Behaviorist approaches to investigating memory and learning: A primer for synthetic biology and bioengineering

Learning capacity, in the sense of specific changes of future responses in light of past experiences [91], has been suggested in many unconventional substrates, including cells [92–96] and even subcellular components such as molecular pathways and gene regulatory networks [97–102]. Moreover, synthetic biology is increasingly providing micro-designed [37,38,103–106] or emergent [42,43] active living constructs. The space of possible subjects for learning experiments is vast, and is growing all the time given developments in smart materials, synthetic bioengineering, brain-computer interfaces, and other fields. The ability to train synthetic living machines for useful functions [25–27] will be a very important new toolbox for the bioengineer, in addition to the design of novel bodies with hardwired operation. It is also likely to have implications for machine learning and robotics, as novel learning architectures discovered in aneural and neural systems could improve performance if imported to silicon-based (or other) media.

Developing effective training protocols for novel model systems is important in two other broad ways. First, it will shed essential light on the fundamental aspects of learning, distinct from the frozen evolutionary accidents of the phylogenetic history of life on Earth. Indeed, many debates about the locus and mechanism of memory could be enlightened by experiments in unconventional substrates [107–109]. Second, training offers workers in regenerative medicine and bioengineering a path toward outcomes that are too complex to micromanage by physical (genetic, pathway) rewiring. By exploiting the learning and basal problem-solving capacities of cells and tissues (in vivo or in vitro), biomedical strategies could push much of the complexity onto the system itself: using stimuli, not hardware rewiring, to achieve desired endpoints such as specific morphogenetic outcomes, whether in the patient or in synthetic living machines with useful functions [110]. Much as evolution exploits learning to achieve outcomes far faster than is possible at the genetic level alone, scientists and engineers can leverage the same advantages to overcome the inherent complexity of the mapping between biological structure and function.

We thank Julia Poirier, Douglas Blackiston, and Haleh Fotowat for helpful comments on the manuscript.

M.L. is supported by the John Templeton Foundation (grant id 62212) and by the Templeton World Charity Foundation (the opinions expressed in this publication are those of the author(s) and do not necessarily reflect the views of Templeton World Charity Foundation, Inc.). The participation of C.I.A. was supported in part by NSF-REU grant 1950805 and NSF-PIRE grant 1743753.