Michael Levin Proposes Using AI-Guided Experiments to Find Desired Living Tissue Forms

Michael Levin Proposes Using AI-Guided Experiments to Find Desired Living Tissue Forms

On August 21, developmental biologist Michael Levin released a lecture on “free lunches.” In it, he describes Mambbot, a collaborative project in which AI proposes how to stimulate cells with light, vibration, temperature, or chemical signals to search for a biobot with a specified form and function.

Levin defines a “free lunch” as the gap between what a system produces and the effort explicitly invested through design, selection among variants, or training.

“The degree of ‘free lunch’ is the difference between what was achieved and the effort invested.”

In his view, this gap determines the next experiment: researchers ask which property of the system produced the result and what intervention could trigger it again.

This idea builds on years of morphogenesis research in his laboratory at Tufts University. Morphogenesis is the process by which groups of cells assemble, repair, and alter the form of a body. In a 2021 paper, the authors described xenobots, motile structures made from frog embryo cells. These clusters gathered loose cells into new clusters, while an algorithm selected forms that reproduced more effectively. In a study of anthrobots, adult human airway cells self-organized into motile structures. In neuronal cultures, they accelerated closure of a damaged area. These results gave researchers a system in which they could change the conditions and then observe the resulting form and function of cell collectives.

In the lecture, Levin presents Mambbot as the next step in this work. A researcher specifies the desired form or function of a biobot, and AI proposes a hypothesis about the appropriate stimulus, which may involve light, vibration, temperature, or a chemical signal. Levin frames the question as follows:

“What stimulus should be applied to the cells to produce a biobot with the desired form and function?”

Researchers can then apply that stimulus to the tissue, compare the resulting form and function with the target, and select the next test.

Levin calls this type of search synthetic morphology. It examines the conditions under which a cell collective can assemble different forms and perform different functions. AI helps select an experiment, and the tissue’s response indicates which hypothesis should be tested next. Levin thus describes an experimental program in which each intervention is based on the observable response to the previous one.