The Strong Thesis for Artificial Life has a problem
The problem is pessimism :(
[This is a parred down (and heavily stylized) version of a paper I’m working on for an undergrad conference. Kind of out of theme with what I have and will continue to post, but I think it’s a fun argument, and you get to see some cool moves in Philosophy of Science, so thought it was worth typing up nonetheless.]
What is A-Life?
Artificial life is the study of artificial systems that exhibit the behavioral characteristics of living systems. Put another way, A-life is the study of different kinds of very cool robots. You have fleshy robots (wet artificial life), you have normal robots (hard artificial life) and you have digital robots (soft artificial life), all of which exemplify various paradigmatic features of living things. Neato, cool, why do you care? Well, in the field of A-life there is a pair of two thesis, one of which is quite controversial, the other less so.
The Weak vs the Strong Thesis
The Weak Thesis states that artificial organisms (those things studied by the field of A-life) can be genuinely informative about traditional living systems. This is largely non-controversial, so I’ll gloss over it a bit here. The big kahuna is the Strong Thesis (ST): Artificial organisms can be genuinely alive. Yea, those fleshy robots could be more than just robots, they could be fleshy robots that are alive like you and me! ST was articulated around the conception of the A-life field with Christopher Langton’s paper “Artificial Life” where he justified the thesis under a functionalist understanding of life.1 The basic idea was that life is a substrate neutral property meaning it could be actualized in digital organisms just as well as biological ones, so there was no reason to think (in principle) that artificial organisms couldn’t be genuinely alive. Now, whether or not Langton was right about a functionalist understanding of life being correct, the important point to note here is that he motivates the Strong Thesis by appealing to a particular understanding of life. He moves from saying that life is functional, to arguing that digital organisms can fulfill all the functions, to concluding that artificial organisms can be genuinely alive. This is pretty standard fair for motivating ST, it’s the way the dialectic typically proceeds: the proponent of ST (or the critic) presents a definition of life, and claims that artificial organisms can satisfy all the necessary and sufficient conditions, and then their interlocuter argues that at least one of those conditions can’t be fulfilled by artificial organisms. Then they go back and forth for a while until both get tired and decide it would be much more fun to watch BattleBots.
One last preliminary note about ST, it’s important to recognize it’s global character. What I mean by that is, if ST is true than if anyone, anywhere, runs into an organism (artificial or otherwise) that thing could be living. Or, put another way, if that thing is artificial, it doesn’t rule out it being living. This may seem pretty obvious but it’s important to note so we can head off some objections later on.
And everything was fine… until philosophers attacked!
Okay so we know what the Strong Thesis is, and how to prove it true or false, right? Wrong, bucko! This standard articulation of ST, and the method of confirming/disconfirming it is contingent on having the correct definition of life. Now, the debate over what definition of life is the right one is a whole thing in the field of artificial life (Carl Sagan actually wrote up this super nice piece summarizing the popular ones), but thankfully adjudicating which definition of life is right is ultimately irrelevant to the point I’ll be making today. Instead, we are going to focus on views that say something like “No matter what we do right now, we won’t be able to get close to a univocal, coherent definition of life.” There are two main papers that I’m interested in that articulate views like this, one is by Cleland and Chyba and the other is by Edouard Machery. Call the people trying to give an adequate definition of life ‘life definitionists.’ The two views I’m going to talk about here cover a good range of ways in which we should be skeptical of the life definitionist project, and I think if you find either view plausible, the project of proving ST is either impossible, incredibly difficult, or ST becomes super duper uninteresting. A quick note, I’m going to be focusing on outlining these views more so than motivating them. Both papers are quite digestible even without a ton of background knowledge, so if neither of these views strike you as plausible on a read through of this post, I recommend digging through them. Alright, lets go over those views, then look at the implications.
Cleland and Chyba (Life as a natural kind)
Okay, firstly we need to get a bunch of terminology out of the way, but it’ll be rather quick. Firstly, an ideal definition is the conjunction of all necessary (those conditions which must be met to be part of the definiendum/extension of the concept being defined) and all sufficient (those conditions which, if met, make whatever meets them part of the definiendum) within reasonable constraints of vagueness. As an example, they provide the idea of a bachelor as an “unmarried man”. Ideal definitions work well for answering ‘what is’/identity questions about characteristically ‘human’ terms (i.e. bachelor, birthday cake, bouncy ball, etc) but do not supply good answers to the identity of natural kinds. A natural kind is a category that is ‘carved up’ by nature, rather than through our own demarcation points generated through linguistic convention.2 For instance, to provide an answer to the question “What is water?” (that is, to provide the correct answer that water is H20), one needs a sufficient theory of water that goes beyond merely its surface level sensible properties.
