Saturday Morning NeuroSim Discussion Thread: Physical Computing

 

From the “Macy Conference Redux” feature form our July 1 meeting

Over the past three years, the Saturday Morning NeuroSim group has met weekly on Saturdays (mornings in North America). The Saturday Morning format continues in the tradition of Saturday Morning Physics and covers a wide variety of topics.

One recent lecture/discussion thread is on Physical Computation. Our approach to the topic begins with the debate around the role of computation in Cognitive Science and the Neurosciences. And so we begin in Week 1 with a discussion of the connections between computation, information processing, and the brain, largely focusing on the work of Gualtiero Piccinini and Corey Maley. A starting point for this session is their Stanford Encyclopedia of Philosophy article on “Computation in Physical Systems”. Many current assumptions about computation in the brain stem from the Church-Turing thesis, which often leads to a poor fit between model and experiment. Piccinini and Maley propose that the Church-Turing-Deutsch thesis is preferable when talking about systems that perform non-digital computations. Amanda Nelson pointed out the it makes sese to think of evolved biological systems (brains) as instances of analogue computers. Another interesting point from the session is the distinction between the digital (Von Neumann) computers and alternatives such as “physical” or “analog” computation, which would be picked up on in the next session.

Physical Computation Session I from June 24 (roughly one hour in length).

The second session focused on physical computation, and led us to discuss the idea of pancomputationalism. While pancomputationalism is the fundamental assumption behind the phrase “the brain is a computer” [1], we we also introduced to pancomputationalism in ferrofluidic systems and mycelial networks. We discussed the works of Richard Feynman (Feynman Lectures on Computation) and Edward Fredkin (Digital Physics), which helped us form an epistemic framework for computation in nature [2]. We also discussed Andy Adamatsky’s work on unconventional computation, particularly his work on Reaction-Diffusion (R-D) Automata, that while discrete in nature has connections to excitable (e.g neural) systems via the Fitzhugh-Nagumo model.

Physical Computation Session II from July 1 (roughly one hour in length)

After taking a break from the topic, our July 15 meeting featured an alternative viewpoint on pancomputationalism. This was made manifest in a shorter discussion on physical computation, with views from Tomasso Toffoli and Stephen Wolfram. We covered Toffoli’s paper “Action, or the funcgability of computation”, which connects physical entropy, information, action, and the amount of computation performed by a system. This paper is of great interest to the group in light of our work and discussions on 4E (embodied, embedded, enactive, and extended) cognition [3]. Toffoli makes some provocative arguments herein, including the notion of computation as “units of action”. A concrete example of this is a 10-speed bicycle, which is not only not a conventional computer, but also has linkages to perception and action. Amanda Nelson found the notion of transformation from one unit into another particularly salient to the distinction between analogue and digital computation. The physical basis of all forms of computation can also be better defined by revisiting “A New Kind of Science” [4], in which Wolfram sketches out the essential components and analogies of a computational system with a physical substrate. We can then compare some of the more abstract aspects of a physical computer with neural systems. This is particularly relevant to engineered systems that include select components of biological networks.

Physical Computation Session III from July 15 (about 15 minutes in length)

The next session followed up on computation in natural systems as well as Wolfram’s notion of universality, particularly in terms of computational models. In particular, Wolfram argues that cellular automata models can characterize universality, which is related to pancomputationalism. Universality suggests that any one computational model can capture system behavior that can be applied across a wide variety of domains. In this sense, context is not important. Rule 30 produces an output that resembles pattern formation in biological phenotypes (the shell of snail species Conus textile), but can also be used as a pseudo-random number generator [5]. In “A Framework for Universality in Physics, Computer Science, and Beyond”, this perspective is extended to understand the connections between computation defined by the Turing machine and a class of model called Spin Models. This provides a framework for universality that is useful form defining computation across the various levels of neural systems, but also gives rise to understanding what is uncomputable. This sessions natural system examples featured computation among bacterial colonies embedded in a colloidal substrate along with computation in granular matter itself. The latter is an example of non-silicon based polycomputation [6].

Physical Computation Session IV from July 22 (about 12 minutes in length).

After talking a more extended break from the topic, we returned to this discussion four weeks later (August 19). Our sixth (VI) session occurred in our August 19 meeting, and covered three topics: physical computation and topology, morphological computation, and RNA computing/Molecular Biology as universal computer.

We have discussed category theory before in our discussions on Symbolic Systems and Causality. In this section, we revisited the role of category theory, but this time with reference to Physical Computation. John Carlos Baez and Mike Stay give a tour of category theory’s role in computation via topology. The idea is that category theory forms analogies with computation, which can be expressed on a topological surface/space.

Computable Topology, Wikipedia.

Baez, J. and Stay, M. (2009). Physics, Topology, Logic and Computation: A Rosetta Stone. arXiv, 0903.0340.

Mapping category theory operators to a topological description.

