Here’s a summary of ‘Why Intelligence Fails’ by the political scientist Robert Jervis. It’s a book analysing two cases where the U.S. intelligence community ‘failed’: being slow to foresee the 1979 Iranian Revolution, and the overconfident and false assessment that Saddam Hussein had weapons of mass destruction in 2003.
I’m interested in summarising more books that contain valuable insights but are outside the typical EA canon. If you’d like more of this or have a book to suggest, let me know.
- Good intelligence generally requires the relevant agency and country office to prioritise the topic and direct scarce resources to it. Good intelligence in a foreign country requires a dedicated diplomatic and covert collection corps with language skills and contextual knowledge.
- Intelligence analysis can be deficient in critical review, external expertise, and social-scientific methodology. Access to classified information only generates useful insight for some phenomena.
- Priors can be critical in determining interpretation within intelligence, and they can often go unchallenged.
- Political pressure can have a significant effect on analysis, but is hard to pin down.
- If the justification of an intelligence conclusion is unpublished, you can still interrogate it by asking:
- whether the topic would have been given sufficient priority and resources by the relevant intelligence organisation
- whether classified information, if available, would be likely to yield insight
- whether pre-existing beliefs are likely to bias analysis
- whether political pressures could significantly affect analysis
- Some correctives to intelligence failures which may be useful to EA:
- demand sharp, explicit, and well-tracked predictions
- demand early warning indicators, and notice when beliefs can only be disproven at a late stage
- consider negative indicators - 'dogs that don't bark', i.e. things that the view implies should not happen
- use critical engagement by peers and external experts, especially by challenging fundamental beliefs that influence what seems plausible and provide alternative hypotheses and interpretations
- use red-teams, pre-mortems, and post-mortems.
- Overall, I’ve found the book to somewhat demystify intelligence analysis. You should contextualise a piece of analysis with respect to the psychology and resources involved, including whether classified information would be of significant benefit. I have become more sceptical of intelligence, but the methodology of focusing on two known failures - selecting on the dependent variable - mean that I hesitate to become too pessimistic about intelligence as a whole and as it functions today.
Why it’s relevant to EA
The most direct application of this topic is to the improvement of institutional decision-making, but there is value for any cause area that depends on conducting or interpreting analysis of state and non-state adversaries, such as in biosecurity, nuclear war, or great power conflict.
This topic may also contribute to the reader's sense of when and how much one should defer to the outputs of intelligence communities. Deference is motivated by their access to classified information and presumed analytic capability. However, Tetlock’s ‘Expert Political Judgment’ cast doubt on the value of classified information for improving prediction compared to generalist members of the public.
Finally, assessments of the IC’s epistemic practices might offer lessons for how an intellectual community should grapple with information hazards, both intellectually and socially. More broadly, the IC is an example of a group pursuing complex, decision-relevant analysis in a high-uncertainty environment. Their successes and failures may be instructive to similar EA efforts.
I think I capture around 80% of the key content in this summary, but if you wanted to read it yourself it’s a short book and much of it is interesting. Reading chapters 1 and 4 would give the best sense of the implications and Jervis’ overall view. Chapter 3 on Iraq WMDs is short and useful to exemplify Jervis’ points, as well as its content being especially relevant to WMD concerns. Chapter 2 on the Iranian revolution is long and dry and I wouldn’t recommend reading it unless you were interested in the specific history.
Chapter 1 - Adventures in Intelligence
Jervis introduces the importance of intelligence failures by noting that conflicts are often partly caused by failures to predict what others will do. He notes it is hard to rigorously study intelligence because successes and failures are often secret. However, he feels confident in claiming that:
- intelligence failures are not rare
- intelligence failures are inevitable
- we have no reason to think intelligence failures are becoming less frequent.
Two important kinds of intelligence failure
He defines two key kinds of intelligence failure:
- A failure of accuracy, in which reality doesn’t turn out to match the assessment.
