After five modules dedicated to AI's risks and problems, it is time to explore the other side of the coin: applications where artificial intelligence produces concrete and measurable benefits for society. Not to fall into naive technological optimism, but to understand which conditions make it possible to use AI in service of the common good — and how to replicate them.
Climate and environment. AI is a powerful tool for addressing the climate crisis. Global Forest Watch uses satellite imagery and machine learning to monitor deforestation in near real time, allowing indigenous communities and activists to intervene before destruction is irreversible. DeepMind (Google) has developed models that predict wind energy production 36 hours in advance, enabling more efficient electricity grid management. ClimateTrace, co-founded by Al Gore, uses AI to monitor greenhouse gas emissions from over 70,000 facilities worldwide, making it impossible for companies to hide their impact.
But there is a paradox: AI itself has an enormous environmental footprint. Training GPT-4 consumed an estimated amount of energy equivalent to that of 1,000 American households for a year. Data centres use billions of litres of water for cooling. AI for climate must reckon with its own footprint — a paradox that requires transparency and conscious choices about when AI is the right solution and when it is not.
Health. AI has demonstrated remarkable capabilities in medical diagnosis. Google's system for diabetic retinopathy detects the disease with accuracy equal to or exceeding that of specialists, potentially enabling mass screening in areas with doctor shortages. AlphaFold (DeepMind) has predicted the 3D structure of nearly all known proteins — a result that is accelerating pharmaceutical research worldwide. Aidoc uses AI to identify brain haemorrhages in CT scans, reducing intervention times.
But here too the challenges are real. Medical AI trained on data from wealthy, white populations works worse for Global South populations. Diagnostic systems can create a false sense of certainty, leading doctors and patients to over-trust algorithmic output. And access to these technologies is profoundly unequal: those who benefit first are the countries and communities that would need them least.
Education. AI can personalise learning, adapting content and pace to each student's needs. Khan Academy has integrated an AI tutor (Khanmigo) that helps students reason without giving direct answers. Duolingo uses AI to adapt language lessons to each user's level. But the risk is replacing the human teacher-student relationship with an algorithmic interaction, and massive data collection on minors without adequate protections.