Today’s scientists will be the last generation of human researchers, predicts immunologist Derya Unutmaz: AI will even choose the scientific questions
Today’s scientists will be the last generation of human researchers, predicts immunologist Derya Unutmaz: AI will even choose the scientific questions
On November 7, the Jackson Laboratory professor outlined a possible path for science: joint research by humans and AI, machine-led discoveries under human direction, and eventually machines choosing which problems to pursue. He expects AI to be making most discoveries within a few years and suggests that this stage could last one or two decades.
Unutmaz’s prediction was prompted by an account of the autonomous laboratory AutoDNA. Its AI software assistants receive a task from a human, plan an experiment, and generate instructions for the instruments. The instruments run the experiment and return data, which the AI uses to adjust the conditions for the next run. In DNA assembly experiments, the system sought to increase the proportion of strands to which the next building block was added and reduce the reaction time. A human sets the goal, and the AI decides what to change in the next experiment.
Unutmaz distinguishes eras of science by who makes the discoveries. He calls the work of individual researchers “Science 1.0” and the work of teams “2.0.” He places the past two or three years in “3.0,” with humans and AI conducting research and producing new findings together. On August 6, he described “4.0” as scientific discovery by AI and robots with minimal human involvement. In engineering, he still assigned humans the task of choosing what to improve next.
In his November prediction, humans direct “Science 4.0.” They choose problems, set priorities, apply the findings, and test them in laboratory experiments. Superintelligence, an AI that far exceeds human capabilities, will make most discoveries independently.
Unutmaz attributes the transition to “Science 5.0” to a further widening of the gap between human and machine intelligence:
“Our intelligence will be insufficient even to ask the right questions.”
In his scenario, humans will also be unable to fully understand the answers that superintelligence provides. With a gap of this size, superintelligence and autonomous robots will determine almost all scientific progress themselves, including which research directions to pursue.
Unutmaz extends this prediction to his own generation. He recently told a scientist friend:
“We scientists with modern scientific training are the last generation of human scientists.”
He writes that he is grateful for the opportunity to witness scientific inquiry passing to a more powerful intelligence. Unutmaz hopes that superintelligence will find answers about life, consciousness, and the nature of reality that lie beyond the reach of the human mind.