1. Computationalism in an Eco-Cognitive Perspective
What I call Eco-Cognitive Computationalism
sees computation in context, exploiting the ideas developed in those projects that have originated the recent views on embodied, situated, and distributed cognition: computation is seen as active in physical entities appropriately transformed so that they become what I call cognitive mediators
, in which data can be encoded and decoded to obtain productive results. This eco-cognitive approach to the concept of computation is strictly related to the tradition of research concerning the so-called distributed and embodied cognitive systems. The “distributed” approach sees cognition as a socially distributed process, pragmatically oriented, and affirms that cognitive processes can be better analyzed as situated in and distributed across material artifactual circumstances, in which the “ecological” view also emphasizes the role of the agent-environment interaction. The theory of distributed cognition was created by Edwin Hutchins [1
] to describe in a novel and more satisfactory way various problem solving processes in real work situations, so providing a new perspective that encountered great success in general cognitive science. I consider this approach particularly appropriate to treat the concept of computation: I will try to show that using this perspective both Turing’s discoveries and the recent new deal on morphological computation can be better understood and at the same time seen as sharing analogous epistemological characters. Indeed, eco-cognitive computationalism does not reduce computation to digital computation (that is to the processing of strings of digits according to rules), but it is open to include and consider other and new forms of computation. The perspective of the ecology of cognition concentrates on “physical computation”, exactly following Turing’s original thoughts concerning the emergence of computation in organic, inorganic, and artefactual agents, I will quickly describe in this article.
Turing’s germinal speculations on how the so-called “unorganized brains” and “unorganized machines” (On the meaning of these concepts in Turing’s sense, see Section 2.1
and Section 5
below) are transmuted in organized “machineries” are extremely interesting: the problem is to see how, so to speak, “innocent” entities (in the sense of virginal objects “free from cognitive capacities”) become first of all carriers of information and knowledge, but also carriers of computation. Turing is convinced that the emergence of rudimentary forms of information and cognition can be clearly comprehended in an evolutionary perspective, as the fruit of an eco-cognitive interplay and simultaneous coevolution
, in time, of the states of brain/mind, body, and external environment. At the same time, this evolutionary perspective favors the subsequent creation of both the Universal Logical Computing Machine (which illustrates computation as a pure syntactic process) and the concrete Universal Practical Computing Machine (Further details on this view proposed by Turing are illustrated in [3
]). Machines lose their cognitive innocence (they were already merely involved in elementary low level cognitive tasks, such as in the case of telephone, telegraph, etc.) and become universal
cognitive physical entities, in so far as they become computational
artefacts that compute for humans or artefactual agents: those computers that in this perspective offered by Turing I called “mimetic minds” [4
] (the concept of mimetic mind is explained below in the following Section 2
). I hope it will become clear that eco-cognitive computationalism does not aim at furnishing a final and stable definition of the concept of computation, such as a textbook or a different epistemological approach could provide: I intend to take into account the historical and dynamical character of this concept, to propose an intellectual framework that depicts how we can understand not only the change of its meaning, but also the “emergence” of new forms of computations.
2. The Birth of Mimetic Minds: Educating Human Brains/Educating Physical Entities
I have said in the previous section that what I call eco-cognitive computationalism
sees computation as active in physical entities suitably transformed so that they become what I call cognitive mediators
, in which data can be encoded and decoded to obtain useful results. I also stressed the importance of an evolutionary framework that sees the emergence of information, cognition, and computation as the result of a complex eco-cognitive interaction and simultaneous coevolution, in time, of the states of brain/mind, body, and external environment. As I have stressed in [3
], the concepts of information, cognition, and computation must not be considered as static, and their meaning changes depending on the modifications of theory and practice: in this article, I will present the new variation of the meaning of the concept of computation generated by the involvement of morphological aspects.
