This page is in an early draft. It is very incomplete and may contain errors.
Some people who are working in AI Safety have published quantitative estimates for how likely they think it is that AI will pose an existential threat.
This page draws heavily from this database made by Michael Aird at Convergence Analysis.
The table below includes estimates from individuals working in AI Safety of how likely very bad outcomes due to AI are.
Many of the individuals expressed Knightian uncertainty when making their estimates, saying that their probability varies day-to-day, or that the estimate is currently in development, or that this is a very quick-and-dirty estimate. People who have explicitly said something like this include Katja Grace, Joseph Carlsmith, Peter Wildeford, Nate Soares, Paul Christiano, and others. These estimates should not be treated as definitive statements of these individuals' beliefs, but rather as glimpses of their thinking at that moment.
Each estimate includes:
The estimates are in no particular order. The table can be sorted by clicking at the top of each column.
|Estimator||Date||What is Estimated?||Probability||Source||Most Recent?|
|Katja Grace||2023||Bad future because AI agents with bad goals control cognitive labor||0.19||Will AI end everything? A guide to guessing||Yes|
|Joseph Carlsmith||2021||Existential catastrophe by 2070 from advanced, planning, strategic AI||0.05||Is Power-Seeking AI an Existential Risk?||No|
|Joseph Carlsmith||2022||Existential catastrophe by 2070 from advanced, planning, strategic AI||0.1+ (Greater than 10%)||Update to: Is Power-Seeking AI an Existential Risk?||Yes|
|Nate Soares||2021||Existential catastrophe by 2070 from advanced, planning, strategic AI||0.77||Comments on Carlsmith's "Is power-seeking AI an existential risk?"||Yes|
|Toby Ord||2020||Existential catastrophe by 2120 as a result of unaligned AI||0.1||The Precipice||Yes|
|Toby Ord||2020||Humanity does not survive and is in charge of our future, if something that is more intelligent than us is created this century||0.2||Toby Ord on the precipice and humanity's potential futures||No|
|Eliezer Yudkowsky||2022||AGI "killing literally everyone"||~1||AGI Ruin: A List of Lethalities||Yes|
|Rohin Shah||2019||Things with AI do not go well, without additional intervention by long-termist community doing safety research||0.1||Conversation with Rohin Shah||No|
|Paul Christiano||2019||How much worse the future is in expectation by virtue of our failure to align AI||0.1||Conversation with Paul Christiano||Yes|
|Peter Wildeford||2023||X-risk, including several scenarios||0.22||Slack channel & private conversation||Yes|
|Adam Gleave||2019||Chance that AI does cause a significant risk of harm, without intervention from AI safety efforts||0.6 - 0.7||Conversation with Adam Gleave||No|
|Adam Gleave||2019||Chance that AI does cause a significant risk of harm, with median AI safety efforts||0.3 - 0.4||Conversation with Adam Gleave||Yes|
|Adam Gleave||2019||Chance that AI does cause a significant risk of harm, with best case AI safety efforts||0.1 - 0.2||Conversation with Adam Gleave||No|
|Rohin Shah||2020||Probability of AI-induced existential risk||0.05||AI Alignment Podcast: An Overview of Technical AI Alignment in 2018 and 2019 with Buck Shlegeris and Rohin Shah||Yes|
|Buck Shlegeris||2020||Probability of AI-induced existential risk||0.5||AI Alignment Podcast: An Overview of Technical AI Alignment in 2018 and 2019 with Buck Shlegeris and Rohin Shah||No|
|James Fodor||2020||Unaligned AI usurps and establishes permanent dominance over humanity||0.0005||Critical Review of 'The Precipice': A Reassessment of the Risks of AI and Pandemics||Yes|
|Buck Shlegeris||2023||Likelihood of AI coup||0.25||The current alignment plan, and how we might improve it||Yes|
|Stuart Armstrong||2014||Probability of humanity's non-survival in the context of artificial superintelligence||0.33 - 0.5||The future is going to be wonderful if we don't get whacked||No|
|Stuart Armstrong||2020||Whether AGI could threaten humanity's survival or permanently curtail its potential||0.05 - 0.3||Is AI an existential threat? We don't know, and we should work on it||Yes|
|Rohin Shah||2021||Chance of human extinction if literally no one tries to address problems with AI||0.33 - 0.7||Rohin Shah on the State of AGI Safety Research in 2021||No|
|Eli Lifland||2022||Misaligned takeover this century||0.35||My take on What We Owe the Future||Yes|
|Katja Grace||2022||AI destroys the world||0.07||Katja Grace on Slowing Down AI and Surveys||No|
|Andrew Critch||2023||Humanity not surviving the next 50 years||0.8||My May 2023 priorities for AI x-safety: more empathy, more unification of concerns, and less vilification of OpenAI||Yes|
|Andrew Critch||2023||Humanity not surviving the next 50 years, without a major international regulatory effort to control how AI is used||0.9+||My May 2023 priorities for AI x-safety: more empathy, more unification of concerns, and less vilification of OpenAI||No|
|Scott Aaronson||2023||The generative AI race, which started in earnest around 2016 or 2017 with the founding of OpenAI, to play a central causal role in the extinction of humanity||0.02||Why am I not terrified of AI?||Yes|
Different people use different framings to arrive at their estimate of AI risk. The most common framing seems to be to describe a model of what the risk from advanced AI looks like, assign probabilities to various components of that model, and then calculate the existential risk from AI on the basis of this model. Another framing is to describe various scenarios for the future of AI, assign probabilities to the various scenarios, and then add together the probabilities of the different scenarios to determine the total existential risk from AI. There are also some people who give a probability without describing what framing they used to get this number.
Below is an example of each of these two framings, due to Joseph Carlsmith and Peter Wildeford, respectively. Both individuals have updated their estimates since publishing their framing, so neither probability breakdown reflects the author's most recent estimate of AI risk. They are included to show how these framings work.
One example of using a model to calculate the existential risk from AI is due to Joseph Carlsmith. He calculates AI-risk by 2070 by breaking it down in the following way:
The total AI risk is the product of the probabilities for each part of the model.
This same model has been used by Nate Soares and Eli Lifland to calculate their estimates of AI risk. Several other people in the table have also used similar models.
One example of describing possible scenarios to calculate the x-risk from AI is due to Peter Wildeford. He calculates x-risk from AI by 2100 by breaking it down in the following way:
The total AI risk is the sum of (3), (5), (6), (7), & (8).
Primary author: Jeffrey Heninger