
Tim Harford on Adapt and the Virtues of Failure 05/23/2011
What this covers
Tim Harford talks with Russ Roberts about his book Adapt, and the conversation starts on the battlefield. Harford describes the 1991 Eagle Troop engagement, where nine US tanks met roughly a hundred Iraqi vehicles in a sandstorm and the captain had to decide alone. He uses it to illustrate Hayek's point that decision-relevant knowledge sits with people on the ground. The same argument carries over to economies. Harford covers Chile's Cybersyn project, the 1920s–30s calculation debate, and the three petabytes in the Lehman bankruptcy examiner's report, all as cases where central oversight cannot absorb tacit, local information.
The middle of the conversation turns to why organizations fail to learn. Harford draws on Groupthink research from the Bay of Pigs and on Kennedy's attempt to break it during the Cuban Missile Crisis by splitting up his advisers. He then looks at Iraq, where H.R. McMaster's approach in Tal Afar ran against a leadership that avoided the word "insurgent." James Reason's taxonomy of slips, mistakes, and violations explains how errors can lie dormant until a crisis.
The final stretch is more contested and more personal. Harford discusses Arnold Kling's hard to break versus easy to fix distinction and Taleb's warnings about systems too big to fail. He compares peer inspection in nuclear power with its absence in banking, and notes that bankers were paid on profits that were not real. The closing material concerns individual psychology: denial of failure, "going on tilt," and loss aversion that discourages small, survivable experiments.
Harford argues that success in complex systems depends on trial and error, and that political, organizational, and psychological incentives make learning from failure unusually hard.
- Complex systems solve problems through diversity of experiments, survivable failures, and the ability to tell success from failure
- Top-down control and groupthink suppress the local knowledge and candid feedback needed to adapt
- Politics and personal psychology penalize admitting failure, while markets and science tolerate many small failures for a few big successes
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Mistakes and violations share a key feature: both can lie dormant and resurface days, weeks, or years later, such as a valve left open or a disabled safety system that is not discovered until a crisis, so the crucial question is how to spot the error before the critical moment.
“they can both just sit there and come back to bite you days weeks even years after the original mistake”
In the 1991 Gulf War, nine US Eagle Troop tanks ran into roughly a hundred Iraqi Republican Guard tanks and armored personnel carriers in a sandstorm with no satellite or air cover; the captain ordered an attack, destroying enemy tanks within seconds, and the battle was celebrated as proof of US superiority even though the captain concluded it had nearly failed because he had to decide alone.
“they ran into nearly a hundred tanks and armored personnel carriers that dug in defensive emplacements and their Saddam Hussein's Republican Guard”
During the Cuban Missile Crisis, Kennedy counteracted groupthink by splitting his expert advisers into separate discussion groups, bringing them together only occasionally, and bringing in outside experts deliberately to shake things up, because he realized that asking for honest opinions alone would not work.
“he actually split his expert advisors up and forced them to have separate discussions and occasionally would bring them back together”
The Chilean Cybersyn project under Salvador Allende failed because centralized computers processing telex reports could not capture the subtle, local information needed to run an economy: people had no reason to report the truth when it was inconvenient, and the system worked only during strikes, when the information was simple.
“the only time that cybersyn really worked was during the strikes when basically the information was really simple”
Erik Brynjolfsson's research at MIT, roughly a decade old at the time, found a strong connection between adopting new information technology and decentralizing responsibility, so firms must either keep old technology with centralization or decentralize and adopt the new technology together.
“there's a really strong connection between new technology and decentralization of responsibilities they go hand in hand”
When people fail, they either deny that the failure happened or respond badly by 'going on tilt' in the poker sense, doubling down and chasing losses, and both responses prevent the distinction between success and failure that adaptation requires.
“what a poker player would call going on tilt effectively you double down you chase your losses so there were different ways in which we can either respond in a bad way to failure or we can deny that a failures ever happened”
Voters rarely scrutinize political candidates as carefully as they would scrutinize a car purchase, because an individual vote is unlikely to change the outcome, so voters pay attention to issues that affect their own lives and political accountability is weak.
