Second-Order Cybernetics Maintaining Essential Variables Is Not Passive Matthew Egbert

« 8 » Proponents of the enactive ap- sible failure to do so will necessarily result “ the survival standard itself is inadequate for proach agree that the homeostat’s double in the homeostat’s death” (§10). However, at the evaluation of life. If mere assurance of per- feedback architecture made an important the level of its physical body, an organism is manence were the point that mattered, life should contribution, but at the same time they always far-from-equilibrium with respect to not have started out in the first place. It is essen- struggle to overcome its limitations (Ike- its environment. Falling into an equilibrium tially precarious and corruptible being, an ad- gami & Suzuki 2008). Franchi briefly refers is the same as dying, because the organism venture in mortality, and in no possible form as to evolutionary robotics models inspired by would lose its ability to do the work of self- assured of enduring as an inorganic body can be. this ultrastability mechanism, but he does producing its own material identity, i.e., the Not duration as such, but ‘duration of what?’ is the not mention that further progress has been very process that ensures that the double question.” (Jonas 2001: 106) difficult. Ezequiel Di Paolo (2003) showed feedback loop between behavior and inter- that implementing Ashby’s mechanism is a nal homeostasis is intrinsically connected Tom Froese graduated with a D.Phil. in cognitive science significant step toward more organism-like within a whole. Only non-living matter can from the Univ. of Sussex. He is currently a postdoctoral robots, but the contingent link between (in- be in physical equilibrium with its environ- researcher at the National Autonomous University of ternal) homeostatic mechanisms and (ex- ment. Franchi conflates the living with the Mexico. He investigates the relationship between life, ternal) behavior is a continuing source of non-living in another way when he notes mind and sociality using a variety of methods, including problems. that the homeostat “will continue acting the evolutionary robotics, systems theory and phenomenology. « 9 » Switching the connections of an way it normally does until an outside force ultrastable system may in some cases lead compels it to change course of action” (§26). Received: 2 October 2013 to an adaptive re-organization of original This is precisely the way in which Newton’s Accepted: 18 October 2013 behavior, as Franchi’s model of two self-con- first law of motion describes the behavior of nected homeostat units shows, but it is far objects. However, living beings – subjects from guaranteed in more complex systems. – can change their behavior even in the ab- It is possible that the switch is not sufficient sence of a change in external conditions. For Biological Systems, to “break” the essential variables, thereby Maintaining Essential failing to provoke an adaptive response at Overcoming the autonomy/ all, or, if adaptation does occur, the new heteronomy dichotomy Variables Within Viability equilibrium may not reestablish the origi- « 12 » Franchi enticingly hints at the Limits Is Not Passive nal behavior. From Ashby’s perspective, this possibility of a “reevaluation of the tradi- failure to recover the original behavior is not tional distinction between heteronomous Matthew Egbert a problem, since he simply equates all equi- and autonomous behavior” (Abstract). Is University of Hertfordshire, UK libriums with survival – without asking how perhaps the relational view of life defended they might differ in terms of desirability (an by the enactive approach (e.g., Di Paolo mde/at/matthewegbert.com idea that is later echoed by Maturana’s claim 2009) such a reevaluation? As I see it, there that autopoiesis does not admit gradations). is a natural development of ideas from first- > Upshot •The target article proposes « 10 » However, this neutrality is not order cybernetics via second-order cyber- that Ashby’s investigations of the ho- found in nature. For example, inverted netics toward enactive cognitive science. meostat and ultrastability lead to a view goggle experiments can be safely conducted The early approach was too heteronomous, of living systems as heteronomous, pas- 109 without any irreversible consequences to the while the second-order approach was too sive “sleeping” machines and thus are participants. How then is the appropriate re- autonomous, but the relational stance of in fundamental conflict with concepts configuration achieved? There have been a the enactive approach formulates a dialectic of autonomy developed by Jonas, Varela couple of attempts at addressing the practi- between these two extremes (Froese 2011; and others. I disagree, arguing that (1) cal problems stemming from contingent de- Froese & Stewart 2013). Relatedly, it aims the maintenance of essential variables pendencies with improved evolutionary ro- for a middle ground between the extreme of within viability limits is not a passive botics models, in particular by more closely computationalism, which explicitly repre- process for living systems and (2) the tying the desired behavior to the satisfaction sents goals, and the extreme of Ashby’s (and purpose of Ashby’s investigations of the of the homeostatic conditions (Iizuka & Di Maturana’s) insistence on absolute non-con- homeostat was to investigate adaptiv- Paolo 2008; Iizuka et al. 2013). However, tingency. Crucial for these two theoretical ity, a subject that is related to, but clearly these attempts have met with only partial shifts is to go beyond traditional theories distinct from, autonomy. As such, I find success. that portray life’s ultimate purpose, no mat- Ashby’s work on adaptivity to be neither « 11 » We need to move beyond the tra- ter whether by active or passive means, as in opposition to nor in direct support of ditional fixation on equilibrium dynamics in equilibrium, stability, or survival. As Hans modern concepts of biological autono- order to make real progress on these issues. Jonas recognized, only when we face up to my and suggest that a productive way According to Franchi, an Ashbyan organ- the essential precariousness of living exis- forward involves the investigation of the ism is continuously required to behave so as tence can we hope to understand the mean- intersection between these two funda- to go to equilibrium, since “its always pos- ing of life: mental properties of living systems.

