Search
SLOW AI: Promoting the Mindful, Respectful, and Fair Use of Generative AI
Large Language Models (LLMs) such as ChatGPT are increasingly being used by the public as a source of health information.
Slow AI is an initiative developed in collaboration with Slow Medicine to promote a thoughtful approach to the use of artificial intelligence in healthcare. It provides guidance for both the public and healthcare professionals, helping to encourage appropriate use while reducing the risk of misuse.
The project was carried out by a multidisciplinary team of experts from Zadig, the Italian Society of Artificial Intelligence in Medicine (SIIAM), the Italian Group for Evidence-Based Dermatology (GISED), the Mario Negri Institute for Pharmacological Research, the University of Foggia, the Turin Medical Association (OMCeO Torino), and Slow Medicine.
The research was organised into three complementary studies, designed to evaluate:
- how accurately a generative AI system provides information consistent with clinical practice guidelines;
- how reliably it answers health questions from members of the public;
- how people use generative AI to seek medical information.
The findings provided valuable evidence on both the opportunities and the risks associated with generative AI in healthcare. They also informed a broader reflection on what a mindful, respectful, and fair use of AI should look like when responding to people’s health-related questions.
The Slow AI Decalogue
One of the project’s outcomes is a concise set of ten recommendations designed to support the informed and responsible use of generative AI for health information.
This is the short version.
- AI-generated health information may seem reliable, but it isn’t always.
- AI answers are generated through statistical prediction rather than medical understanding.
- Although AI responses are usually clear and easy to read, they are often too general and may contain errors in diagnosis or treatment recommendations.
- AI responses are usually consistent with clinical guidelines, but they may not reflect the latest evidence.
- AI systems can produce “hallucinations”: plausible-sounding statements that are entirely unfounded.
- AI-generated text is in excellent Italian and often indistinguishable from human writing, making it difficult to recognise when information has been produced by a machine.
- Sharing your personal health information with an AI system in the hope of obtaining a diagnosis or treatment may put both your privacy and your health at risk.
- Asking AI to recommend a treatment for your condition may result in inappropriate advice because it cannot take your individual clinical circumstances into account.
- Asking AI to interpret the results of a medical test is risky because those results need to be considered alongside your medical history, symptoms, and overall clinical picture.
- Use AI with caution and healthy scepticism. Its answers are general in nature and cannot replace personalised medical advice from a qualified healthcare professional.
A more detailed version of the Slow AI Decalogue is available on the websites of Zadig and Slow Medicine.