These failures of accuracy can be reduced but are inevitable. Why? Because “Intelligence is a game between hiders and finders and the former usually have the easier job”. In addition to this structural disadvantage, intentions are hidden and fast-changing, and deception is easy. And on top of all this adversaries are strategically aware of the ease and utility of deception and can exploit it.
- A failure to provide what is needed.
This can be historical, such as a past failure to not have developed the appropriate intelligence-gathering means. As an example for the Iraq WMD case, Jervis suggests the U.S. could have prioritised recruitment of Iraqi sources in the 1990s. It can also be a present failure to optimally use the available information. This can be a problem of judgement rather than accuracy - e.g. having all the relevant information, but not drawing the right conclusions or influencing action in the right way.
Two core theses
There are two overarching theses of the book. The first is the importance of ‘political psychology’ - that is, understanding the individuals who make up the critical organisations, and accounting for their imperfect access and analysis of information, their incentives, and organisational culture.
The second thesis is that the methods of social science - akin to ‘critical thinking’ - are powerful and under-used in these cases by both the IC and their critics. His specific charges are worth quoting at length:
“they do not formulate testable hypotheses and so often rely on beliefs that cannot be falsified, leave crucial assumptions unexplicated and unexamined, fail to ask what evidence should be present if their arguments are correct, ignore the diagnostic value of absent evidence, and fail to employ the comparative method and so assert causation without looking at instances in which the supposed causal factor was absent as well as cases in which it is present. All too often, intelligence and critics rely on intuitive ways of thinking and rhetorical forms of exposition.”
Chapter 2 - Failing to see that the Shah was in Danger
Jervis’ work on Iran derives from a de-classified report he wrote as a scholar in residence at the CIA-housed National Foreign Intelligence Assessment Center in 1979. The book reproduces his original report with a contemporary introduction, and adds a series of review comments made by CIA analysts.
Below I’ll summarise the elements Jervis criticises and possible correctives he proposes. To guide those interested in reading specific subtopics within the Iran case, Jervis covers:
- the White Revolution and the Shah’s attempts to liberalise
- the Shah’s willingness to use force
- the unity or division of the opposition forces
- religious opposition
- the Shah’s position and mood and how it was perceived
- the intensity of anti-Shah feeling
- Iran’s domestic economic situation
- the (lack of) contacts between U.S. intelligence and various factions in Iran
- biases by ignoring U.S. policy and its effects
Jervis' reasons for the intelligence failure in Iran
Jervis suggests a number of reasons why intelligence collection was poor in the years leading up to the Iranian revolution.
- Good information requires a diplomatic and covert collection corps with good language skills - there were almost no Americans with good Farsi. This meant they were isolated, and also that their few connections were with a secular, middle-class part of the opposition, which was minor.
- Iranian politics was not a priority - intercepting Soviet comms, communism, and anti-U.S. terrorism were.
- There was a paucity of resources dedicated to Iran, with only two political and two economic analysts in the CIA dedicated to Iran and no Iran experts in other bodies. The CIA station in Tehran was small with little political intelligence.
- There was a paucity of critical analysis by superiors or peers, and few even read the critical intelligence reports. There was no peer review, and the key publication vehicle was a daily brief, which incentivised short, up-to-date, headline-like articles, while deep analysis was not incentivised.
- There was little contact with external experts, especially from academia.
- Analysts wrote “more like journalists than social scientists” - they reported facts and made a coherent story, but didn’t try to use explicit methods or frameworks, draw generalisations, pose alternative hypotheses, or go beyond information directly implied by their sources.
- There were pre-existing beliefs that determined much of the interpretation of events, and these beliefs went unchallenged. What distinguished optimists from pessimists about the Shah were “general beliefs about Iran which long predated the recent protests.” These pessimists were mainly in the State department and outside the government, so there was little analytic exchange between the camps. Other specific priors preventing accurate analysis were:
- that religion was not important
- that U.S.-backed regimes are only threatened from the left
- that the Shah would crack down if things got bad.