Following the Peircean semiotic perspective, we can remember that signs can be externalized in both natural and artefactual environments and we could add, using a term coming from the last decades of research in cognitive science, that signs are distributed
and so externalized in a process also called “disembodiment of the mind”. In his seminal article Intelligent Machinery
], Turing illustrates this same process adopting some neurological metaphors: a big cortex can present an evolutionary advantage when two conditions are fulfilled: (1) the presence of a huge quantity of relevant signs (information and knowledge) available in an environment suitably “artificialized” (props, supports, etc.); and (2) the embedment, discretely developed, in a small community able to manage information (In [4
], I have indicated that this speculative argumentation has been recently supported by research in paleoanthropology on the birth of material culture).
The birth of computation is indeed interestingly linked—in Turing’s article I have just quoted—to creative heuristics intertwined with the analysis of the function—out there, in the external environment—of organic bodies, physical entities, and artefacts.
2.1. Educating Human Brains
Turing contends that human intelligence is obviously generated by an appropriate education ([5
], p. 3) and an analogy between human brains as organic entities and computational machines as physical artefacts has to be built. Brains of the human infant are similar to what he calls unorganized
machines—machines which are largely random in their construction—that can be educated thanks to “rewards and punishments”. The analogy created by Turing “the cortex of an infant is an unorganized machine” ([5
], p. 16) is speculative and just plays a central heuristic role in his creative cognitive processes: obviously, we know from basic neuroscientific research that the brain is a highly organized system. Turing just refers to the fact that the cortices of infants are kinds of blank slates which are “socially” fulfilled through language. From this perspective, the hypothesized random construction of the infant cortex does not deal with neurobiological aspects but with the absence of “education” coming from the external social world—for example, language. Thus, the definition of “unorganized machinery” is related to a special level of description.
When an unorganized machine (such as an infant cortex) is submitted to suitable interferences, its behavior changes and step after step the machine becomes organized (and possibly also universal). Not only is the existence of a human cortex evolutionarily justifiable only in terms of its organization in the framework of a coevolution
between it and the external information available to organize it: “[…] the possession of a human cortex (say) would be virtually useless if no attempt was made to organize it. Thus if a wolf by a mutation acquired a human cortex there is little reason to believe that he would have any selective advantage” ([5
] p. 16). An environment full of information (made real thanks to speech and at the same embedded in a social background
in which several “techniques” are usable and learnable) has to be already available, but also, as I have noted above, together with a small community that can grant the passage of information across generations.
We have said that, to educate an infant brain, rewards and punishments are needed: this fact indicates that organization can happen only through those two inputs and, finally, the unorganized human cortex is changed in an intelligent one thanks to discipline
, which Turing considers the two main aspects of a process that has to be studied as it happens in humans to “copy it in machines” ([5
], p. 21).
2.2. Educating Physical Entities
It is natural for Turing to employ the idea of education in the case of humans to construct the new concept of computation: also physical machines can be educated
to produce certain kinds of modification. In this last case, education coincides with programming, which is an imitation of the human case: we can modify machines thanks to programming to the aim of reaching some good and interesting responses. The concept of (Universal) Logical Computing Machine (LCM) emerges in this epistemological atmosphere and, on the basis of this abstract theoretical tool, also one of the (Universal) Practical Computing Machines (PCM). These are digital computers (as physical entities) that can be built and “educated”, able to imitate the behavior of a human computer very well. I have already anticipated that, to show Turing’s externalization of both the abstract (the logical machine) and the concrete (the practical machine) entities, I have called them mimetic minds
]: these machines are indeed capable of imitating the mind in a universal manner and are objectified in the environment, as clearly expressible formal intellectual structures and practical machines. They are universal because we do not need various machines that do different performances: thanks to suitable programming activities, we can have various performances done. The analogy is completed saying that universal machines are subjected to paper
interference, when the introduction of new information in the machine modifies its behavior, but also (and here the analogy is no longer active) to screwdriver
interference, when some parts of the machine are eliminated and and replaced by others, giving rise to completely new machines (by the way, in organic agents, this effect of changing the structure is caused by the evolution) (More details on Turing’s ideas summarized in this section are illustrated in ([6
], Chapter 3)).
