i

Description

Un bocal de verre rempli de billes de couleurs diverses, sur fond noir.

Source

Anneke from Groningen · original page

Licence

PD Public domain

Original photo, unretouched. The background is a version of it stylised by the site.

Ambiguity aversion

Between a risk we know how to quantify and a risk we do not know how to quantify, we choose the first, even when nothing justifies it.

In brief

• What : for equal gain, we avoid the option with unknown probabilities in favor of the one with known probabilities.
• Why : the unknown raises fears that something is being hidden from us or that we are mistaken; we feel more competent with what we can quantify.
• Detect : “we'll go with the one we know,” “we don't have enough hindsight,” a choice justified by familiarity rather than numbers.
• Avoid : turn ambiguity into a range, test on a small scale, compare both options in the same terms.

On this page

1. Definition
2. The mechanism
3. Where it shows up
   1. In project management
   2. In management
4. Famous cases
   1. The Ellsberg paradox 1961
   2. Betting on what we think we know 1991, 1995
5. Detection signals
6. Avoidance reflexes
7. With AI: detect, fix, mitigate
   1. Detect
   2. Fix
   3. Mitigate
   4. Prompts to incorporate into your workflows
   5. How to integrate them
   6. Precautions
8. Limitations
9. Sources and verification
10. See also
    1. Combines with, or reinforces
    2. Related biases
    3. Counterweights and related reflexes

Definition#

Fictitious illustration

Fictitious illustration

Ambiguity aversion is the tendency to prefer options whose probabilities are known over those whose probabilities are unknown, even when expected value is identical or better for the latter. It was described by Daniel Ellsberg in 1961, in an article that became famous as the Ellsberg paradox.

Since economist Frank Knight (1921), we distinguish risk (probabilities are known: a die, an urn whose contents are known) from uncertainty or ambiguity (they are not). Almost every IT project falls into the second category, while we reason as if it fell into the first, or we simply avoid anything that falls into the second.

The mechanism#

• Fear of being duped. An unknown probability may have been chosen by someone who knows more than we do. We protect ourselves by choosing the transparent option.
• Feeling of competence. Heath and Tversky (1991) showed that people are willing to bet on what they believe they know, to the point of paying to do so, and that in domains where they feel ignorant, they prefer a bet with known probability.
• Comparative ignorance. Fox and Tversky (1995) observed that aversion appears mainly when we directly compare the known option and the vague option; evaluated alone, each is judged about the same. It is the side-by-side comparison that makes the unknown uncomfortable.
• The cost of an attributable error. Being wrong with an unusual option is hard to justify; being wrong with the standard solution is much easier.

Where it shows up#

In project management#

• Choosing a vendor: the provider whose flaws are known is preferred over a cheaper, more promising newcomer whose risks we do not know how to quantify. “Nobody ever got fired for choosing the market leader's solution” is the popular version.
• Choosing a technology: an established framework, with known limits, beats a recent tool whose limits are still to be discovered, even if the latter would solve real problems.
• Migration: a platform change is postponed because “the risks are poorly known,” even though the risks of the current situation (ending support, debt) are just as vague but familiar.
• Estimation: a discovery task (“we don't know what's in it”) is avoided or pushed to the end of the project, instead of being explored early.

In management#

• Hiring and promotion: we choose the profile whose track record we know rather than the atypical profile whose chances of success we do not know how to estimate.
• Delegation: we assign a topic to the person who “has already done it” rather than someone whose level we do not know, and we never discover other people's skills.
• Innovation: initiatives with uncertain return are rejected in favor of improvements whose gain is measurable, but weaker.

Famous cases#

Detection signals#

• “We'll go with the one we know”: familiarity is given as the reason, with no numbers.
• A newcomer is rejected because it does not have “enough references” or “enough hindsight,” even though no one tried to verify its risk.
• The risks of the new option are listed precisely; those of the current option are never listed.
• Discovery tasks are systematically pushed back.
• A choice is justified by “at least we know what we have,” without comparing the expected value of both options.

Avoidance reflexes#

1. Turn ambiguity into a range. “This provider is an unknown” becomes “cost between 80 and 160 days, with a dropout risk we estimate between 5% and 20%.” A range, even a wide one, is comparable; a “we don't know” is not.
2. Apply the same formalism to both options. Describe the familiar option with the same sections: cost, timeline, risk, with its own uncertainties (ending support, people leaving).
3. Reduce the unknown with a small test. A one- to two-week proof of concept, a pilot on a limited scope: ambiguity becomes a risk we know something about.
4. Run a pre-mortem on each option. “In one year, the choice is a failure: why?” for the familiar option as well as the new option, to restore balance in the arguments.
5. Evaluate separately before comparing. Each option is described and rated by different people before being put into a table, to limit the comparison effect.
6. Ask what it costs not to try. Rejecting the vague option has an opportunity cost, rarely estimated.
7. Plan reversibility. A decision you can undo (short-term contract, abstraction isolating the tool) makes ambiguity more bearable.

With AI: detect, fix, mitigate#

Prompts to incorporate into your workflows#

1. Make the unknown comparable. When an option is rejected because it is not well known.

We are hesitating between a familiar option [describe] and a less familiar option [describe], for [objective].
1. List the uncertainties for EACH option (cost, timeline, quality, dependencies, people leaving, market evolution), without favoring one.
2. For each one, suggest a cautious range, specifying what it is based on and what would make it change.
3. Indicate which uncertainties would be resolved by a one- to two-week trial, and which trial.
Do not decide: I want to be able to compare both options in the same terms.

2. Symmetrical pre-mortem. Before deciding between a known option and a new option.

We chose [option X] for [objective]. One year later, the choice is a failure.
Write the five most plausible causes, from most likely to least likely.
Then do the same exercise for [option Y]: the same choice, in the same situation, would have been a failure.
Compare both lists: which causes are common, and which apply to only one option? Which ones would we have dismissed out of habit without thinking them through?

How to integrate them#

• Before the decision committee: prompt 1 for each disputed option, then a shared table.
• During: prompt 2 in a workshop, comparing both lists of failure causes.
• After: keep the ranges and check, at the end of the project, which one was right.

Precautions#

• A generated range is not a measurement: require the source and verify it.
• AI tends to suggest “safe” options and recommend the solutions most cited in its data: it can amplify the preference for the familiar.
• Do not share bids, pricing, or supplier data with an unauthorized tool.

Limitations#

• Aversion is not universal: according to reviews, it can reverse (preference for ambiguity) when the probability of gain is low or that of loss is high.
• Refusing ambiguity can be rational: an unknown risk may hide a selection bias (the person proposing knows more), and a fragile organization may not be able to absorb a bad draw.
• The literature distinguishes ambiguity aversion from status quo bias and loss aversion, with which it is often mixed in reality.

Sources and verification#

• Ellsberg (1961), “Risk, Ambiguity, and the Savage Axioms,” The Quarterly Journal of Economics, 75(4) — the urn paradox.
• Knight (1921), Risk, Uncertainty and Profit, Houghton Mifflin — the distinction between risk and uncertainty.
• Heath & Tversky (1991), “Preference and Belief: Ambiguity and Competence in Choice under Uncertainty,” Journal of Risk and Uncertainty, 4(1).
• Fox & Tversky (1995), “Ambiguity Aversion and Comparative Ignorance,” The Quarterly Journal of Economics, 110(3)
• Camerer & Weber (1992), “Recent Developments in Modeling Preferences: Uncertainty and Ambiguity,” Journal of Risk and Uncertainty, 5(4) — review.

See also#