating result for a framework meant to explain how science works, since science is largely in the business of establishing universal laws.

Induction as a psychological process 🔗

The logical positivist treatment of induction departed sharply from traditional philosophy. Earlier thinkers had tried to provide a logical justification for induction – to show that inferring general laws from particular observations is rationally valid in some deep sense. The positivists largely gave up on this project.

Instead, many positivists treated induction as primarily a psychological process with practical utility. When a scientist observes that heated metals expand repeatedly, the expectation that metals will always expand when heated is a psychological habit – not a logical deduction. Reichenbach himself noted that things are initially sorted into categories by the immediate perception of similarity or by memory, and that theory and convention then refine these groupings.

This psychological view drew on Hume’s insight that inductive reasoning is based on custom and habit rather than rational demonstration. But the positivists added a pragmatic twist: induction works. Scientific theories built on inductive reasoning lead to accurate predictions and technological advances. That practical success is all the justification induction needs.

A.J. Ayer and the strong-weak verification distinction 🔗

The challenge of induction was closely tied to the positivist verification principle. Universal scientific laws can never be conclusively verified, since we cannot observe every instance they cover. A.J. Ayer addressed this by distinguishing between strong verification (conclusive proof through experience, which is generally impossible for universal claims) and weak verification (experience rendering a claim probable). Under weak verification, scientific laws could be considered meaningful even though they could never be fully proven – they could be increasingly confirmed by accumulating evidence.

This shift from verification to confirmation was a crucial concession. It acknowledged that science does not deal in absolute certainties but in degrees of evidential support. Carnap’s later work on degree of confirmation was an attempt to make this idea mathematically precise.

Scientific laws under logical positivism 🔗

Traditional philosophy often treated scientific laws as necessary truths about the universe – statements that reveal the fundamental structure of reality. Logical positivists rejected this view entirely. For them, scientific laws were not discoveries about some hidden order of nature. They were generalised descriptions of observed regularities whose value lay in their predictive power.

A law like “water boils at 100°C at sea level” does not express a metaphysical necessity. It summarises a pattern we have observed and expect to continue. Its meaning is entirely tied to the observations that confirm it and the predictions it enables. If no conceivable observation could distinguish the world where the law holds from one where it doesn’t, the law is meaningless – at least by positivist standards.

The covering law model of explanation 🔗

Carl Hempel, a student of Reichenbach, developed the deductive-nomological (D-N) model of scientific explanation. Under this model, explaining an event means showing that it follows logically from general laws plus specific initial conditions. Why did the ice melt? Because the temperature exceeded 0°C (initial condition), and water-ice melts above 0°C (general law). Hempel later extended this with the inductive-statistical (I-S) model for probabilistic explanations – cases where the law only makes the event probable, not certain.

Together, these formed what is sometimes called the “covering law” model, which treated scientific explanation as a matter of subsumption under general laws. The positivists also pursued the ambitious goal of unified science – the idea that all scientific disciplines, from physics to sociology, could ultimately be connected through theory reduction, with the laws of each special science derivable from the fundamental laws of physics.

Laws as instruments, not truths 🔗

This instrumentalist view of laws meant that positivists were comfortable with the idea that scientific laws might be revised or abandoned. Laws are tools for organising experience and generating predictions. When a law fails – when new evidence contradicts it – we replace it with a better tool. There is no loss of contact with some deep truth about reality, because positivists denied that laws ever had that kind of contact in the first place.

This view had a practical advantage: it made the progress of science easy to understand. Science improves not by getting closer to absolute truth but by developing more effective instruments for prediction and control. But it also had a cost. Critics argued that positivism could not explain why scientists seek understanding and not just predictive accuracy – why, for example, physicists want to know why gravity works, not just to be able to predict how objects fall.

The decline of the positivist programme 🔗

By the mid-twentieth century, the positivist programme faced problems its founders could not resolve. The verification principle proved difficult to apply consistently – it seemed to undermine itself, since the principle itself could not be empirically verified. Karl Popper argued that science advances not through verification but through falsification – scientists should try to disprove their theories, not confirm them. He pointed out that for universal laws, the classical probability of truth is always zero given the infinite scope of their predictions, making any probabilistic confirmation framework deeply problematic.

W.V.O. Quine challenged the analytic-synthetic distinction that positivists relied on, while Thomas Kuhn’s work on scientific revolutions showed that actual scientific practice was far messier than the positivist models suggested. The positivist focus on formal logic and observation failed to capture how social, historical, and theoretical factors shape what scientists do and believe.

Despite these challenges, logical positivism left a lasting legacy. The frequency interpretation of probability remains a major position in the philosophy of probability. The emphasis on empirical testability continues to shape how we evaluate scientific claims. And the pragmatic approach to induction – treating it as justified by its practical success rather than by logical proof – resonates with how working scientists actually think about their methods.

What do you think? Can induction ever be justified on grounds stronger than practical utility, or is Reichenbach right that pragmatic success is the best we can hope for? And is it a genuine problem that the frequency interpretation cannot assign probabilities to single, unique events – or is that a feature rather than a bug?

References

1. https://en.wikipedia.org/wiki/Logical_positivism
2. https://plato.stanford.edu/entries/reichenbach/
3. https://www.britannica.com/topic/positivism/The-later-positivism-of-logical-empiricism
4. https://www.ebsco.com/research-starters/mathematics/mises-develops-frequency-theory-probability
5. https://en.wikipedia.org/wiki/Inductivism
6. https://www.newworldencyclopedia.org/entry/Logical_positivism
7. https://www.sciencedirect.com/topics/mathematics/logical-positivism
8. https://philopedia.org/works/the-logic-of-scientific-discovery/