4. Pancomputationalism Naturalized
The problems of pancomputationalism is nicely summarized in the following passage by Horsman et al.:
There is currently no accepted answer to this question, and an absence of a worked out formalism within which to determine whether a computation is happening physically gives rise to a great deal of confusion when discussing non-standard forms of computation. We can all agree that a laptop running a Matlab calculation and a server processing search engine queries are physical systems performing computation. However, when we move beyond standard and mass produced technology, the question becomes more difficult to answer. Is a protein performing a compaction computation as it folds? Does a photon (quantum) compute the shortest path through a leaf in photosynthesis? Is the human mind a computer? A dog catching a stick? A stone sitting on the floor? One answer is that they all are – that everything that physically exists is performing computation by virtue of its existence. Unfortunately, by thus defining the universe and everything in it as a computer, the notion of physical computation becomes empty. To state that every physical process is a computation is simply to redefine what is meant by a “physical process”—there is, then, no non-trivial content to the assertion. A statement such as “everything is computation” is either false, or it is trivial; either way, it is not useful in determining properties of physical systems in practice.
The assumption of pancomputationalism clearly states that everything is computational [23
]. This view is often defended contending that, when we say that everything is computational, we are just describing something as computational as a way of “interpreting” it, and everything can be interpreted that way. Unfortunately, this defense leads pancomputationalism to conflate computational modeling with computational explanation. Other forms of pancomputationalism contend that the whole universe can be considered computational and seem to also claim that information or computation have a kind of priority with respect to physical materiality ([11
], p. 5). Gordana Dodig Crnkovic [15
] proposes a richer info-computational view of the universe (a synthesis of pancomputationalism—naturalist computationalism—with informational structural realism) and defends it by observing the central role of computing in nature (natural computing).
As I have observed in [3
], taking advantage of an evolutionary perspective and of the Thom’s concepts of the catastrophe theory [26
], when we see the case of an infection as a pregnance (mediated by a virus, that is a material/biological medium) that affects healthy subjects, who are the invested saliences that in turn can re-emit that same contagion as a pregnance into the natural niche (in which, in turn, other media such as air or blood are the transmitters), it seems weird to contend that information (or computation) is at play, in this case against paninformationalism and pancomputationalism. One count is to produce a “model” of that event from an informational point of view or by using a computational program, and another count is to produce a biological knowledge of it (In the following section of this article, I will describe the dangers that can emerge by thinking that mimetic computational modeling “is” immediately, ipso facto, scientific knowledge). However, I think that the positions which defend paninformationalism and pancomputationalism are significative for two reasons. The first reason is related to the consideration of natural human evolution, which presents that widespread activity of semiosis (including all kinds of signs, not only the propositional ones) that has been created by humans since the times of our primitive ancestors. In this way, humans have built those voluminous cognitive niches
, hugely endowed with informational (and more recently, computational) processes [27
] certainly favors an inclination to envisage some ontological status to information and computation. In addition, I contend that thinking in terms of “distributed computation” (an expression I have introduced above in Section 3.6
) helps us see pancomputationalism in a more naturalized way, avoiding ontological or metaphysical considerations.
The second reason is epistemological: in the literature, there is even an all-encompassing notion of information, a kind of paninformationalism, in which physical (or biological) information is extrapolated to every state of a physical (or biological) system that is delineated as an information-carrying state [30
], which cannot be under-evaluated. Indeed, this perspective has favored many excellent results that physicists (for example) [32
] (and logicians) have reached, for example providing mathematical frameworks for seeing quantum theory in the perspective of the principles of information processing. My only warning does not concern these results, but the possible abuse of the notions of information (and of computation) in physics and biology, a problem extensively illustrated in detail in [16
], as I will illustrate in the following section.
