Availability heuristic AI governance: how the ease of recalling AI-related incidents distorts risk perception and governance decisions in nonprofits
Availability heuristic AI governance describes the tendency to judge the likelihood of an event based on how easily examples come to mind rather than on systematic evidence. First identified by Tversky and Kahneman in 1973, and extended by Schwarz and colleagues in 1991 to account for the role of retrieval ease itself as information, availability heuristic AI governance risk shapes how community organisations assess the risks and benefits of AI adoption. When dramatic AI failures are easy to recall from media coverage, those risks feel larger than base rates justify. Conversely, when AI errors in the organisation’s own practice go undocumented and unshared, they are unavailable for recall and therefore feel less likely than they actually are. The result is a systematic distortion of governance decision-making that favours salient, memorable risks over mundane but statistically more significant ones. Availability heuristic AI governance connects to attribute substitution, normalisation of deviance, and security theatre in the NFH Lexicon, and underpins the NFH Behaviour Risk Card framework for nonprofits and community organisations in Australia.
Availability Heuristic
Risk judgements are not based on statistics. They are based on what comes to mind. When AI failures are dramatic and well-publicised, they feel more likely than they are. When AI errors in your organisation go unrecorded, they feel less likely than they are. Both distortions shape governance in opposite but equally damaging directions.
Availability heuristic AI governance: how the ease of recalling AI-related incidents distorts risk perception and governance decisions in nonprofits
Definition
What is Availability Heuristic AI Governance Risk?
The availability heuristic is the cognitive shortcut by which people estimate the likelihood or frequency of an event based on how easily examples of that event come to mind. Events that are vivid, recent, personally experienced, or heavily covered in media feel more probable than events that are abstract, distant, or rarely discussed, regardless of their actual statistical frequency.
The heuristic is not irrational in origin. In most everyday contexts, frequency and retrievability are genuinely correlated: common things tend to be more memorable because they are encountered more often. However, this correlation breaks down systematically when salience and frequency diverge, which is precisely the condition that AI governance operates in. Media coverage of AI failures is selective, dramatic, and concentrated on high-profile cases. Meanwhile, routine AI errors in small organisations go undocumented and unshared.
Retrieval Ease as Evidence
Schwarz and colleagues extended the original account by demonstrating that it is not just the content of what is recalled that influences judgment, but the ease of retrieval itself. When examples come to mind easily, people interpret that ease as evidence that the event is common. When recall is effortful, they interpret the difficulty as evidence of rarity. Applied to AI governance, this means that organisations with poor incident documentation will systematically underestimate the frequency of AI errors in their own practice, because those errors are genuinely hard to recall.
Research Origin
Tversky and Kahneman (1973) identified the availability heuristic as one of a small set of cognitive shortcuts that people use to make judgments under uncertainty. Through a series of experiments, they demonstrated that frequency and probability judgments are reliably distorted by the ease with which examples come to mind. Crucially, factors that influence retrievability but are unrelated to actual frequency, such as vividness, recency, and media exposure, produce systematic and predictable errors in risk assessment.
Schwarz, Bless, Strack, Klumpp, Rittenauer-Schatka and Simons (1991) refined the theory by showing that the subjective experience of retrieval ease, not just the content retrieved, functions as information in its own right. In their experiments, participants who were asked to recall many examples of assertive behaviour rated themselves as less assertive than participants asked to recall fewer examples, because retrieving many examples felt difficult and that difficulty was interpreted as evidence of scarcity. This finding has direct implications for AI governance: the difficulty of recalling AI errors becomes itself a signal that errors are rare, compounding the underestimation already produced by poor documentation.
Kahneman (2011) situated the availability heuristic within his broader account of System 1 and System 2 thinking, characterising it as a fast, automatic judgment that System 2 can override but typically does not. He documented the conditions under which availability distortions are most pronounced: high cognitive load, time pressure, emotional salience, and the absence of statistical information. These conditions closely describe the environment in which most community organisation governance decisions about AI are made.
In AI Contexts
How the Availability Heuristic Shows Up With AI
AI governance decisions are particularly susceptible to availability distortion because the information environment around AI is highly uneven. High-profile failures, such as AI systems producing discriminatory outputs, generating dangerous misinformation, or being used in harmful ways, receive extensive media coverage and are therefore highly available for recall. Lower-stakes but more frequent errors, such as AI outputs that are subtly wrong, contextually inappropriate, or missing critical nuance, are rarely reported and consequently feel much less common than they are.
The Documentation Gap
Within organisations, availability distortion is further amplified by the absence of systematic AI incident recording. When AI errors are not documented, they cannot be recalled as a class of events with a frequency. Instead, each error is experienced as isolated, and the general impression is that errors are rare. Moreover, staff who have encountered AI errors but not reported them hold that knowledge individually rather than organisationally, making it unavailable to governance decision-makers who rely on collective recall.
