LAUSR

The exciting sci-tech arena of synthetic biology (SB) provides concepts, tools, and approaches for fundamental revolutions in basic and applied research. SB plays a key role when it is conceived as one of the branches of the “sciences of the artificial” (Cordeschi, 2002; Damiano et al., 2011), together with artificial intelligence (AI) and robotics. In particular, SB contributes to the wetware approaches, which are complementary to the software and hardware ones that characterize the other two most well-known branches. In this perspective, SB offers the unique opportunity of devising novel chemical versions of AI, whose main feature is embodiment, i.e., forms, systems, networks that compute through physical interactions (not based on the abstract representations typical of AI), and that potentially display autonomous adaptive/plastic dynamics (in contrast to mechanical robots). SB, then, can be seen as an experimental platform for unconventional computing, based on (bio)chemicals, organized structures, and reactions. Operations, even when interpreted by observers in terms of logical representations, actually lie in the material domain. As such, operations (reactions, interactions, synthesis and degradation of the operators) and the operators themselves (the molecules performing or subjected to the operations) are truly interwoven, and definitely cancel the distinction between hardware and software, typical of non-chemical machines. In this opinion paper, we aim at sketching a possible implementation of embodied, chemical AI by means of SB tools. In particular, we will focus on bottom-up approaches and on the socalled synthetic (or artificial) cells (SCs or ACs) (Luisi 2002; Salehi-Reyhan et al., 2017; Göpfrich et al., 2018; Guindani et al., 2022), Figure 1A. In the past few years, indeed, the worldwide community of SC practitioners has generated a very relevant momentum, promoted by the onset of numerous consortia and projects (Schwille et al., 2018; Frischmon et al., 2021). The question we would like to deal with is the following: is it possible to devise minimal forms of perceptive chemical AI in SCs? Because of its widespread relevance since the beginning of AI, the system we look at is a chemical perceptron (a chemical neural network), and we will discuss its possible implementation inside SCs. We will first introduce the motif of “phospho-neural networks” (Hellingwerf et al., 1995) and a plan for implanting such networks in SCs, calling for a specific design that would address both experimental feasibility, detailed modeling, and non-trivial behavior. We also suggest that the exploration and the interpretation of chemical networks’ dynamics, especially when they are based on macromolecular elements, is best pursued according to fuzzy logic. Finally, a short comment on the theoretical relevance of these approaches on the more general problem of embodying AI in the chemical domain will complete the paper. Edited by: Ovidiu Radulescu, Université de Montpellier, France