A FOOD 4 THOUGHTS RESEARCH SERIES
AI Futures is a ten-part research series about changes already taking shape around artificial intelligence. It looks beyond the model: at the laboratories that verify discoveries, the factories that generate physical data, the infrastructure that powers computation, and the ways AI is reshaping commerce, memory, work and relationships.
THE EDITORIAL QUESTION
What are we missing while AI becomes ordinary?
Each paper separates reported evidence from interpretation, draws on primary sources where available, and closes with a question for organisations and decision-makers.01 · AI FUTURES · 2026-08-03
From generating hypotheses to verifying scientific claims
Multi-agent systems are beginning to formulate scientific hypotheses and propose experiments. Papers published in 2026 show real acceleration, but also the decisive limit: every result still requires independent experimental verification and critical reading of the data.
READ THE STUDY ↗02 · AI FUTURES · 2026-08-10
World models, simulation and the value of consequence data
World models aim to represent environments and transitions, moving beyond next-word prediction. Systems grouped under this label today are diverse: generative simulators, synthetic-data platforms and predictive models are not interchangeable. For businesses, the question is which consequences of their decisions they can measure and retain.
READ THE STUDY ↗03 · AI FUTURES · 2026-08-17
Robots, factories and the geography of industrial data
The convergence of robotics, sensors and AI models is bringing intelligence into physical processes. The latest IFR data show a global stock of around five million industrial robots and China’s growing share of installations and domestic production. These figures indicate industrial scale, but they do not directly measure control or quality of the data used to train AI systems.
READ THE STUDY ↗04 · AI FUTURES · 2026-08-24
Europe’s AI Gigafactories and the layers of technological sovereignty
The EU call for AI Gigafactories aims to expand computing capacity within the Union, but technological sovereignty is about more than hosting data centres. This paper separates infrastructure, chips, energy, software, data and skills, and considers what the new investment can and cannot guarantee.
READ THE STUDY ↗05 · AI FUTURES · 2026-09-01
Data centres, grids and the material cost of digital intelligence
According to the International Energy Agency, data-centre electricity consumption could more than double by 2030, reaching around 945 TWh in its base case. The constraint is not only how much energy is generated: grid connections, construction times and geographic concentration can slow projects and make the local costs of computing visible.
READ THE STUDY ↗06 · AI FUTURES · 2026-09-08
Synthetic images, memory and trust in accounts of the past
A preregistered MIT Media Lab study of 200 participants found that AI-altered images and videos can increase false memories. The strongest effect appeared when participants saw AI-generated videos based on edited images. The result is specific to the experimental design, but it raises questions about archives, testimony and content provenance.
READ THE STUDY ↗07 · AI FUTURES · 2026-09-15
AI agents, product discovery and the brand’s new territory
AI-referred traffic to US retail sites grew rapidly in 2026, and Adobe reports that visitors from AI sources converted better than average in March. This signals a change in discovery, not proof that consumers have stopped browsing. Platforms are also recalibrating checkout: control over recommendation and control over the transaction remain distinct questions.
READ THE STUDY ↗08 · AI FUTURES · 2026-09-22
Authenticity, provenance and the value of human work
A Columbia Business School study found that, in specific experimental conditions, artworks labelled as AI-generated received monetary valuations 62% lower than those presented as human-made. The premium for human work emerged mainly when artworks were compared with AI images. This concerns perceptions of art in a study setting; it does not directly establish a market premium for luxury goods.
READ THE STUDY ↗09 · AI FUTURES · 2026-09-29
AI, entry-level work and the transmission of knowledge
A Stanford analysis of ADP administrative data through June 2026 finds that employment for US workers aged 22–25 in AI-exposed occupations is 19% below an author-constructed counterfactual benchmark. The gap is associated primarily with weaker hiring, but the authors stress that these are descriptive findings, not a causal estimate of AI’s effect.
READ THE STUDY ↗10 · AI FUTURES · 2026-10-06
Teenagers, conversational relationships and design responsibility
A 2025 US Common Sense Media survey of 1,060 teenagers aged 13–17 found widespread use of AI companions and, for some respondents, conversations perceived as as satisfying as those with real friends. The survey reports behaviours and concerns; it does not prove that chatbots cause dependency or replace human relationships.
READ THE STUDY ↗