AI is among 17 potentially ‘catastrophic threats’ that researchers say require targeted action

Arun Sankar/AFP/Getty Images

More than a dozen “interconnected catastrophic threats” pose significant risk to global health and wellbeing through the remainder of the 21st century, researchers say – including the potential for extinction. But targeted action from world leaders can help avoid worst-case scenarios.

Using data from the Institute for Health Metrics and Evaluation’s Global Burden of Disease study, a commission of 35 leaders from various industries forecast the total years of good health that could be lost by the global population by due to sickness, disability or premature mortality from a range of threats.

Their findings, published in the journal The Lancet , identify 17 threats that pose the greatest danger – but also have the potential for realistic intervention.

The threats include health conditions and behaviors – such obesity, high blood pressure, smoking and mental health – as well as economic conditions, environmental change, technology and conflict.

The commission found that if current trends continue, these threats could collectively account for tens of billions of years of lost health.

“The striking finding is not only the scale of these threats, but that humanity still has the agency to act,” said Dr. Christopher Murray, co-chair of the new commission and director of the University of Washington’s Institute for Health Metrics and Evaluation.

“Across all 17 catastrophic threats, the decisions made by leaders today will shape the health and wellbeing of the next generation. But delaying action will only serve to shift and increase the costs into the future,” Murray said in a statement.

Two threats pose a risk of extinction, the commission found: nuclear conflict and the malicious use of artificial intelligence.

Last month, an initial public offering for the AI company Anthropic said that its models could pose a “ catastrophic or existential risk to humanity ,” and people who worked in the industry have warned that AI could kill the entire human species , perhaps within the decade.

Various forecasters who the new Lancet commission worked with showed “wide divergence” in estimating the risk of AI-related catastrophe, according to the new report.

But the commissioners also note that the emergency of powerful AI systems is an example of “how quickly assumptions about stability can be overturned” – along with evidence from other “recent global shocks,” including the Covid-19 pandemic, the Ebola outbreak ongoing in Africa, and accelerating climate impacts.

The potential for healthy years of life lost due to malicious AI – about 10.4 billion years worldwide by 2100 – is about the same as some other risks including naturally occurring pandemics and obesity.

But because “no malicious AI event of a truly global magnitude” has yet occurred, the estimates for the risk posed by AI are theoretical. The risks posed by obesity, however, are based on more empirical data.

Data shows that overweight and obesity are increasing in every country, and estimates suggest that the global rate could reach 60% of adults and a third of children by 2050. Obesity is related to dozens of other poor health outcomes, particularly associated with increased caloric intake and decreased physical activity, and the commission forecasts a “major global impact of this cluster of behavioral risks.”

The goal of the new commission is not to rank threats but to “catalyse international, regional, and national actions to protect and strengthen the future of health and humanity,” according to a related editorial published in The Lancet.

“Recent global events show us just how quickly stability can give way to crisis, and how costly it is to respond only after disaster strikes. And the global threats to health risk amplifying existing inequalities, with the greatest burden falling on the most vulnerable communities. Yet the future is not inevitable,” ​Dr. Natalia Kanem, former executive director of the United Nations Population Fund and co-chair of the new commission, said in a statement. “Governments and communities must act decisively starting today, to protect health and build a fairer, more peaceful and more secure future for the coming generations.”

Trust is key to meaningful intervention, according to the new report.

It is a “stabilising force, enabling societies to anticipate, mitigate, and adapt,” the commissioners wrote. Distrust, however, can increase potential harm.

Along with trust, seven other factors will “shape how societies respond to threats in the future,” they wrote. They include collaborative global leadership, resilient food systems and efforts to foster more equitable societies.

“Public health is full of examples in which potential threats were known, and protection plans were in place, yet effective action failed,” the new report says. “Achieving meaningful progress will require not only technical solutions but also robust societal and institutional foundations that foster collaboration and accountability.”