- The revolution was a political development. The intelligence community does not have a comparative advantage in political analysis. The intelligence community’s main asset - access to secrets - was not very relevant or helpful to predicting or understanding the dynamics of the revolution, or revolutions in general.
- The Iranian context, especially that of Shi’ism, was alien to most analysts and so was poorly modelled and under-appreciated.
While he’s aware that his approach selects on the dependent variable, he speculates on efforts in other countries:
“Like many people who did not know the government from the inside, I had assumed extensive coverage of every country. In fact, this was out of reach and remains so...
My fear is that what I saw was fairly typical. I would like to believe that things have greatly improved in the subsequent years, but in fact the increased burden on embassies to carry out other chores, especially escorting congressional and business delegations, has led to a decrease in the amount of political reporting.”
It’s difficult to contemporaneously assess the U.S. intelligence gathering capability on a particular issue in a particular context. But I think Jervis’ analysis suggests some useful questions to ask when considering the views of intelligence officials or organisations even if their justification remains unpublished:
- Does the relevant intelligence collecting body have sufficient resources, attention, and priorities to enable quality analysis of the specific issue?
- Is the issue amenable to insight from classified sources?
- Are pre-existing beliefs likely to over-determine analysis?
- Are political pressures likely to significantly sway analysis?
Correctives to causes of the Iran failure
Jervis also suggests some correctives, some of which concur with EA orthodoxy, and some of which may be useful additions:
- judgements should require sharp and explicit predictions, which should be tracked and compared to subsequent events.
- judgements should require early warning indicators, or at least it should be noted when a belief can only be disproven at a late stage.
- judgements should also suggest negative indicators - meaning events that should not occur if the view is correct. This is sometimes called 'dogs that don't bark' - referring to a Sherlock Holmes story where Holmes deduced that because the guard dog didn't bark during the crime, the murderer must have been known by the dog.
- jargon should be carefully controlled: “All those involved in the production process - managers, analysts, and editors - need constantly to keep in mind the prospective readership of the document they are working on and to question whether a given term will give a reader a correct, and not just a brief, understanding of the phenomenon it purports to characterise.”
- Critical engagement is needed, which can be provided by peer review or by institutional competition.
- The pool of analysts should be drawn from a variety of backgrounds and kept in communication with other organisations and external experts, in order to increase sensitivity to information that contradicts the majority view.
Chapter 3 - The Iraq WMD Intelligence Failure
A broad consensus on the Iraq WMDs case blames the U.S. intelligence community for egregious errors. Jervis’ overall view agrees that:
- there were significant and avoidable errors,
- that there was far too much confidence,
- that alternatives were not sufficiently considered
- that there was insufficient imagination as to other interpretations of the available evidence (e.g. that Saddam was bluffing).
However, Jervis contends that a better process would have reduced the confidence of conclusions but not changed their content. He also argues that better intelligence would not have led to improved policy, as the decision to invade Iraq was essentially already made.
Popular explanations of the failure
Jervis argues against a number of popular explanations for why analysis was so wrong about WMDs in Iraq:
- groupthink - Jervis believes the number and diversity of analysts was too large and dynamic over time for simple groupthink to be a fitting explanation
- excessive consensus - Jervis thinks this is somewhat true, but points to the fact that there were still vigorous disagreements between agencies. Also, there was a substantial amount of background information that was agreed upon by all agencies, and this agreement was sensible, practical, and correct. So this explanation needs to distinguish which parts had excessive consensus vs. other parts that had true and useful consensus.
- failure to challenge assumptions - Jervis agrees this is true regarding the belief that Saddam’s policy was “consistent, coherent, and unchanging.” However, he suggests this explanation is too simplistic. Making assumptions is inevitable to any substantial analysis, and they are often true and useful. Furthermore, many problematic assumptions were also unchallenged by outside observers. So this explanation needs to point to which assumptions should have been challenged and why.