We aslo covered the role of Morphological Computation by reviewing three papers on this form of physical computation that intersects with digital computational representations. Morphological Computation is the role of the body in the notion of “cognition is computation”. One idea that is critiqued with in these papers is offloading from the brain to the body. Offloading is moving computational capacity from the central nervous system to the periphery. If you grab a ball with your hand, you recognize and send commands to grasp the ball, but you must grasp and otherwise manipulate the object to fully compute the object. Thus, this capacity is said to be offloaded to the hand or peripheral nervous system.

Interestingly, offloading and embodiment are integral parts of 4E (Embodied, Embedded, Enactive, and Externalized) Cognition, which itself critiques the brain as computation idea. But as an analytical tool, morphological computation is much more utilitarian than Cognitive Science theory, and is concerned with how the robotic bodies and other mechanical systems interact with an intelligent controller. In non-embodied robotics, body dynamics is treated as noise. But in morphological computation, body dynamics play an integral role in the intelligent system and contribute to a dynamical system.

Muller, V.C. and Hoffmann, M. (2017). What Is Morphological Computation? On How the Body Contributes to Cognition and Control. Artificial Life, 23, 1–24.

Fuchslin, R.M., Dzyakanchuk, A., Flumini, D., Hauser, H., Hunt, K.J., Luchsinger, R.H., Reller, B., Scheidegger, S., and Walker, R. (2013). Morphological Computation and Morphological Control: Steps Toward a Formal Theory and Applications. Artificial Life, 19, 9–34.

Milkowski, M. (2018). Morphological Computation: Nothing but Physical Computation. Entropy, 20, 942.

The three insights from our morphological computational discussion.

While these papers do not get too deeply into the role of pancomputation in Morphological Computation, it is implicitly stated and plays a central role in our last topic: RNA computing and Molecular Biology. For more information, see this talk on YouTube and the paper below. Basically, while the pancomputationalism perspective is missing from biology, the structure and potential function of DNA and RNA provide a route to phycial computation.

Akhlaghpour, H. (2022). An RNA-based theory of natural universal computation. Journal of Theoretical Biology, 537, 110984.

Bringing pancomputationalism into biology? What is its value?

Thanks to Morgan Hough for joining us from Hawaii (4:00 am!) on August 19.

References

[1] Richards, B.A. and Lillicrap, T.P. (2022). The Brain-Computer Metaphor Debate Is Useless: A Matter of Semantics. Frontiers in Computational Science, 4, 810358.

Should we just simply “shut up and calculate”, or debate some more?

[2] Fredkin, E. (2003). An Introduction to Digital Philosophy. International Journal of Theoretical Physics, 42(2), 189.

This work is the Rosetta Stone for many comparisons between modern AI systems and human-like intelligence, at least in terms of computation.

[3] Newen, A., DeBruin, L., and Gallagher, S. (2018). The Oxford Handbook of 4E Cognition. Oxford University Press.

[4] Wolfram, S. (2002). A New Kind of Science. Wolfram Media.

This is a link to the 20th Anniversary edition, with a full set of Cellular Automata rules, defined by number.

[5] Zenil, H. (2016). How can I generate random numbers using the Rule 30 Cellular Automaton? Quora post.

[6] Bongard, J. and Levin, M. (2023). There’s Plenty of Room Right Here: Biological Systems as Evolved, Overloaded, Multi-Scale Machines. Biomimetics, 8(1), 110.

The World of Physical Intelligence

As part of the Embodied Intelligence workshop this past week, I saw a presentation by Metin Sitti of the Max Planck Institute for Intelligent Systems on the emerging paradigm of Physical Intelligence. What is Physical Intelligence? It is the intelligent behavior exhibited by motion and takes into account a morphology (embodiment) and action at multiple scales of spatial organization. The talk was wonderful and walked through a number of empirical studies involving both living and non-living systems. I will leave it up to the reader to appreciate all of the points raised in this talk.

Physical Intelligence as an emerging idea.

While Physical Intelligence is not a common paradigm, the general idea is actually not new. Other people have proposed very similar frameworks over the past 15 years or so, including yours truly. Physical Intelligence involves some form of motor or movement behavior, which is generated by an embodied agent, that in turn interacts with the physical environment that can be defined by features such as inertial and gravitational forces, surface textures, and even light energy. 

Physical stimuli for soft robots or other autonomous agents from Figure 1 in Shen et.al, Journal of Materials Chemistry B, 8, 8972-8991 (2020).

At its most human-centric, physical intelligence can be just another word for embodied intelligence, which is where the body shapes and determines what is experienced by the nervous system. In cases where the behaving agent has no nervous system, the body geometry shapes the agent's behavioral output. This is true both in cases of adaptive (intentional) behavior and reactive behavior. In some cases, physical intelligence is identical to Neuromechanics. In other cases, it resembles a range of fields, from Biophysics to Embodied Robotics. The important contribution of the Physical Intelligence paradigm is the principles that guide this diverse topical terrain. Both my talk and Metin's talk provide some of these potential principles, and Scott Grafton's book provides a few more.