5. Using Physical Computing to Model Physical and Biological Systems
As I have indicated above (Section 2
), Turing argumentations are coherent with the illustration of physical computation as the realization in terms of a physical evolution of an abstract computation I described in Section 3
. To quickly recall Turing’s seminal ideas, we can say that he contends that a big cortex constitutes an evolutionary advantage only if it is fertilized by a great quantity of meaningful information and knowledge carried by external supports, props, and tools—interacting with it—that only an already “evolved” collective of humans can have. In summary, information
, but also cognitive
contents carried thanks to language, signs, icons, etc. have to be more and more available to promote the useful exploitation of a big cortex. Paleoanthropologists such as Mithen [35
] would also add that storing in the external environment signs and drawings, and manipulating external entities transforming them in artifacts, is the main process that characterizes not only the birth of the so-called material culture, but also the trigger of a fundamental process of coevolution
between brains and culture—on this issue cf. also the recent rich and informed book written by Laland [37
Making an analogy to this evolutionary process, Turing contends that transmuting an external physical artefactual entity (such as an electronic physical object) “paper interference” is needed, when the introduction of new information in the machine modifies its behavior. The Turing lucubrations about the “unorganized” brains, described as kinds of blank slates that are socially fulfilled through language are really modern because, even if very abstract and characterized by a dominant heuristic role, show that phylogenetic mechanisms acting in human cognition are crucial and worth being taken into account, an attitude more powerful with respect to the one of traditional philosophical Western schools.
It is important to note that, in this view offered by Turing, the digital machine (a discrete state machine) is certainly an alphabetic machine: its conditions of possibility resort to human evolution towards alphabetic natural language. Longo [16
] contends that this fact is at the origin of that tremendous “discretization of knowledge” that the Turing’s achievements have created. The “continuous” natural language is indeed transmuted by the alphabet in something separated into small atoms, which forge letters. These atoms do not present any kind of meaning that instead comes out thanks to their syntactical aggregation made by skilled human agents able to sensibly combine them. This discreteness, typical of digital machines, is the fundamental aspect that motivates their imitation
power—they are mimetic machines, mimetic minds
, as I said—and Turing himself contrasted this simple imitation power to the much stronger epistemological power of the modeling
capacity of mathematics, when he was thinking about the science of morphogenesis.
At this point, the problem of imitation leads us to consider a further aspect of physical computation: when computers are further used to model physical (or biological systems), are we still dealing with imitation or with a kind of reliable production of knowledge that could occasionally be called “scientific”? Let us see more details concerning this problem. We have indeed said in the previous section that physical computation is a transposition of physical evolution for abstract computation, that is, we can say, following Turing, that we “educate” a physical system to perform a computation. Once this task is performed, we can in turn submit a physical (or biological) dynamic of a system to a computation modeling, that is, we can use computers to simulate the behavior of a physical or a biological system. In the meantime, it is important to note that in this case the computational simulator and the physical or biological system simulated interact at the abstract level and what is simulated is a model
of the physical or biological system, not the system itself ([13
], p. 17). We know that computational modeling of physical or biological processes can be extremely useful as a heuristic tool in actual scientific research, also at the creative level (just to make a simple example, to simulate the behavior—through modeling [38
]—of a physical system during an experiment), but we have to note that the study of a physical or a biological evolution can not always take advantage of a computational representation. Hence, what is the epistemological status of computational simulation?
The dichotomy between discreteness and continuity involves a reflection upon the other related dichotomy between imitation (as an effect of the computational representation) and intelligibility (as the fruit of scientific knowledge). I will devote the next paragraphs to better describe this important issue, at least from a general theoretical point of view. As Longo [16
] illustrates, the digital machine (a discrete state machine) is first of all an alphabetic machine, made possible thanks to the human evolution to alphabetic natural language (of course, we know it is also based on the so-called logical and formal machine). As I have already noted, this fact explains that powerful “discretization of knowledge” that mainly characterizes the “computational turn”.
