Debunking the 10% AI threat claim

Debunking the 10% AI Threat Claim

I recently wrote an article on “AI doomsday or promising future”, where I tried to balance the risks and the opportunities of AI. During my research, I found plenty of references to a claim that “half of AI researchers believe that there’s a 10% chance AI will kill us all”.

This and other threats are commonly raised. The claim is from a 2022 survey. But is the claim a fair conclusion of the study? Before losing sleep, let’s look at the facts behind this claim.

The 2022 Expert Survey on Progress in AI, which forms the basis of this claim, had a response rate of 17% from the 4271 people who published papers at NeurIPS and ICML conferences in 2021. However, only 162 of these respondents answered the question of AI causing human extinction. The median probability of these 162 responses was 10%, with 81 people estimating the probability as 10% or higher.

Given these facts, it’s worth asking if the results support the media claims. The survey question is vague, lacking a specific time period for the probability estimation. The sample size of 162 is small and could suffer from response bias. We also don’t have information on the expertise levels of the respondents or their confidence in their estimates.

In conclusion, the 10% claim in the article is not well-supported and should not be taken at face value. However, this doesn’t mean there aren’t any risks associated with AI. It’s crucial to continue researching AI safety and potential consequences, but let’s be cautious of sensationalism and strive to base our concerns on well-founded evidence.

It’s important to have nuanced and informed discussions about the potential implications of AI technology. Misrepresenting experts’ opinions in the field can lead to unnecessary panic and hinder the development of responsible AI.

Read the full article by Melanie Mitchell here:

Was this article helpful?

Leave a Reply