“I didn't really think that my vote was gonna make the difference and so I was kind of paying more attention to stuff that actually made a difference to my life”
Leonard Read's essay 'I, Pencil' shows the economy is incredibly complex, a distributed system in which people cooperate toward common goals without knowing them; Harford's toaster project, in which a design student tried to build a toaster alone, illustrates how hard it is to reproduce such complexity, and Harford concludes that the more complex the system, the harder it is to solve its problems through deliberate change.
“the more complex and amazing you think the system is the more respects you give the system the more challenging this project of actually trying to solve problems in a modern economy becomes”
Hayek's knowledge problem means local knowledge is often tacit, so it is hard to extract from people's heads into a central authority, and even if it could be, the knowledge burden is basically impossible, a problem not solved by computers; experts also fail because of groupthink, ego, and power.
“often they don't even know what that knowledge is until they have to use it how do you get it out of their heads into a central authority”
Safety systems change incentives: people become less worried when they see something going wrong because they know a backup exists, and they fail to realize the backup may be disabled for maintenance at the critical moment, so rare failures line up into catastrophes.
“what you notice is that the incentives get changed by the safety systems so absolutely a lot of times in industrial accidents people aren't so worried when they see something going wrong because they know there's a backup”
Arnold Kling distinguishes 'easy to fix' from 'hard to break'; society has gravitated toward trying to build systems that never fail, which makes the rare failure that does occur catastrophic, when it would be better to accept occasional small failures that are cheap to fix.
“he makes an interesting distinction between easy to fix versus hard to break and the argues that we have a tendency to try to find systems that will never fail”
Rare catastrophes are almost impossible to measure, so a strategy of making catastrophes rare without limiting their size is unwise; Nassim Taleb's point is that systems should be robust so that when parts break, the damage is contained, and nothing should be too big to fail.
“these rare catastrophes it's very hard to say exactly how rare they really are they ugly it's almost impossible to measure them”
Friedrich Hayek's concept of local knowledge, the 'particular circumstances of time and place', matters because decision-relevant knowledge is held by people on the ground and is often tacit, so it cannot be transmitted to a central authority that then decides and sends orders back down.
“Hayek talks a lot about how local knowledge that's in the heads of individuals it's incredibly difficult to get it out of their heads often they don't even know what that knowledge is until they have to use it”
Colonel H.R. McMaster pioneered a responsive, intensive counterinsurgency approach in Tal Afar in 2005 while Donald Rumsfeld and senior officers avoided even using the word insurgent, so official language blocked discussion of what was going wrong and why strategy needed to change.
“he was responsible for US operations near a city called talofa what Colonel McMaster did was pioneer a new way of dealing with insurgents”
The Lehman Brothers bankruptcy examiner's report covered an estimated three petabytes of data, roughly 350 billion pages, which makes top-down oversight of an organization practically impossible even as technology makes information sharing easier.
“the universe of Lehman's email and other electronically stored documents is estimated at three petabytes of data roughly the equivalent of three hundred and fifty billion pages”
Distinguishing success from failure is hard in finance because bankers were paid bonuses on accounting profits that were not real economic profits, since the risks embedded in those profits could not be measured.
“bankers were being paid bonuses based on apparent profits that turned out you know there were accounting profits but they weren't real economic profits because there were risks embodied that couldn't be measured”
Colonel H.R. McMaster's account of the Vietnam War, and his later counterinsurgency work in Tal Afar in 2005, showed that the military needed to give more responsibility and attention to frontline soldiers and captains, and his career risk reflected how hostile top leadership was to bottom-up adaptation.
“he campaigned wrote a series of papers and analyses saying you need to give more attention to the training and responsibility of the troops on the ground”
Friedrich Hayek's concept of local knowledge, the particular circumstances of time and place, is illustrated by the 1991 Eagle Troop tank battle, where a captain with only local information had to make a split-second decision that a top-down command could not have made.
“it emerged in a tank battle of all places”
Humans have a strong impulse to gloss over their mistakes and see them as successes; growing up involves learning to live with and admit mistakes, a point Harford connects to Adam Smith's observation about the self-delusion that hides one's own conduct.
