There's Plenty of Room Right Here: Biological Systems as Evolved, Overloaded, Multi-Scale Machines
As Hoel points out about the ad hoc status of claiming one single privileged perspective within a system, according to the integrated information theory (IIT) account of consciousness [257]: “…There are so many viable scales of description, including computations, and all have some degree of integrated information. So, the exclusion postulate is necessary to get a definite singular consciousness. This ends up being the most controversial postulate within IIT, however.” [99].
We maintain that, as our understanding of polycomputing biological and technological systems increases, it will eventually exclude the exclusion postulate from any attempt to explain human consciousness as a mental module operating within the brain.
Prior skeptical debates about whether biological systems are computers reflect both an outdated view of computation and a mistaken belief that there is a single, objective answer. Instead, we suggest a view in which computational interpretations are not simply lenses through which diverse observers can all understand a given system in the same way, but indeed that several diverse interpretations of the information being processed by a set of events can simultaneously be useful to different subsystems (observers) at the same time. It is now seen that there is no one-to-one mapping between biological form and function: the high conservation of the biological form and function across evolutionary instances implements a kind of multiple realizability. At the same time, biological components are massively overloaded with regard to polycomputing. Indeed, their competency, plasticity, and autonomy [2,22,258,259] may enable a kind of second-order polycomputing, where various body components attempt to model each other’s computational behavior (in effect serving as observers) and act based on their expected reward, from their perspective. Thus, modern computer engineering offers metaphors much more suited to understand and predict life than prior (linear and absolute) computational frameworks. Not only are biological systems a kind of computer (an extremely powerful one), but they are amazing polycomputing devices, of a depth which has not yet been achieved by technology. In this sense, biological systems are indeed different than today’s computers, although there is no reason why the future efforts to build deep, multi-scale, highly plastic synthetic devices cannot take advantage of the principles of biologic polycomputing.
A key implication of our view is that that blanket pronouncements about what living or non-living machines can do are worthless: we are guaranteed to be surprised by outcomes that can only be achieved by formulating and testing hypotheses. It is already clear that synthetic, evolved, and hybrid systems far outstrip our ability to predict the limits of their adaptive behavior; abandoning the absolutist categories and objective views of computation is a first step towards expanding our predictive capabilities.
At stake are numerous practical outcomes, in addition to fundamental questions. For example, to make transformative advances in our ability to improve the health in biomedical settings [260], we must be able to control multiple scales of biological organizations which are heavily polycomputing—from cellular pathways to patient psychological states. It is essential to begin to develop computational frameworks to facilitate that kind of control. The ability to construct and model a kind of computational superposition, in which diverse observers (scientists, users, the agent itself, and its various components) have their own model of the dynamic environment, and optimize their behavior accordingly, will also dovetail with and advance the efforts of synthetic bioengineering, biorobotics, smart materials, and AI.
We thank Oded Rechavi and Aimer G. Diaz for useful pointers to relevant biological phenomena, and Julia Poirier and Susan Lewis for assistance with the manuscript. We thank the reviewers as well, who in addition to providing much useful feedback also proposed several specific and actionable research paths forward from the described work.
J.B. and M.L. contributed equally to this work. All authors have read and agreed to the published version of the manuscript.
M.L. and J.B. are co-founders of Fauna Systems, an AI-biorobotics company; we declare no other competing interests.
M.L. gratefully acknowledges the support of the Templeton World Charity Foundation (grant TWCF0606) and the John Templeton Foundation (grant 62212). J.B. gratefully acknowledges the support of the National Science Foundation (NAIRI award 2020247; DMREF award 2118988).