http://www.univie.ac.at/constructivism/journal/9/1/093.franchi 110 Computational Concepts in Second-Order Cybernetics concepts of biological autonomy. biological of concepts with work Ashby’s integrates that research subsequent the of and Ashby’sresearch, of emerges view consistent more a general, in life of concept adaptivity,broader a not and “just” targeting as seen is that homeostat the below if argue will I (§11). it embodies” principles whose homeostasis, eralized gen as life of view a for proxy a is “Ashby’s device life: of view broad a targeting as homeostat the seeing interpretation, ferent dif a have to appears Franchi book. the of for Braina works and of those others. frame- own their of implications the investigate methods for philosophers to explain, criticise and new valuable provide rather are but philosophy, hide not do models computational where light, see I covers” (§4). scientific-looking under presented are they unless frameworks conceptual abstract from away shies that age positivistic-minded a to suited more garb technical a in presented […] philosophy disguise (also) in is cognitive sciences the in work computationally-intensive technical seemingly “sometimes, that comment Franchi’s bevs ht hs os o tk it ac that behaviour the into of complexity the count take not does this that observes Franchi (§10). possible” as quickly as sleep to back go to is function ideal its machine “because sleeping a is homeostat the that Walter’s comment Grey William with gins be heteronomous and passive is meostat intended to be a model model a be to intended is homeostat the concept overwhat Franchi with disagree may However,I brain. the of model a than rather concept a of model a as considered is it when productive most is homeostat the that Franchi with agree I tary, provided that they are not conflated. not are they that provided tary, complemen be fact in can and compatible in not are approaches two the that rightly, Heobserves, constructs. theoretical explain cepts, and systems, real-world of understanding our of accuracy the evaluate tween pnn snec of sentence opening the is behaviour?” adaptive produce brain the does “How adaptivity. investigating for

C truons computationalduplicitousless philosophya in « 1 » 1 « 2 » 3 « » 2 « | which are used to refine, explore or explore refine, to used are which models of objects of models As a side note, I do not entirely agree with c

tfn Faci itnuse be distinguishes Franchi Stefano t rnh’ agmn ta te ho the that argument Franchi’s I see the homeostat as a system system a as homeostat the see I i v (1960), and, in my and, in view, (1960), goal the t F t is ou nda t R ion of. , s Ashby’s oss

which are used to used are which s vol. 9,N°1 oes f con of models Design Design ------2

remain at that state.at that remain system’s the tendency teract inherent coun to needed is effort constant – asleep or passive as pole the balancing system the describe to mistake a be would it but tical, ver roughly remains pole the – tips finger your on pole vertical a balancing of act the metaphor, a as consider, We can 67). 1960: volve the vigorous activity of others” (Ashby in may variables some of “the constancy it: puts Ashby as or activity, dynamic stantial sub require can limits viability within able vari essential an of maintenance the that recognise we when apparent is (A) in error The (A). to according Franchi by being described is system this but (B), in described as performs that system a presents clearly B: A: say: to thing same the not is Itlatter. the doing as above quote the in described is butformer, the of capable “asleep.”is or homeostat The passive, inactive, being system the (2) and equilibrium) from away far are that ditions con at change, of absence an i.e., stability, for allows that and equilibrium of that than general more is that concept (a state steady tion with of the idea (1) a atbeing variable a confla a is there particular, In passive. as homeostat the of view incorrect an to leads believe I that here used language confusing some is there but passive; is non-action,”it “derives from homeostat the action because original) in emphasis “ agree: to seems he that, than other but, generate can homeostat the to go back to rest (i.e., to equilibrium). to (i.e., rest to back go to order in just – can it action whatever take will it sleep to back to order in needed is as much as act will stat homeo the disturbing: more something far philosophically means this] [and non-action from action derive to effort its in […] lies life about sis

