AI 2027 underestimates synthetic biology: it could create a self-replicating economy before robotic factories do
AI 2027 underestimates synthetic biology: it could create a self-replicating economy before robotic factories do
On July 14, a LessWrong author argued that the transition could begin earlier with the smallest systems capable of replicating themselves from available raw materials. One of the scenario’s coauthors agreed with part of this criticism. AI 2027 itself already includes a more distant possibility involving biological infrastructure.
In the AI 2027 scenario, the United States and China establish special economic zones with simplified construction regulations. In the American zones, Agent-5, the superintelligence in the scenario, designs robots and directs human workers as they build factories and laboratories. The new factories then produce more robots. As a benchmark, the authors cite a car factory that produces vehicles with a total mass comparable to its own in less than a year. They then suggest that an autonomous robotic economy might be able to replicate itself in under a year.
In a response to the scenario, Thomas B. argues that the future leap may occur below the scale of a factory. He suggests that a superintelligence would benefit from reducing the unit of self-replication as far as possible and designing organisms and nanotechnological systems that grow and reproduce using raw materials from their surroundings.
The argument rests on experimental speed. According to Thomas B., the AI 2027 assumption of human-level robots would imply automated laboratories that are not constrained by shortages of skilled human labor. He proposes the following sequence: these laboratories could create large numbers of inexpensive living sensors and effector systems, run experiments on them in parallel, and feed the results into simulations. Better simulations would help design the next generation of organisms and devices. Each round of research would therefore shorten the time needed for the next round.
The disagreement comes down to the speed of the feedback loop. A factory requires buildings, machine tools, supply chains, and large machines. A self-replicating biological system could simultaneously serve as a sensor, an object of parallel experimentation, and a component of the next production system. Thomas B. argues that this possibility should be central to the scenario.
AI 2027 itself already points in this direction. The authors cite plants, insects, and bacteria as benchmarks for reproductive speed. In a more distant version of the scenario, they consider an economy resembling an ecosystem of novel algae: some organisms grow, while others process them into materials for floating factories. The disagreement concerns when this biology would appear. The main scenario develops a robotic economy first. Thomas B. considers biological self-replicating systems a plausible early driver of acceleration.
In the comments, Thomas Larsen, one of the coauthors of AI 2027, agreed with the central criticism and wrote that the ending should have been connected more strongly to nanotechnology. The debate shifts the central question about superintelligence away from the number of robots and toward the duration of the cycle from experiment to data to a new system. If that cycle can be compressed to match the reproductive speed of engineered organisms, factories will no longer be the only measure of productive capacity.