Nearly One in Five AI Researchers Expected an Extinction Scenario Back in 2024
A significant minority of AI researchers already anticipated a catastrophic outcome from advanced artificial intelligence, according to survey data from 2024.
The finding, drawn from a large-scale poll of machine learning professionals, reveals that 19 percent believed there was a real possibility that superintelligent AI could drive humanity to extinction.
This number, reported as a 5 percent or higher chance of human extinction from unaligned AI, was not just a fringe view held by doomsayers.
It represented nearly one in five researchers actively working in the field, a remarkably high proportion for such a severe prediction.
Why This Survey Matters
The survey was conducted by researchers at the AI Impacts initiative and involved over 2,700 AI researchers who had published work at major conferences.
Participants were asked to estimate the probability of various outcomes, including existential catastrophe, stemming from the development of superintelligent AI.
The 5 percent threshold is a critical benchmark. It is the same probability used in historical risk assessments for nuclear war, asteroid impacts, and other civilizational-scale threats.
The fact that a substantial body of experts assigned a similar or higher risk to AI is a data point that cannot be easily dismissed.
It suggests that the “AI extinction” scenario is not a niche concern but a mainstream worry among the very people building these systems.
Scope of the Survey
The survey included researchers from both academia and industry, offering a broad cross-section of the AI community.
- Publication record: All respondents had at least one paper accepted at a top-tier AI conference, such as NeurIPS or ICML.
- Geographic diversity: Respondents came from leading institutions around the world, including North America, Europe, and Asia.
- Timeframe: The survey was conducted in late 2022, but the findings were published and widely discussed in 2024, making the data still highly relevant to current debates.
These details are crucial because they show that the concern is not limited to a small group of computational ethicists or technology critics.
It is held by the same researchers who are publishing the algorithms, building the datasets, and training the models that power today’s most advanced AI systems.
What the 19 Percent Represents
To put that number in perspective, 19 percent is nearly double the proportion of researchers who said the same thing in a similar survey conducted just a year earlier.
This upward trend suggests that concerns about extinction-level AI risk are growing, not shrinking, as large language models and other generative AI tools become more powerful.
The survey also asked about other outcomes, such as “extreme negative outcomes” short of extinction.
When combining all pessimistic scenarios, the median researcher estimated a 40 percent chance that AI would lead to outcomes significantly worse than most humans expect.
This means that for a large segment of the field, the “worst-case scenario” is not a science fiction fantasy but a plausible, if not probable, outcome that must be actively mitigated.
Implications for Today’s AI Development
The finding has direct implications for how we approach AI safety and regulation.
If the people building these systems think there is a 1 in 5 chance of catastrophe, then the current pace of unregulated deployment is a massive gamble.
This is not an argument for halting AI research; it is an argument for devoting significantly more resources to alignment research, interpretability, and fail-safe mechanisms.
It also strengthens the case for international AI safety agreements, similar to the treaties that govern nuclear weapons or biological agents.
While the survey does not prove that extinction is likely, it proves that the risk is real enough to warrant serious, coordinated action.
Ignoring the views of 19 percent of your own field is a dangerous form of groupthink, especially when the downside is the end of human civilization.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.