“I think we have this very human impulse to gloss over our mistakes to see them as successes and I think part of growing up is learning to live with them and admit them”
Loss aversion, identified by Daniel Kahneman and Amos Tversky, makes people overweight the pain of a downside such as rejection, so they avoid low-cost experiments like asking someone out even when the upside is great and the downside is survivable.
“loss aversion the idea that taking a loss is disproportionately painful and that makes it really hard to just try out these new things”
Groupthink, as Irving Janis showed in the Bay of Pigs fiasco, arises from social processes rather than a domineering leader: advisers stayed silent because everyone else seemed to agree, and agreeing with the room is the low-risk strategy for individuals.
“it's not that Kennedy was a substan and esque figure who suppress dissent he was trying to get people to tell him what they thought but there were just social processes”
Nuclear power's peer monitoring, where operators visit and inspect each other's plants, creates a strong incentive to catch errors, whereas banking lacked an equivalent peer-inspection force, though such monitoring is hard to operationalize.
“they inspect each other's inks of power stations and they issue reports and they learn from each other and that's something that I think was a very weak force in banking”
James Reason's taxonomy distinguishes slips (doing something unintended, like pressing the wrong button), mistakes (acting on a flawed view of the world, such as buying a house on a subprime mortgage assuming prices would keep rising), and violations (knowingly breaking rules, like falsifying an income claim); mistakes and violations can sit dormant and bite days, weeks, or years later.
“the psychologist James Reason talks about three kinds of error he says the slips mistakes and violations”
The socialist calculation debate of the 1920s and 1930s showed that a planner with enough information and computing power still could not assign correct prices and quantities, because the subtle information about substitutes, supply increases, and time-and-place circumstances cannot be easily put into a computer or captured on a survey.
“that information can never be easily put into a computer or answered on a survey so it's it's really a utopian fantasy”
Politicians have incentives not to measure whether their policies succeed or fail, because rigorous measurement is often inconvenient knowledge for them.
“it's often not in politicians incentives interests to measure whether their particular policies are succeeding or failing obviously that's often inconvenient knowledge”
The fog of war makes top-down battlefield control fail: at Gettysburg, Lee missed Stonewall Jackson's freelancing and did not trust his subordinates on the ground, and waited for information that was badly distorted, which helped cause the defeat.
“I think one of the reasons we failed at Gettysburg is that he didn't trust the guys he had on the ground and he was waiting for information which he was horribly misinformed”
Expectations of government bailouts reduced the vigilance of firms in the financial crisis, especially lenders to banks, so the bailout expectation weakened focus on mistakes and slips, though Harford does not believe the crisis was mainly caused by laziness or willful fraud.
“I don't fully buy the idea that all of this stuff was was extreme laziness or or even sort of willful fraud because everybody knew the government would would bail them out”
Steve Jobs's Newton, a commercial flop, did not diminish his reputation because his iPhone and other successes outweighed it, but Alan Greenspan's reputation has collapsed into 'loser' status over failures such as the 2002-2004 interest rate policy and his support for the 1995 Mexican bailout, showing that a single high-profile failure can dominate a politician's legacy.
“Alan Greenspan appears to be coming down in history as a loser he was a genius and then all of a sudden because of his interest rate policies of 2002 through 2004”
Markets and science are asymmetric: one big success compensates for many failed firms or theories, because selection processes let the surviving success dominate, whereas in politics a single embarrassing failure can end a career and so discourages trying a diverse set of policies.
“in the case of market and in the case of the scientific method basically one success makes up for a lot of failures”
Economists have little to say about industrial accidents, a literature dominated by engineers, psychologists, and sociologists studying how complex systems go wrong, including the Challenger shuttle disaster, Three Mile Island, Chernobyl, and Deepwater Horizon, and this literature is highly relevant to understanding financial crises.
“this is a literature that economists have pretty much nothing to say about and it's dominated by engineers psychologists and a couple of sociologists”
Top-down control of language, such as the Pentagon refusing to call the Iraq violence an insurgency, is more damaging than a single press-conference gaffe because when strategy is failing, an organization that cannot even name the problem has no way to change the strategy.
“more important than the fact the strategy was failing because strategies do fail a lot if there was no way to change the strategy you couldn't even talk about what was going wrong”
Organizations implement lessons learned on the front line quickly because lives depend on it, but rarely implement lessons learned at the top because senior decision-makers face no pressure or incentive to change.