One human-centric interpretation of Physical Intelligence (by Scott Grafton).

Metin brings up the example of the Strandbeest, which is a kinetic sculpture with no nervous system or centralized control. This is an example of a purely reactive system that also generates seemingly intelligent behavior. Closer to the human experience is the Passive Dynamic Walker, which produces human-like bipedalism without a central nervous system. This is the essence of the physical: a particular physical configuration can exhibit reactive behavior independently of a central controller.

While serving as part of the peripheral nervous system, and in fact being controlled by a central nervous system, muscles can also play a key role in physical intelligence. While muscle cells can be spontaneously active without being inputs by motor neurons, there is a elaborate coordination between the central nervous system and muscular control. Muscular control can produce both very fast and very slow adaptive movements. In addition, the overall shape of a body in relation to its muscles can constrain the behavior of the agent in question.

The main takeaway is that the brain, body, and environment all work interdependently to shape the behavior that emerges from this complex system. Even in cases where there is no brain (or neural network), the interactions between body and the environment are enough to generate reactive behaviors that appear to be intelligent. This is true for both individual morphologies and the collective behavior of many agents (e.g. swarm intelligence). In fact, the brain can be supplanted by control mechanisms that regulate the conformity and response to physical forces in the environment. Future work should focus on the differences between "neural" and "physical" behavior, as well as the necessity and sufficiency of each component in the triad.

A diagrammatic example of this relationship (with a brain in the feedback loop) from Chiel and Beer, Trends in Neuroscience, 20(12), P553-P557.

In conclusion, DARPA has also engaged in the idea of physical intelligence, and there was a proposer's conference in 2009. This version of Physical Intelligence has a strong cybernetics flavor, particularly in incorporating the EGRT (Every Good Regulator Theorem) into the mix.

Cybernetics of the firm (deemphasizing the role of individual morphology). COURTESY: New World Encyclopedia.

Virtuality: a new view on virtual experience

The first images when you Google "virtuality". Products from the Virtuality Group. LEFT: Virtual Reality Society, RIGHT: Wikimedia (Dr.  Jonathan D. Waldern/Virtuality Group).

At Neuromatch 2.0, I gave a talk on the concept of virtuality titled "Computational Virtuality as a Form of Artificial Intelligence" (slides available on Figshare). Virtuality is a means to summarize all brain and body systems involved in a virtual experience. This talk features two big ideas, the second of which the Saturday Morning NeuroSim group will be building on in the coming months.  

1) "Being Virtual" as Human Performance (click to enlarge).

2) Differentiating Allostasis machine (click to enlarge). Allostatic load is induced collectively by stimulus [1], leads to dysregulation of system and ultimately to a new homeostatic setpoint.

Overall, virtuality can be characterized as a constructive sensory experience (h/t to Anson Lim). The first big idea stems from the observation that current theories are too limiting to properly characterize virtual experiences at the systems-level. Indeed, the very notion of objective reality is based in part on the synchronization of perception and action. When this synchronization is disrupted or perturbed in any way, we observe virtuality. This does not even require a virtual environment -- virtuality can be experienced in our interaction with the physical world [for an example from fly-human interaction, see 2]. But virtual environments are the most common way to elicit the allostatic load required to observe virtuality.

There are two aspects of embodied performance essential to understanding the virtuality effect. The first is the idea of cognitive gaps, or disruptions to spatiotemporal representations caused by the perturbation of reality. The second involves inducing allostatic load in the internal model the represents perception-action coupling. The latter has wide-ranging effects in the homeostat (nervous system), and can lead to many different effects of varying length scales.

But where does the artificial intelligence part come in? In the second part of the talk, I conclude that the effect sizes for human experiments are too small and formal experimental design is too limiting to demonstrate virtuality. This is particularly true for observing virtuality over long periods of time. Therefore, we can use computational agents! To propose a general approach, we want to identify two attentional and three sensorimotor features essential for any agent to exhibit virtuality.

Naturally, we want our agents to be embodied in some manner. We propose working with two types of agent: Morphogenetic Agents, a novel agent type that can exhibit both pattern recognition and morphogenesis (percepetion-action), and the well-known Braitenberg Vehicle, in this case experiencing incongruous environmental physics which leads to gaps in the perception-action loop. 

NOTES:

[1] this requires a ecological view of perception. For a short introduction, please see: Lobo, L., Heras-Escribano, M. and Travieso, D. (2018). The History and Philosophy of Ecological Psychology. Frontiers in Psychology, doi:10.3389/fpsyg.2018.02228.

[2] Alicea, B. (2013). Perceptual Time and the Evolution of Informational Investment. Synthetic Daisies blog, September 24.