- Leah PearsonPh.D. 5th year1 2,
- Bruce Tsairesearch analyst3
- 1Harvard TH Chan School of Public Health Department of Global Health and Population
- 2Harvard-MIT PhD, Harvard Medical School
- 3Rethink your priorities
Advances in artificial intelligence (AI) have the potential to transform healthcare by creating new treatments, improving diagnosis, making treatments more accessible, reducing costs, and reducing the workload for clinicians. conceals sexuality. 1 These technologies have the potential to help people live healthier and longer, but many doctors and AI researchers say AI also poses health risks. 23 The challenge is to ensure that algorithms “capture our norms and values, understand our meaning and intentions, and above all, do what we want” . This is called the coordination problem. 4 The risks associated with uncoordinated AI, where system behavior is inconsistent with the purposes and principles of human creators and users, are a growing threat to public health, and the medical community has no right to do so. can and should. corresponds to
Algorithmic misalignment is already putting millions of people’s health at risk. For example, the goal of one commercial algorithm used throughout the US healthcare system was to identify patients who would benefit from additional care. However, the algorithm uses health care costs as a measure of medical need, resulting in a preference for white patients over sicker black patients, many of whom face significant barriers to access treatment. was 56 Such missteps have raised concerns from various stakeholders in the medical community. 7
Outside the healthcare system, algorithms designed to improve health can also have the opposite effect. In 2021, Mark Zuckerberg announced Facebook’s goal to expedite the deployment of a COVID-19 vaccine by promoting vaccine-related content from groups such as the World Health Organization. But Facebook posts from official sources were often inundated with critical comments, including conspiracy theories and misinformation. In an effort to promote vaccine-promoting content, Facebook’s algorithm eventually forced users to view anti-vaccine comments as many as 775 million times per day, potentially discouraging people from taking vaccines. It is reported that there are8.
AI is already negatively impacting people’s health in tangible ways, but as algorithms become more powerful and complex, and their scope and pervasiveness increase, experts are also raising more speculative concerns. I’m here. A recent statement by the Center for AI Safety states that “mitigating the risk of AI-induced extinction should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” Signed by pioneers, hundreds of AI researchers, and the CEO of the AI Safety Center. This includes major AI companies, government officials, and even global health leaders, medical professors, and bioethicists. 9
AI development is advancing rapidly, with billions of dollars pouring into companies aiming to develop artificial general intelligence (AGI), or machine intelligence that surpasses human capabilities. 1011121314 Public health is often an important issue for private actors on the edge of AGI research. , is at most a secondary consideration. Many experts believe AGI is possible for him, but even if it isn’t, the algorithms these companies develop along the way could cause a lot of damage. 15
Role of the medical community
The growing risk of misaligned AI concerns the medical community for two reasons. First, the potential health benefits of AI are often used to justify certain development avenues. For example, some people working to build AGI suggest that their technology could improve access to healthcare16. But with every benefit comes risk, and the medical community is well-positioned to weigh them when it comes to advancing certain AI capabilities. In fact, the medical community may have a particularly important role, as computer scientists sometimes exaggerate the medical benefits that come with advances in AI. For example, one prominent computer scientist argued in 2016 that we should stop training radiologists because they will soon be replaced by AI. This prediction is well out of date 1718.
Second, any problem that could harm the health of millions of people and that experts fear could be catastrophic is, by definition, a public health risk. The nature of theoretical future threats is difficult to pinpoint, but AI already has the power to cause great harm, even in the hands of well-intentioned agents. Researchers recently generated 40,000 potential chemicals instead using an AI model designed to aid drug discovery. 19 Even greater risks can arise when powerful algorithms are misaligned with users’ values, or fall into the hands of nefarious agents.
As others have recently noted, physicians have a successful history of shaping discourse about transformative technologies that pose public health risks. 20 Doctors played a key role in the formation of Nobel Peace Prize winning organizations such as the International Physicians for the Prevention of Nuclear War (IPPNW). ) and the international campaign to abolish nuclear war. Both organizations have played pivotal roles in the global effort to ban nuclear testing and establish an international treaty for nuclear disarmament21. The parallels between achieving nuclear capability and advanced AI are clear, both dual-use technologies with the potential for significant benefit and harm. An important technology that defines the century. 22 IPPNW had data from the atomic bombings of Hiroshima and Nagasaki to support its advocacy efforts, but wait for a similar catastrophe to occur at a time when AI experts are already sounding the alarm. would be wrong.
To mitigate health risks from AI, the medical community should advocate for regulations endorsed by AI experts and consistent with those governing conventional science, medicine, and public health. Such regulations may include the development of safety standards such as pre-implementation reporting, public incident reporting, external review processes, licensing requirements, and clear liability rules23. These measures can help ensure AI systems are predictable and interpretable. 2425
More generally, as governments consider AI policy proposals, the medical community needs to ensure that the risks posed by AI miscoordination are viewed from a public health perspective. The drive to use AI in new capabilities and to develop its capabilities must always be appropriately tempered by understanding the impact of AI on population health and health equity.
Finally, just as physicians prescribe only drugs that regulators have determined are safe, healthcare systems, scientific laboratories, and other medical practitioners should be evaluated and adequately tested against misalignment and misuse. Only AI models that have been deemed robust by industry should be employed26. It helps create ethical codes and market forces that encourage adherence to safety standards.
The potential of AI in medicine is undeniable, but so are the dangers of uncoordinated AI. By advocating for policies that address the growing risks posed by inappropriate AI, the medical community will help ensure that these transformative technologies are used to build a healthier and safer world. can do.
footnote
-
Competing Interests: LP receives research grants from the Effective Altruism Funds, which are supported by Open Philanthropy. BT is a global health researcher at the Open Philanthropy-supported nonprofit Rethink Priorities, but the opinions expressed in this manuscript are the author’s own. Open Philanthropy has separately funded her AI safety-related efforts, but neither LP nor she received any funding related to this.
-
Provenance: Not commissioned. Not externally peer-reviewed.
