p(doom)
p(doom) denotes the subjectively estimated probability of a catastrophic outcome caused by advanced artificial intelligence. The value expresses personal conviction and varies among experts from near zero to clearly double-digit percentages.
p(doom) is an individually estimated probability that advanced AI leads to a catastrophe for humanity — a subjective belief expressed as a percentage, not a measured value.
Explanation
The notation comes from probability theory, but the underlying event is not defined consistently: some mean human extinction, others the permanent loss of human control, others any irreversible civilisational decline. The horizon usually stays implicit too. As a result two figures are not comparable even when the percentage is identical. The value is a credence — a willingness to bet — not a frequency measurement, and it is not falsifiable. Inside companies the number therefore often does harm: it creates false precision, blends existential with operational risk and shifts responsibility from concrete processes to global scenarios. The operationally useful translation is p(fail): how often a defined AI-supported workflow breaks, under which failure definition and at what cost. That question is measurable, loggable and boundable.
Marketing Relevance
For marketing and technology leaders p(doom) is primarily a communication topic: the term surfaces in board, client and press conversations and requires a defensible stance. What is defensible is that no expert consensus exists and estimates diverge widely. Decisions about AI deployment instead depend on per-use-case error rates, approval chains and logging.
Example
A board asks about a press statement on p(doom). The team does not answer with a number of its own but names the spread of public estimates and switches to its own level: the customer chat is reviewed for false commitments over a week, the rate is documented and sign-off becomes mandatory for binding statements.
Common Pitfalls
Presenting a single percentage as "the state of the science" exceeds what sources support. Equally problematic are comparisons of values without a clarified event definition and horizon, and using p(doom) as a substitute for concrete risk analysis in your own operations.
Origin & History
The expression grew out of AI safety research forums and became widely visible from 2023, when leading experts stated estimates publicly. The Center for AI Safety's Statement on AI Risk (May 2023) deliberately avoided any number and only called for prioritising risk reduction. The AI Impacts survey of AI researchers (2023) showed a median in the low single-digit percentages for extremely bad outcomes, with wide spread. Since then p(doom) has served mainly as a position marker in debates, not as a research finding.
Comparisons & Differences
p(doom) vs. AI Alignment
Alignment describes the technical work of tying system behaviour to human intent. p(doom) is merely an estimate of how likely that work is to fail.
p(doom) vs. KI-Risikomanagement
Risk management works with defined errors, frequencies and controls inside concrete processes. p(doom) names a global, unverifiable scenario with no operational implication.
Marketing Use Cases
Performance marketing teams use p(doom) to generate campaign concepts faster and roll out A/B tests in hours instead of weeks.
Content teams deploy p(doom) to accelerate editorial pipelines — from research and outline through to multilingual localization.
In customer support, p(doom) powers intelligent chatbots that resolve Tier-1 tickets automatically, cutting ticket volume by 40–60%.
Analytics and insights teams combine p(doom) with BI dashboards to interpret large datasets in real time and surface proactive recommendations.
Product and innovation teams prototype new features with p(doom) without locking up deep engineering resources.
Compliance and legal teams apply p(doom) to automatically check contracts, briefings and marketing assets against regulations like the EU AI Act.
Frequently Asked Questions
Is there a scientifically valid p(doom) value?
No. No consensus value exists and no method could ground such a figure empirically. Public statements by prominent experts range from near zero to clearly double-digit percentages, and even broad surveys show wide spread rather than a defensible average.
Why is p(doom) unsuitable for corporate decisions?
Because event definition, horizon and data basis are missing. A percentage without those components cannot be translated into action. What is usable instead is the failure probability of a specific workflow, because errors there can be defined, measured and bounded.
How should you answer when asked for your own p(doom)?
Best with the actual situation rather than a number: clarify the event, name the spread of expert opinion and switch to the operational level. There you can show which errors your system can make, how often that happens and who signs off on such cases.
Related Services
Go deeper: Agentic AI Hub · Model comparison 2026