Quick Overview

  • A study based on estimates from 272 global AI experts indicates that by 2030, there is a 10% probability that uncontrolled or weaponized AI could cause 1 million deaths, at least $100 billion in losses, or other "catastrophic harms," with the financial industry being particularly vulnerable.
  • The study, conducted jointly by MIT FutureTech and the University of Queensland, points out that AI deployed in the financial sector could trigger large-scale losses by enhancing capabilities such as market manipulation, fraud, cyberattacks, computer sabotage, and privacy violations.
  • The researchers stated: "Even with pragmatic mitigation measures, dangerous capabilities, weapons and cyberattacks, concentration of power, inequality and unemployment, and environmental harms still have a greater than 10% probability of leading to catastrophic outcomes." Even with mitigation efforts, "experts still believe that across 24 'risk areas,' the probability of catastrophic outcomes exceeds 5%."

In-Depth Analysis

These findings from MIT and the University of Queensland echo other recent studies that have raised alarms about the risks AI poses to humanity, sparking public opposition and prompting initial regulatory actions.

For example, in the 2022 "Existential Risk Persuasion Tournament," AI experts estimated a 3% probability that AI could cause human extinction.

According to the MIT/Queensland study, the biggest risk of AI is that by 2030 it could acquire capabilities such as large-scale deception, weapons development, public manipulation, political abuse, and cyberattacks.

The study also notes that other major risks should not be overlooked, including the possibility that in an "AI race," countries or companies might rush to deploy systems without adequately checking for errors or security threats to gain an advantage.

Malicious actors could use AI to launch cyberattacks or create chemical, biological, radiological, or nuclear weapons, the study says.

AI could concentrate power and resources, leading to unequal distribution of benefits and social inequality, or it could inadvertently generate false or deceptive information that distorts people's perceptions and causes physical, emotional, or material harm, the study points out.

The researchers said that efforts to reduce AI risks face an obstacle—AI users and stakeholders are more vulnerable to various threats than AI developers, regulators, legislators, and standard-setters who have the power to reduce risks.

"Providers of computing, cloud, and data infrastructure are largely insulated from many of AI's risks," they said.

This vulnerability gap creates "a misalignment of incentives in addressing the most important AI risks," the study says.

"Even well-understood risks can persist when responsibility is dispersed, mismatched with capabilities, or disconnected from those who bear the harm," the researchers said.

Furthermore, "commentators note that the emerging regulatory landscape looks more like a patchwork than a coordinated framework, and it remains unclear which risks governments should prioritize," they added.