The Limits of Understanding: Do the Social Sciences Actually Explain Anything?
Do political science and economics give us genuine understanding, or an elaborate illusion of it? Here is the sceptical case, the best defence I know of against it, and where I think the argument actually lands.
Closed Systems Versus Open Realities
Start with chess. It is a closed, deterministic system with absolute and complete rules, and it is still not reducible to simple maxims. There is no unbreakable rule like "never sacrifice your queen," because a master understands the rare positions in which the sacrifice is exactly right. Learning to play well demands enormous quantities of data: machine learning needed millions of games to approximate mastery of a board with sixty-four squares.
If a closed system that restricted resists rigid rules and demands that much data, then human reality - many orders of magnitude more complex, with no fixed rules and no repeatable positions - presents something close to an intractable problem. That gap is where the doubts about political and economic "understanding" live.
The Sceptical Position
The sceptic argues that what is labelled understanding in the social sciences serves a psychological purpose rather than a scientific one.
- Skimpy data. Unlike physics or a digital game, social systems offer limited, noisy, unreliable historical data. We cannot run controlled repeatable experiments on macro-political events. The sample size for "what happens after a revolution" is measured in dozens, and no two cases share initial conditions.
- Description is not explanation. Statistical correlation is descriptive. Knowing that A tends to accompany B is not a grasp of the causal mechanism. Without knowing why a pattern holds, and precisely when it breaks, you do not understand it.
- The failure of minimal prediction. If a model cannot identify even a small fraction of the crucial exceptions - the equivalent of spotting when the queen sacrifice is correct - its predictive utility is close to nil. Notice how few economists called 2008 and how many explained it afterwards.
- Psychological utility over truth. Humans are deeply uncomfortable with chaos. Just as religions supplied narratives for the unexplained, the social sciences serve as secular equivalents, soothing existential anxiety with a comforting impression of order and authority.
- Active harm. The illusion is not free. Policies built on overconfident models cause severe unintended consequences, which makes the pretence of knowledge worse than an honest admission of ignorance.
The Defence
Defenders concede the limitations but argue that dismissing the fields entirely misreads what they claim to do.
| Defence | Core argument | Example |
|---|---|---|
| Institutional design | Individual behaviour is unpredictable, but structural rules shape outcomes reliably across large populations | Duverger's law: single-member plurality voting tends to produce two-party systems |
| Probabilistic utility | Absolute certainty is unavailable, but narrowing probabilities still lets societies mitigate risk | Central banking, inflation targeting, macroeconomic risk management |
| Conceptual frameworks | The fields supply shared vocabulary for diagnosing systemic failure | Opportunity cost, incentive structures, collective action problems, applied to pollution or resource depletion |
The strongest of these is the first. Duverger's law does not predict what any voter does, and it does not need to. It predicts an aggregate structural outcome from a rule, and it has held up across many countries and decades. That is not nothing, and it is not the same kind of claim as forecasting next year's growth rate.
Intellectual Company
The sceptical view has a distinguished pedigree.
Karl Popper attacked historicism - the belief that history obeys discoverable laws that let us predict the future course of societies. His argument was elegant: future knowledge changes human behaviour, and we cannot predict future knowledge, so we cannot predict the long-term future of a social system. He advocated piecemeal social engineering, small and reversible, over grand models.
Friedrich Hayek, in his 1974 Nobel lecture "The Pretence of Knowledge," warned against modelling complex social systems with the quantitative precision of physics. The data required to understand a market is decentralized, subjective, and constantly changing, which makes centralized mathematical planning both flawed and dangerously overconfident.
Nassim Nicholas Taleb argues that the social sciences lean on thin-tailed Gaussian models in fat-tailed domains, where rare high-impact events dominate outcomes. In those domains the illusion of prediction actively increases vulnerability, because it encourages trust in fragile models.
Where This Lands
The real question is whether an incomplete and badly flawed map beats no map at all.
The sceptic says an inaccurate map is worse than none, because it grants travellers false confidence and walks them off cliffs they would have approached carefully in the dark. The defender says continuous refinement, with the limitations openly acknowledged, is the only route to incremental improvement in governance.
My own view is that the answer depends entirely on the claim being made. The structural results are real knowledge: change the voting rule and the party system changes, and that survives replication. The forecasting results largely are not, and the discipline's worst failures come from lending the credibility earned by the first kind to the second. The honest position is not to abandon the social sciences but to hold them to the standard they claim: state the confidence interval, state the conditions under which the model breaks, and treat a prediction that cannot fail as a prediction that says nothing.