Artificial intelligence is becoming part of the conditions under which people seek care, form relationships, obtain information, and make health-relevant decisions. Yet AI systems are not neutral public health tools. They are often embedded in commercial infrastructures that optimise engagement, enable targeted persuasion, extract intimate data, and simulate care without the duties, competencies, or accountability expected of human carers. These risks are especially pronounced in AI companionship, mental health support, targeted advertising, and emotionally responsive systems used by young, lonely, distressed, or otherwise vulnerable people.
This keynote argues that AI should be understood as an emerging public health infrastructure. At present, its dominant design and business logics create significant risks. Harmful marketing may enter private conversational spaces with renewed persuasive force. AI companions may intensify harmful ideation, foster over-reliance, or displace help-seeking. Weak governance may also leave early warning signals invisible until harms accumulate and become difficult to reverse.
However, AI does not need to develop along this path. Drawing on moral philosophy and information systems research, the keynote proposes compassion as a foundational ethical principle for aligning AI with societal wellbeing, the alleviation of suffering, and the common good. Compassionate AI begins by asking who is vulnerable, whose suffering remains hidden, and which harms are avoidable. It requires design that protects dignity and autonomy, governance that treats critical AI systems as public-interest infrastructure, and use that strengthens rather than substitutes for human care.
The keynote concludes by outlining a public health agenda for AI grounded in systems thinking, policy innovation, independent monitoring, and solidarity across health, technology, government, civil society, and affected communities.