“we always implement lessons learned on the frontline because lives are lost or saved by how quickly you do that but we very rarely implement lessons learned at the top because there's just no pressure and no incentive to do so”
Harford argues that mistakes and violations matter less than detecting that an error has occurred, because safety systems that are supposed to catch errors, like the financial-crisis insurance and hedges that regulators and top managers approved, can silently fail until a crisis reveals them.
“if the safety system doesn't work and whether it's a mistake or a violation maybe doesn't matter as much as hey how do we spot that an error has been made in the system”
The deeper reason politicians can be framed as losers for past failures is the measurement problem: most political outcomes are vague, so proponents and opponents can each claim success or failure, whereas a failed consumer product is clearly discontinued and cannot be credibly held up as a success.
“the difficulty of assessing success in failure so there's some political things that are undoubtedly failures there's some that are undoubtedly success but most of them are vague”
Regulators can be captured by the industries they oversee, so the best safeguard is systems that work through incentives, such as peer monitoring in the nuclear industry, where operators inspect each other's plants and share reports because no one wants a nuclear accident.
“they inspect each other's inks of power stations and they issue reports and they learn from each other and you can sort of see how that would no nobody in the nuclear industry wants a nuclear power plant to blow up”
The captain of Eagle Troop concluded the battle did not validate top-down command but showed that the battle was won only because he had to make a split-second decision himself, and he subsequently campaigned for more training and responsibility for troops on the ground.
“the captain who experienced that situation he came away with a different conclusion he said look we we all nearly died and the reason we didn't die is because I had to make a split-second decision”
Successful adaptation in a complex system requires three conditions: diversity (trying many different things so failures are common), survivability (failures must be survivable and not bring down the system), and the ability to distinguish success from failure so that successes are replicated.
“it's basically three things one is you try a bunch of different stuff a variety of things everyone can be doing the same thing next since failures will be common it needs to be okay for some of that stuff to fail”
Bottom-up organization does not mean unplanned or random; it means knowledge and its use come from the bottom, but the top must create structures for bottom-up learning to emerge and must trust it, and planning still goes on in every organization.
“when we talk about bottom-up we don't mean unplanned it doesn't mean that everything's random everybody's freelancing it means that this the knowledge and the use of knowledge is coming from the bottom”
New technology does not obviously push organizations toward centralization; the common assumption that better communication and computing strengthens top-down control is a fantasy, because information volume overwhelms central review.
“it's not at all obvious that new technology is a force for centralization”
University is a uniquely safe space for experimentation: students try new friends, places, ideas, and hobbies, yet the stakes are low because they can focus and pass exams, and adults become more conservative because they lose this structure of safe experimentation.
“it's a very safe space for experimentation that's why it's so thrilling but we really struggle to do that I think lay to an adult life we get very conservative”
Politicians such as George W. Bush and Barack Obama almost never admit mistakes when asked, which Harford suggests is less a strategic calculation than a pathological inability to concede error, even though admitting error could be necessary to move forward.
“both George Bush and Barack Obama just to pick two recent examples when asked if they'd made any mistakes they just they don't seem to find them coming to mind”
Changing one's mind during a conflict is seen as toxic in politics: John Kerry was painted as a flip-flopper in the 2004 campaign, even though many reasonable people changed their minds about the Iraq war as it went badly, which is a sign of sanity rather than weakness.
“a lot of perfectly reasonable people change their minds halfway through that's not that's an indication of sanity and yet politically very toxic”
Politics is an adversarial system where opponents exploit any past failure in attack ads, while business rivals such as Apple and Microsoft rarely dredge up competitors' failed products like the Newton, which Roberts argues is part of why politicians are punished more for failure.
“maybe one of the things is it isn't a politics is an adversarial system this is less so so you don't have though there's some needle between Steve Jobs and Bill Gates”
Senior officers become so attached to their roles in hierarchical organizations that they will sacrifice adapting to the war they are fighting; an officer's statement during Vietnam, 'I'll be damned if I'm going to ruin this army just to win the war,' captures this preference for preserving organizational structure over effectiveness.
