From chatbot to agent
AI systems are beginning to use computers and tools to complete sequences of actions. On OSWorld, agent performance rose from roughly 12% to 66.3% in a year, yet agents still fail about one in three structured attempts.
Room 9, trends, scenarios and uncertainty
The future of artificial intelligence
This room does not treat forecasts as facts. It separates what is already happening, strong technological trends and futures that remain genuinely uncertain.
Open the interactive room →Exhibits in this room
AI gets a body
Robotics is increasingly connected with vision-language models. Yet the gap between a controlled lab and an unpredictable home remains large: the 2026 AI Index reports only 12% success on real household tasks.
Laboratories that choose the next experiment
Self-driving laboratories combine AI, robotics and automation: a system proposes an experiment, runs it and uses the result to choose the next step. Today they mostly operate in narrow, well-defined domains.
Medicine and digital models of patients
AI is already used in diagnosis, drug discovery and research on medical digital twins. Early results are promising, but clinical reliability and generalisation require rigorous evaluation.
The classroom with personal AI
Generative AI is already widespread in study and education. The question is shifting from whether students will use it to what humans should learn when machines can write, solve and explain.
Work changes tasks before it changes occupations
The ILO estimates that one in four workers is in an occupation with some GenAI exposure. In the near term, task transformation is more likely than the automatic disappearance of whole occupations.
Smaller, local and specialised models
Progress is not only about scaling up. Smaller and specialised models can be highly competitive on some tasks, while local execution can reduce cost, latency or data exposure.
The invisible factory behind AI
AI depends on physical infrastructure: chips, data centres, cooling and electricity. In the IEA base case, global data-centre electricity use roughly doubles to around 945 TWh by 2030.
More AI, more rules
As AI enters higher-stakes decisions, requirements for transparency, evaluation, provenance and accountability grow. In Europe the AI Act is being phased in over several years.
AGI: the great uncertainty
There is no scientific consensus on whether or when artificial general intelligence will exist. Current systems show increasingly autonomous capabilities, but that does not establish an inevitable path to AGI or loss of control.