- analysis was under strong political pressure - Jervis agrees that policymakers cherry-picked and misrepresented intelligence, and that politicisation likely constrained critical analysis and encouraged over-confidence. However, he argues that no analysts he interviewed confidentially claimed pressure towards a particular conclusion or alteration of conclusions by superiors. Furthermore, Clinton-era intelligence estimates were similar, and the most politically sensitive intelligence body, the Department of State, was one of the most sceptical of the WMD claim. And other countries’ intelligence estimates believed Iraq had WMDs, even in countries that opposed the war.
Jervis’ debunking of politicisation as a cause appears to be the most controversial claim in the book. In the New York Review of Books, Powers argues that the Bush administration’s pressure on the intelligence community was strong and consequential (this claim is also made by Farideh Farhi’s 2011 review). He notes Jami Miscik’s threat to resign as deputy director of intelligence of the CIA unless the White House ceased pressuring the CIA to come up with a link between Iraq and 9/11. He also notes that Cheney visited the CIA headquarters 8 times to argue about Iraq. Powers claims that the focus on Iraq at the time did not arise exogenously due to some development, but rather because of White House pressure after 9/11.
A letter from a former National Intelligence Officer to the Powers review agrees on there being strong politicisation, and describes elaborate and flagrant motivated reasoning in the national intelligence process justifying the Iraq-WMDs claim.
Powers suggests Jervis “demands a strangely high standard of proof of White House meddling”, and repeats Jervis’ quotation of a CIA analyst:
“Politicization is like fog. Though you cannot hold it in your hands, or nail it to a wall, it does exist, it is real, and it does affect people’s behavior.”
I think the difficulty in assessing the influence of power on analysis suggests EA should be thoughtful about how the world-views of funders and leading figures may influence apparently independent convergence of the community to key claims.
Jervis' list of causes of failure
In opposition to the consensus, Jervis argues the actual causes of intelligence failure in the Iraq case were:
- overlearning - following the 1991 Gulf War the U.S. I.C. realised it had greatly underestimated Iraq’s WMD activities.
- denial and deception - Iraq had tried to conceal the program in the past, and the I.C. took this as a given rather than a hypothesis - meaning that failing to find evidence was counted as evidence of denial/deception, rather than evidence against a WMD program.
- shortfalls of human intelligence - few individuals that were misleading and over-relied on, and context on source reliability didn’t survive transmission between agencies and up command chains. Worst case is Curveball, whose information only came through German intelligence, with attendant language difficulties and unreliability.
- poorly structured inter-agency debate, especially existing grievances between CIA and DOE. Specifically there were poor communication channels, tribalistic motives, biassed overseers, misguided expertise, and premature cognitive closure.
- failed integration of technical and political analyses - e.g. CIA WINPAC assessed weapons, but didn’t take into account Iraqi geopolitics and political psychology.
- failure of imagination, with an inability to appreciate that opponent plans and beliefs were incoherent and strange.
- cognitive biases including premature cognitive closure, confirmation bias, aversion to considering negative evidence, and over-reliance on plausibility and past knowledge than actual evidence and clear argument.
However, Jervis is largely exculpatory for the I.C., and his overall take is worth quoting at length:
“The fundamental reason for the WMD intelligence failure in Iraq was that the inferences were very plausible, much more so than the alternatives. Saddam had vigorously pursued WMD in the past, had used chemical weapons to good effect, had powerful incentives to rebuild his programs, had funds, skilled technicians, and a good procurement network at his disposal, and had no other apparent reason to deceive and hinder the inspectors…
Saddam sought to maintain the appearance of WMD in order to keep power at home and deter Iran. The United States was a secondary concern... Saddam’s policy, then, was foolish and self-defeating, and this goes a long way to explaining the intelligence failure. When the situation is this bizarre, it is not likely to be understood…
Even if there had been no errors in tradecraft, I believe that the analysts would and should have judged that Saddam seemed to be actively pursuing all kinds of WMD and probably had some on hand.”