the main source of resistance to Ashby’s the Ashby’s to resistance of source main the « 5 » 5 « terpretation of Ashby)of terpretation in (My same. the (roughly) remain to is in as order needed for an variable essential much as act will homeostat The sleep.” to §10) (Franchi back go to order in needed is as much as act will homeostat “The « 4 » 4 «

In ee Faci s ugsig that suggesting is Franchi Here, . It will go to extraordinary lengths, lengths, extraordinary to go will It . ein o a Brain a for Design Ashby , ” not (§10, (§10, to go go ------viability limits as passive being is amistake. T limits. viability within maintained be to are organismic ongoing require all these that clear is it blood, the in oxygen and sugar of levels the temperatureor body as such variables, essential archetypalabout think weIf 1944). Schrödinger (Gánti2003; degradation their counteract to material resources and energetic of ongo acquisition- ing require that far-from-equilibrium structures dissipative are organisms All systems. living in than evident vari- more ables essential of maintenance the for ity mous systems adaptive would be also systems. autonoof instances- real-world all case, which in persist, to are they if environment) variable a to robustness dynamical minimal (or adaptivity of and/ornaturalorganisms musthavedegree some tinct. However, it may be that in practice, artificial dis- as kept be should autonomy and adaptivity operation. They are onlylabelled as essential or existence its to essential actually not are magnets) its of state (the homeostat the of variables” “essential The autonomous. not butadaptive, is that system a ofexample an spite their overlap.spite their orthogonal are properties two The not. is autonomous,autopoieticthator one is or that system adaptive an of conceive to sible that is not adaptive; and conversely, it is - pos one or adaptive, is that autonomoussystem it is possible to conceive of an autopoietic or autonomy.way,entailanother notPut does Ashby by developed adaptivity of concept automatically entail adaptivity. Similarly, the not does autopoiesis of concept the (2005), Paolo Di Ezequiel by observed as however, vari - frameworks; both in roles central playables essential instance, for frameworks: systemitself. There these is overlapbetween emerge and how these can be satisfied bythe can needs own autonomous,its how be can frameworks for understanding how a system HumbertoconceptualMaturana,developed Francisco later and Jonas, Hans describing and studying adaptive behaviour. defining, for framework a autonomy.developed Ashby studying for developed works frame- recent relatively the Ashby’s opposes that work §§13–14 in proposition the o assume the maintenance of these within these ofmaintenance the assume o « 7» « 6» 3 | ee I m run ta te ocps of concepts the that arguing am I Here, This leads to why I disagree with disagree I why to leads This - activ of necessity the is Nowhere 3 In fact, the homeostatis the fact, In if they they if activity V arela and arela de- Second-Order Cybernetics Interpreting Ashby – But which One? David Vernon