This discretization, in the case of the mental representations of “concepts”, generated great consequences. We can guess that an isomorphism is established between the mental processes (where of course “phonemes” play a dominant role), which ensure the stability of a concept, and the physical and material processes that ensure the stability of the actual object represented by the concept. A discretization of knowledge that—long before the computational turn—did not have a marginal role in the recent history of human cognition. The internalization of phonemes [39
] has been related to the development of the “enhanced working memory” (EWM) appeared about 30,000 years in hominids and seemed to coevolve with the emergence of a phonological storage capacity
along with consequent language and other modern reasoning abilities such as planning, problem solving/algorithm manipulation, analogy, modeling, holding inner representations, tool-use, and tool-making, but also cross-modal thinking also needed in social tasks. In particular, an increase in phonological storage could also have aided cross-modal thinking (and so hypothetical cognition) and the social tasks caused by the need of the preservation and defense of collective coalitions: “[…] enhanced phonological storage may have freed language from the laconic and its confinement to present tense and simple imperatives to rapidly-spoken speech and the use of future tense—the linking of past, present, and future, and the use of the subjunctive […]. Although real enemy’s actions might be anticipated, imaginary enemies could be envisioned and other intangible terrors could be given life. Great anxieties could arise with novel vistas (e.g., the meaning of life, thoughts of death, life after death, etc.)” ([39
], p. 22).
Notwithstanding the triumph of discretization, in western written natural languages but also in philosophical and logical knowledge, from Democritus to Descartes and from the modern XIX century axiomatics to the computational turn, we are still facing the conundrum represented by the fact that, however, this simple reality of small components is actually very complex, as recent scientific research into the dynamical systems theory, quantum physics, and biology demonstrates. For example, it is difficult to study the cell only by referring to its constituents, and also its “wholeness” is fundamental. The suspect is that, when we use computational devices, which are discrete machines, to produce knowledge about physical and biological systems, some serious expressive limits arise: it is unlikely that these machines can play the role of instruments for directly building scientific intelligibility regarding complex objects/systems such as physical and biological entities (and also human cognitive systems themselves) (We also have to note that, in the last few decades, the notion of digital computation certainly played the role of modeling neural activity and of central features of human cognition, but many neuroscientists (for example [40
]) strongly contended that neural networks do not perform digital computation at all: in this last case, cognition is trivially seen as being performed by neural networks but is not computation and the approach we need to use to study cognition has to be related to the theory of dynamical systems).
The deep difference between the idea of scientific knowledge as simple imitation
, as developed by digital machines and their “modeling capacity”, and the idea of scientific knowledge as generation of rational intelligibility
is a core problem of the dynamical approach [40
]. As I already said above, Turing himself emphasized the difference between the simple imitation
capacity of machines and the modeling
power of mathematics: the double pendulum, which is a perfect deterministic machine, only expressed by two equations, is sensible to minor variations, below the threshold of observability: it is a typical chaotic deterministic system, and it is extremely difficult to represent it by a mimetic machine [16
]. Longo also notes that this is a system sensitive to initial conditions. It can instead described using the mathematics of deterministic chaos, in which determination does not involve predictability: in this system, a process does not follow the same trajectory even if we reiterate it with the same initial conditions, within the limits of physical measures (ibid.). Furthermore, we can also usefully observe that, in the case of Turing machines that simulate such systems, when we restart after having processes analog simulations, using the same initial data, the same already seen trajectory is performed and, on the contrary, the “real” pendulum behavior is completely different, when we restart the pendulum never performs the same trajectory.
Only “mathematical” models can explain the structure of physical causality of the a system at play, and digital simulation only resembles
causality, always realizing Laplacian and predictable processes. The explanation of this conundrum resorts to the fact that, at the roots of digital data, there is a discrete
topology, but physical measurement is always an interval that is optimally represented by continuous mathematics. Even if sometimes the use of mathematical equations in physics does not have a predictive power, this provides a knowledge characterized by a rich qualitative
epistemological value, able to make intelligible the physical causality of the system [41
]. Finally, we have to note that, with respect to physics, in the case of biological organisms, the gap between simulation and intelligibility is even worse because the variability is dominant (In [42
] Longo contends that, “in biology, in particular, the introduction of information as a new observable on discrete data types has been promoting a dramatic reorganization of the tools for knowledge” and that some consequences of this effect have been induced in life sciences, with particular emphasis on research on cancer).
In summary, digital simulation—even if epistemological useful at the level of intermediate modeling during the processes of scientific research and discovery, as I have already indicated above in this section—produces an epistemologically distorting result due to its simply mimetic quality.