Chapter 4 - The Politics and Psychology of Intelligence and Intelligence Reform
Tensions between analysis and decision-making
Jervis argues that there are fundamental tensions between the aims and functions of political leaders and the aims and functions of intelligence.
- Political leaders need to oversell their policies, and intelligence will usually detract from that aim.
- Similarly, leaders need psychological confidence in their plan of action. Decision makers also face a psychological need to believe their chosen option doesn’t have steep costs or trade-offs, though these are usually inevitable. Because intelligence is usually descriptive, not normative, it is free to point out these trade-offs, and deflate confidence.
- Leaders also want to avoid considering fallback plans to minimise complexity and the contemplation of failure by themselves and their teams. Intelligence is usually more alive to ongoing or threatened failures and alternative options.
- There is a narrow window “after the leaders have become seized with the problem but before they have made up their minds”. Most intelligence is irrelevant to leaders because they are too busy to consider the specific topic. This is until the topic is urgent and important enough, after which a policymaker will soon need to politically and psychologically commit to an interpretation and plan.
On international politics
Jervis argues that international politics is not like chess, where all information is visible. Instead, deception is powerful and easy. This might suggest an analogy of poker, but this also falls short, as players are often engaged in different games without even realising. Instead Jervis suggests international politics is like Rashomon - the Japanese fable where each participant sees the interaction and each other differently.
A common dictum is ‘When you hear hoofbeats, think horses, not zebras.’ Jervis sympathises with this message, but argues that major intelligence failures will primarily occur with ‘zebras’ - events that are rare and outside expectation. He suggests a number of kinds of ‘zebras’ in international politics:
- foreign states with unknown motivated biases, whose future behaviour could look strange and irrational on a typical rational actor view.
- actors with unusual world-views, such as from domestic politics or idiosyncratic leaders
- states doing something new (e.g. USSR missiles in Cuba) - we’re prone to expect continuity, not change, and we’re slow to adapt our worldviews to new evidence that doesn’t fit.
Intelligence reforms relevant to EA
Jervis also makes a number of suggestions to reform intelligence services, though he seems pessimistic that they will ever be implemented. Below I’ve summarised the recommendations that are most relevant to EA:
- intelligence should keep more careful track of sources, clarify degree of confidence, avoid unwarranted consensus, and cite and build on supporting and dissenting reports.
- there should be a freer flow of information between agencies (though this needs to be balanced with counterintelligence considerations)
- intelligence should conduct post-mortems and maintain product evaluation staffs
- intelligence should strengthen critical middle management and peer review
- intelligence should build deep contextual knowledge and empathy
- intelligence should examine fundamental beliefs that influence plausibility and implausibility and challenge them. It should notice and interrogate how fundamental beliefs interact with and influence pieces of evidence. It should consider:
- alternative hypotheses
- negative evidence (‘dogs that don’t bark’)
- disconfirming evidence
Feel free to suggest some books, either here or by DM. I'm looking for highly regarded books outside the typical EA canon that might contain useful conceptual tools or insights, especially from political science and international relations. Here are some possibilities I’m thinking of:
‘How Statesmen Think’ by Robert Jervis
‘Perception and Misperception in International Politics’ by Robert Jervis
'A History of Global Health', by Randall Packard
'The Globalizers', by Ngaire Woods (on the World Bank and IMF)
'Metazoa', by Peter Godfrey-Smith (on the evolution of subjective experience)
'Health Justice', by Sridhar Venkatapuram (on applying the capabilities approach to health)
'On War', by Claude von Clausewitz
'The Strategy of Conflict', by Thomas Schelling
'How the World Thinks', by Julian Baggini
'Modern Strategy', by Colin Gray
If you really want more but can't be bothered acquiring and reading the book, my rough notes might sate. You can read them here.