because Ashby simply assumes the existence mechanisms of adaptivity can respond to es- position on homeostasis and that it is an of essential variables, and then investigates sential variables to prevent catastrophic sys- adequate model of the brain. how they can be regulated, without delving tem failure. Some work in this area is already into the details of the intrinsic dynamics of underway (Barandiaran & Egbert 2013; Eg- « 1 » The target article claims that Ross the essential variables (for example the ten- bert 2013; Egbert, Barandiaran & Di Paolo Ashby’s generalized homeostasis thesis en- dency of blood sugar levels, in the absence 2010; Egbert, Di Paolo & Barandiaran 2009), tails that living organisms are heteronomous of behaviour, to approach non-viable states). and further developments will not only help rather than autonomous, being controlled by When a magnet of the homeostat leaves us to understand how life differs from non- the environment rather than independently the predefined viability limits, the system life, but also how life could have originated adapting to environmental perturbations. continues to operate. For this reason, the (Ruiz-Mirazo, Pereto & Moreno 2004). In the following, I will explain why such a homeostat itself is actually heteronomous, conclusion is consistent with the second but because the homeostat is a model of an Acknowledgements edition of Ashby’s book (1960). However, I adaptive system and not a model of an au- I would like to thank Kepa Ruiz-Mirazo will also argue that an interpretation of the tonomous or living system, the heteronomy for our discussion related to this commen- first edition (Ashby 1954) can lead one to of the homeostat says nothing about the het- tary. the opposite conclusion, one that supports eronomy or autonomy of living systems. autonomy rather than heteronomy and one « 8 » For Ashby’s investigations, it suf- Matthew Egbert: “A prominent theme in my research is that is compatible with the principles of au- ficed to consider the essential variables in the recursive self-maintaining organisation of life, and topoiesis. I also wish to highlight the prior- only very abstract terms – a variable that how this relates to its profound adaptability. This draws ity of multistability over ultrastability, and must be maintained within limits; nothing me toward the investigation of synthetic protocells the associated limitations of the homeostat more. Because the essential variables were and the origin of life, where mechanisms of behaviour and simple ultrastability (which Ashby him- not included in detail, the homeostat is not are at their simplest, and towards minimally cognitive self acknowledges, albeit in a more obvious particularly effective at demonstrating the robotics and simulated agents, where fundamental manner in the first edition.). dynamic nature of the maintenance of es- concepts underlying adaptive behaviour and autonomy « 2 » In the following, when referring sential variables. It was therefore possible can be clearly defined and examined. As a Research to work in Design for a Brain, first edi- for the homeostat to be mistaken for a sleep- Associate at the University of Hertfordshire, I am tion (Ashby 1954), I will adhere to Ashby’s ing, passive machine that does everything currently investigating self-maintaining behaviours convention of using, e.g., S. 3/9 to refer to it can “to do nothing.” If essential variables in autonomous robots and simulated agents.” Chapter 3, Section 9. I will add a leading were modelled in more detail and the in- Homepage at http://www.matthewegbert.com superscript (1S. 3/9 vs. 2S. 3/9) to differen- trinsic dynamics of the essential variables tiate between the first and second editions, of dissipative structures such as life were Received: 14 October 2013 respectively. I have also retained Ashby’s use included, it would have been more obvious Accepted: 18 October 2013 of italics; words originally in bold face type that Ashbian adaptive regulation, whenever are underlined. employed by a biological system, must be anything but passive. Heteronomy vs. autonomy « 9 » For reasons such as these, there Interpreting Ashby – « 3 » In the abstract, Stefano Franchi needs to be more work modelling the home- But which One? states that Ashby’s thesis “entails that life is 111 ostat and its interesting form of adaptation. fundamentally ‘heteronomous.’” While Ash- In the latter section of the target article, David Vernon by does not use the term “heteronomy” in Franchi pointed out that Ashby’s “environ- either of his books, this conclusion follows University of Skövde, Sweden ments,” when simulated as homeostat units, naturally from Ashby’s development of gen- included an inappropriate, or at least odd, david.vernon/at/his.se eralized homeostasis in the second edition property of self-regulation. This is an impor- (2S. 5/6), according to which an organism tant observation that raises questions that > Upshot • The association of heterono- and its environment form a single state- can be investigated using models such as my with Ashby’s work in the target ar- dependent system (2S. 3/9, 2S. 3/10), the that presented in the target article. Similarly, ticle follows from a direct interpretation variables of which include a set of essential I have argued here that there are assump- of the second edition of Ashby’s book variables, the value of which much be kept tions implicit in Ashby’s work concerning Design for a Brain. However, the first edi- within certain bounds if the organism is to the nature of essential variables that need tion allows for an alternative – opposite survive (2S. 3/14). to be made explicit and investigated. To un- – interpretation that is compatible with « 4 » Homeostasis, the process of regu- derstand how adaptive behaviour relates to autonomy and autopoiesis. Furthermore, lating the essential variables, requires the autonomy and agency, we need to develop a more balanced perspective is suggest- organism to adapt to its environment to a more sophisticated understanding of es- ed to avoid unintentionally giving the achieve stability, i.e., to keep the essen- sential variables, their intrinsic dynamics, casual reader a misleading impression tial variables within physiological lim- the emergence of viability limits and how that the homeostat is Ashby’s ultimate its (2S. 5/3). A region in the system’s field

http://www.univie.ac.at/constructivism/journal/9/1/093.franchi