Malicious use of AI, including AI-created biological weapons, could threaten humanity’s survival by 2100, a new Lancet Commission assessment warns. It describes the likelihood as low. The finding is part of a wider assessment of 17 major threats to human health, each selected because there is also a realistic route to action.
The Lancet Commission on 21st Century Global Threats to Health places malicious AI use alongside nuclear conflict involving nuclear famine. The nuclear scenario involves particles reaching the upper atmosphere, blocking sunlight and causing famine. For AI, the extinction pathway it identifies is biological weapons created using the technology, according to the Guardian’s account of the assessment.
Counting lost healthy years
To assess the threats, the commission modeled how many years of healthy life could be lost under current trajectories. Its measure, disability-adjusted life years, or DALYs, combines health lost through illness and disability with years lost through early death. The assessment also modeled best- and worst-case scenarios for each threat, making the potential effect of action part of the exercise.
Beyond the extinction risks, the commission classified threats capable of causing more than one billion cumulative lost healthy years as “major catastrophic threats.” Inclusion required more than a large potential burden: the experts also had to identify possible solutions. The list therefore brings familiar health problems into the same assessment as catastrophic conflict and malicious AI use.
Those other threats include air pollution, antimicrobial resistance, childhood malnutrition, climate breakdown, conventional conflict and declining overseas aid. The assessment also includes high alcohol use, high blood pressure, income inequality, low educational attainment, naturally occurring pandemics, obesity, poor mental health, smoking and unhealthy ageing.
Climate action illustrates the scale of the modeled opportunity. Limiting global heating to about 1.8C rather than allowing it to reach 4.4C could prevent the loss of 1.89 billion healthy years, the assessment estimates. That is a comparison between scenarios in the commission’s model, rather than a measured gain already achieved.
We are not starting from a point of view of ‘oh look, we want to scare you – we’ve got X many threats facing humanity’. We’re starting from the premise that if you are a policymaker today, whether at the global multilateral level or at the national level, these are the potential threats that you could face, and this is the magnitude by which, if you act, you can actually make a difference.
Ibrahim Abubakar, commission member and professor of infectious disease epidemiology at UCL
The burden differs by region
- In higher-income regions, non-communicable conditions—those that do not spread between people—such as obesity and poor mental health were the dominant threats.
- In lower-income regions, the commission found health losses driven by issues such as low educational attainment. Sub-Saharan Africa also bore the brunt of childhood malnutrition.
The international health-aid scenario shows another uneven burden. A collapse or steep reduction in that aid could cause up to 1.5 billion lost healthy years before 2100, the commission estimates. Nearly 70% of that loss would fall in sub-Saharan Africa. Major donors, including the US and UK, have announced large aid-spending cuts over the past year.
Prevention before crisis
The commission says the threats interact and must be addressed together. Its authors urge leaders to act before risks become crises, including by investing in technological innovations and strengthening health systems. Co-chair Christopher Murray also calls for governments to dedicate money and staff to anticipating, preventing and preparing for a more diverse set of threats.
The 35-expert commission is scheduled to share its findings at the World Health Summit in Berlin on October 11. It plans to update the assessment annually, revisiting the 17 threats with new data and incorporating emerging hazards.
Reader comments
Newest comments first. Replies stay oldest first.