“I'll be damned if I'm going to ruin this army just to win the war”
Battlefield coordination has no price system, so central command is useful for coordinating one-off lightning strikes such as the opening of the Iraq war, but counterinsurgency requires trusting ground forces and local relationships that cannot be coordinated from the top.
“you don't have a price mechanism and you probably don't want to delegate authority completely to online now you want to coordinate that from the top down yeah but but with many problems including the counter insurgency problem”
Harford's closing argument is that the more complex and impressive an economy is, the more respect it deserves and the harder it is to change it to solve problems, so reform must proceed with humility and an acceptance that it is acceptable to fail in the right way while generating information.
“as long as you're failing in the right way as long as you're generating information it's okay to experiment”
In a complex world, failure is pervasive and not necessarily alarming: very successful economies and organizations have high failure rates, so the key skill is managing failure, responding to it, and learning from it.
“very successful economies have very high levels of failure successful organizations have high levels of failure but we can't really get away from this failure and so a key skill is being able to manage it and respond to it and learn from it”
Value at risk is a poor monitor of real risk, and it fails to reveal danger when institutions believe they are insured, because savvy people in 2008 knew everyone was insured by AIG and similar firms but still slept soundly, suggesting incentive or cognitive failure rather than mere ignorance.
“you value at risk it doesn't work so well as a monitor of real risk but yeah you know it's not so bad if it fails or yes I'm insured but what if the insurer fails”
Ronald Reagan is remembered as a staunch free-market principled president, yet he imposed quotas on Japanese cars, expanded government size, and withdrew from Lebanon, showing that principled-seeming figures often break principles while framing pragmatism as principle.
“Ronald Reagan has this great reputation as this great free-market capitalist president he put on the effective quotas on Japanese cars”
Changing public attitudes toward failure, and treating failure as acceptable when it generates information, is essential; political systems currently do not take evidence seriously, and policy is often hand-waving rather than the high-quality evidence a doctor or epidemiologist would demand.
“as long as you're failing in the right way as long as you're generating information it's okay to experiment”
Email and news groups let ground troops in Iraq swap tactical information, such as the latest roadside bomb tricks, and a PowerPoint by Captain Travis Patrick Quinn explaining how to win in Al Anbar spread through the force without official printing or distribution, showing that bottom-up information flows can outrun formal channels.
“it's it was passed around like like thermostat so the these people were able to communicate with each other”
The 1988 Piper Alpha oil platform disaster in the North Sea triggered a small financial crisis in the British insurance industry, which Harford used as a way into understanding the larger 2008 financial crisis.
“Piper alpha triggered a small financial crisis in the British insurance industry so I thought hey I'll write about that financial crisis mini financial crisis because it'll be a way into the big financial crisis”
Captain Travis Patrick Quinn, who created the PowerPoint, died in December 2006 and was honored by local sheiks at his funeral because he understood the province and built alliances.
“all the local sheiks came to his funeral because he was so respected he really understood how things worked in that province”
Britain's most electorally successful prime ministers, Tony Blair and Margaret Thatcher, each won three elections despite slogans that signaled refusal to change course, showing voters can reward inflexibility, yet both leaders also changed policy while presenting themselves as principled.
“Tony Blair I don't have a Geir Margaret Thatcher u-turn if you want to the lady's not for turning”
In business, failure is measured and visible, since a company that goes broke cannot be called a success, but American culture, especially in Silicon Valley, tolerates and even expects multiple failures before success and gives entrepreneurs second chances.
“in the business world mistakes are measured your company goes broke you can't say it was a success”
The UK Independent Banking Commission, chaired by Sir John Vickers, pushed to make banks more modular so that a chunk of a bank can be isolated and rescued while depositors are protected and the rest of the bank continues, but bankers resist such breakup and modularization.
“the banking commission which is run by Sir John Vickers they've definitely pushed towards trying to make banks more modular”
Harford's undercover-economist-style interest in local knowledge led him to examine whether markets can coordinate complex production, but the more important question is how to decide which parts of a working complex system still need problem-solving change.
“hang on are we actually happy with every aspect of the way the world is or have we got some problems”
Greetings and program logistics such as sponsor credits, the sound-engineering experiment, and sign-offs are not substantive.
“welcome to econ talk”