Arun Sankar/AFP/Getty Images

研究人员表示，十几种“相互关联的灾难性威胁”将在本世纪剩余时间内对全球健康和福祉构成重大风险，其中包括物种灭绝的可能性。但世界各国领导人采取有针对性的行动有助于避免最糟糕的情况发生。

利用健康指标与评估研究所的全球疾病负担研究数据，由来自各行各业的 35 位领导人组成的委员会预测，由于疾病、残疾或各种威胁造成的过早死亡，全球人口可能会损失多少健康年限。

他们的研究结果发表在《柳叶刀》杂志上，指出了 17 种最具危险性的威胁，但这些威胁也有可能通过实际干预加以解决。

这些威胁包括健康状况和行为——如肥胖、高血压、吸烟和心理健康——以及经济状况、环境变化、技术和冲突。

委员会发现，如果目前的趋势继续下去，这些威胁加起来可能会造成数百亿年的健康损失。

“最令人震惊的发现不仅是这些威胁的规模，而且是人类仍然有能力采取行动，”新委员会联合主席、华盛顿大学健康指标与评估研究所所长克里斯托弗·默里博士说。

“在所有17项灾难性威胁中，领导人今天做出的决定将影响下一代的健康和福祉。但拖延行动只会将成本转移到未来并增加未来，”默里在一份声明中说。

委员会发现，有两种威胁可能导致物种灭绝：核冲突和人工智能的恶意使用。

上个月，人工智能公司 Anthropic 在首次公开募股时表示，其模型可能对人类构成“灾难性或生存性风险”，而业内人士也警告说，人工智能可能会在十年内毁灭整个人类物种。

根据这份新报告，与新成立的《柳叶刀》委员会合作的各种预测者在估计人工智能相关灾难的风险方面表现出“很大的分歧”。

但委员们也指出，强大的人工智能系统的出现就是一个例子，说明“关于稳定性的假设可以多么迅速地被推翻”——其他“近期全球冲击”也提供了证据，包括新冠肺炎疫情、非洲持续爆发的埃博拉疫情以及日益加剧的气候影响。

到 2100 年，恶意人工智能可能造成的健康寿命损失约为 104 亿年，这与其他一些风险（包括自然发生的流行病和肥胖症）大致相当。

但由于“尚未发生真正具有全球规模的恶意人工智能事件”，因此对人工智能风险的评估仍停留在理论层面。而肥胖带来的风险则有更多实证数据支撑。

数据显示，超重和肥胖在每个国家都在增加，据估计，到2050年，全球成年人的超重和肥胖率可能达到60%，儿童的超重和肥胖率可能达到三分之一。肥胖与数十种其他不良健康后果相关，尤其与热量摄入增加和身体活动减少有关，委员会预测“这一系列行为风险将对全球产生重大影响”。

根据《柳叶刀》杂志发表的一篇相关社论，新委员会的目标不是对威胁进行排名，而是“促进国际、区域和国家行动，以保护和加强人类健康和未来”。

“近期全球事件表明，稳定局势可能迅速被危机所取代，而灾难发生后才采取行动的代价又是多么高昂。全球健康威胁有可能加剧现有的不平等，最脆弱的群体将承受最大的负担。然而，未来并非不可避免，”联合国人口基金前执行主任、新委员会联合主席纳塔莉亚·卡内姆博士在一份声明中表示。“各国政府和社区必须从今天开始采取果断行动，保护民众健康，为子孙后代建设一个更加公平、和平、安全的未来。”

最新报告指出，信任是有效干预的关键。

委员会成员写道，信任是一种“稳定力量，使社会能够预见、缓解和适应”。然而，不信任可能会加剧潜在的危害。

他们写道，除了信任之外，还有其他七个因素将“影响未来社会应对威胁的方式”。这些因素包括全球协作领导力、具有韧性的粮食系统以及促进社会更加公平的努力。

这份新报告指出：“公共卫生领域充斥着这样的例子：潜在威胁早已为人所知，防护计划也已制定，但有效的应对措施却未能奏效。要取得实质性进展，不仅需要技术解决方案，还需要强大的社会和制度基础，以促进合作和问责。”