Reed reports that he attended Cannes Lions, the "festival of creativity," for the first time this year and that Google DeepMind CEO Demis Hassabis wanted to discuss creativity. Reed noted that where AI meets the creative industries the subject is fraught with emotion and legal disputes; while some attendees treated AI as a villain last year, this year they were warmer but remained cautious.
Drawing a moral line between uses of AI
Many skeptics draw a clear moral distinction: AI that aids cancer research or finds climate-friendly materials is good; AI that creates music, films, or art is not. After speaking with Hassabis, Reed says he concluded the technology itself does not naturally make that separation.
The "Einstein test" and a missing kind of creativity
Hassabis argued that for AI to truly transform science it needs a form of creativity that current models lack. Today’s systems can generate images or outputs no one has seen before, but true creativity, he says, would be a system that can formulate a genuinely new concept rather than merely extrapolating from existing human knowledge. Reed reports Hassabis calling this the "Einstein test": given only the evidence available to Einstein, could today’s AI models rediscover relativity? Probably not, Reed conveys.
Physical understanding and multimodal training
To find the missing algorithmic piece, Hassabis suggested AI may need an understanding of the physical world in a way that text-based systems do not. That implies training models on human-made visual and auditory creations. He connects this idea to his earlier neuroscience work: a 2007 paper studying patients with hippocampal damage found that memory and imagination share a constructive mechanism — one reconstructs the past, the other recombines its elements into something new.
Game design, simulations, and scientific applications
Hassabis said he designed games by running simulations in his head: before writing code he would picture a child using the interface, anticipate where they might struggle, and imagine what would make the game fun. At Cannes he summarized succinctly: "Imagination is a type of simulation." Simulations let an intelligence test many possibilities before choosing a path.
DeepMind’s early work on games served as a research ladder toward systems able to tackle protein folding and drug discovery. Reed suggests video models might follow a similar trajectory: capabilities that help generate an advertisement could also help an AI understand the physical world, train a robot, or analyze images of cells and molecules. As Hassabis put it, "some of these things are inseparable."
Costs, beneficiaries and the ethical debate
Technological progress rarely arrives in neat, separable pieces. Who benefits and who pays for AI progress is a legitimate debate. Reed observes it may be difficult to embrace the machinery of scientific imagination while rejecting that same machinery in art.
Financial note
The piece also cites a The Wall Street Journal report that Google is investing about $75 million in A24, the independent film studio, to help bolster the creative process with AI tools.