The governance consequence runs in both directions. Overestimation of dramatic risks, driven by media salience, produces reactive governance responses disproportionate to actual likelihood. Underestimation of routine errors, driven by poor internal documentation, produces insufficient governance for the risks that are actually most frequent. Neither distortion serves the communities the organisation is trying to protect.
Why It Matters for Nonprofits
The Specific Risk for Community Organisations
Community organisations rarely have the capacity to conduct systematic risk analysis of AI adoption. Governance decisions about which AI tools to use, how to oversee them, and what safeguards to implement are typically made by leadership under time pressure, drawing on whatever information is most readily available. That information is disproportionately drawn from external media coverage and from the most memorable internal incidents rather than from systematic evidence.
Building Availability Deliberately
The practical governance implication of the availability heuristic is counterintuitive: making AI errors more available for recall is a governance improvement, not a problem. Organisations that systematically document AI incidents, share learnings across the team, and regularly review AI error patterns are actively counteracting availability distortion. They are building an accurate internal information environment to replace the distorted one that unmanaged availability produces.
Regular review of AI errors also serves a secondary governance function: it provides the evidence base that makes the case for governance investment. Without that evidence, the low retrievability of routine errors will consistently work against governance resourcing decisions.
Availability heuristic AI governance describes the tendency to judge the likelihood of an event based on how easily examples come to mind rather than on systematic evidence. First identified by Tversky and Kahneman in 1973, and extended by Schwarz and colleagues in 1991 to account for the role of retrieval ease itself as information, availability heuristic AI governance risk shapes how community organisations assess the risks and benefits of AI adoption. When dramatic AI failures are easy to recall from media coverage, those risks feel larger than base rates justify. Conversely, when AI errors in the organisation’s own practice go undocumented and unshared, they are unavailable for recall and therefore feel less likely than they actually are. The result is a systematic distortion of governance decision-making that favours salient, memorable risks over mundane but statistically more significant ones. Availability heuristic AI governance connects to attribute substitution, normalisation of deviance, and security theatre in the NFH Lexicon, and underpins the NFH Behaviour Risk Card framework for nonprofits and community organisations in Australia.
Observable Signals
What to Watch For
These patterns suggest availability heuristic distortion may be shaping AI governance decisions in your organisation. They are signals worth naming and discussing, not a definitive checklist.
- Governance discussions about AI risk rely primarily on external examples from media coverage rather than on documented internal incidents, suggesting that availability of external cases is substituting for systematic internal risk assessment.
- Staff who have encountered AI errors describe them as unusual or isolated even when similar errors may have occurred elsewhere in the organisation, indicating that the absence of shared documentation is producing individual rather than organisational recall.
- The organisation has no systematic process for recording, reviewing, and sharing AI incidents, meaning that governance decisions about AI risk are made without access to the internal evidence base that would correct availability distortion.
Availability heuristic AI governance: how the ease of recalling AI-related incidents distorts risk perception and governance decisions in nonprofits
Availability heuristic AI governance describes the tendency to judge the likelihood of an event based on how easily examples come to mind rather than on systematic evidence. First identified by Tversky and Kahneman in 1973, and extended by Schwarz and colleagues in 1991 to account for the role of retrieval ease itself as information, availability heuristic AI governance risk shapes how community organisations assess the risks and benefits of AI adoption. When dramatic AI failures are easy to recall from media coverage, those risks feel larger than base rates justify. Conversely, when AI errors in the organisation’s own practice go undocumented and unshared, they are unavailable for recall and therefore feel less likely than they actually are. The result is a systematic distortion of governance decision-making that favours salient, memorable risks over mundane but statistically more significant ones. Availability heuristic AI governance connects to attribute substitution, normalisation of deviance, and security theatre in the NFH Lexicon, and underpins the NFH Behaviour Risk Card framework for nonprofits and community organisations in Australia.
External References
- Tversky, A. and Kahneman, D. (1973). Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5(2), 207-232.
- Schwarz, N., Bless, H., Strack, F., Klumpp, G., Rittenauer-Schatka, H. and Simons, A. (1991). Ease of retrieval as information: Another look at the availability heuristic. Journal of Personality and Social Psychology, 61(2), 195-202.
- Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.
Related Lexicon Entries
From theory to practice
Understanding the risk is the first step.
The NFH Toolkit gives your organisation practical tools to build the internal documentation and review practices that counteract availability distortion and produce governance decisions grounded in actual organisational experience. The Behaviour Risk Cards name this pattern in language your whole team can use.
Not For Humans: Because AI’s greatest risk isn’t technical, it’s Human.