Alright, terminology shmerminology, what’s the view? Well, C&C want to say that life is a natural kind term. If you find that plausible then we’ve got a big issue on our hands, because the project of trying to generate an ideal definition for life is entirely wrong headed (as articulated above, ideal definitions are very hard to come by for natural kind terms). Worse still, we don’t have any sort of ‘theory of life’ that is analagous to the theory we have for chemistry. That is to say, we don’t have a theory that’s going to spit out something like “Life is XYZ” the way that molecular theory spits out “Water is H20”. If we don’t have the theory then we can’t define the term, at least not adequately. We would acting like medieval alchemists trying to define liquids in terms of their macro properties (transparency, solvency, and so on).
Well that’s easy then, just develop a theory of life, spit out the definition, and problem solved, right? NO! You fool, rube, cretin, etc.! If only it were that easy. C&C are generally skeptical that any such theory is going to be forthcoming because of something called the N=1 problem. The thought here is that all life on earth is going to (plausibly) be explained by Darwinian evolution applied to a single common ancestor. If that’s right, it’s not clear whether the features we observe are going to be essential features of life per se or essential features of life on earth. In fact, it seems like the only way to figure out which of those is true would be to have a definition of life that could do the necessary classificatory work, but to get that definition we need to solve the problem in the first place! The upshot of C&C’s view is this: Life constitutes a ‘natural kind’ term, others have been seeking an ideal definition for life which is antithetical to its nature as a natural kind term, and we lack a sufficient ‘theory of life’ to analyze the relevant natural kind.
Machery
Onto Machery who has a more strongly pessimistic view about the life definitionist project. Machery wants to say that the life definitionist faces a dilemma: Either life should be understood as a folk concept, or as a scientific one. If it is the former, Machery thinks life's indefinability follows, and on the latter horn we can conclude that life is polysemous. The argument for life’s indefinability from being a folk concept is that, on all the popular psychological accounts of folk concepts, they lack definitions (according to Machery). Further, even if we wanted to try to capture the intuitive judgements about what is alive and what is not via a definition we likely would do so in an arbitrary manner. The argument from life being a scientific concept to there being no singular definition of life is that there is a glut of scientific definitions across various fields (evolutionary biology, astrobiology, synthetic biology, etc.), an “embarrassment of riches” of definitions which are extensionally distinct. The idea is that the wide variance amongst the natural sciences in what is appropriate as a candidate definition of life implies that it is unlikely they will all converge on one unified definition. The upshot for any life definitionst is grim. Either the entire project of defining life is misdirected, on account of there being nothing which is apt to be defined in the first place, or there is no univocal usage of ‘life’.
So what’s the problem?
Alright, we’ve got two views on the table, C&C’s and Machery’s. These two views give us three possibilities for the life definitionist project: A. No definition of life will be sufficient until we have developed a ‘theory of life’ (and such a theory does not seem to be forthcoming), B. Life is not definable full stop, C. ‘Life’ is polysemous with different meanings depending on the discipline of use. If A is right, then the standard methodology for the Strong Thesis is bunk, and it’s not even clear that we can state ST properly. If none of our definitions of life are going to properly capture the natural kind we want until we develop a sufficient theory, then any statement of ST will involve a notion of ‘alive’ that doesn’t actually track the natural kind we want it too. So if you accept A it doesn’t seem like you can have a view about ST, at least in so far as ST is supposed to be talking about life the natural kind rather than the insufficient definitions we attempt to generate to track that natural kind, prior to the discovery of an adequate theory. Alright, well obviously we are going to have the same but much worse problem with B. If B is right, then there is no adequate definition of life that can be provided at all, there isn’t even the hope of a sufficient theory. So if you accept B you definitely can’t take a view on ST. Alright, last option, what about C? Well, C at least gives you definitions to work with. We can consider, say, the astrobiological definition of life as legitimate in that field, and use that to construct a Strong Thesis. But notice, that Strong Thesis is going to behave quite differently from the Strong Thesis we have been talking about. It’s domain specific, that is to say, even if the ‘Strong Thesis of Astrobiology’ holds, this doesn’t say anything about the ‘Strong Thesis of Synthetic Biology’. So this supposed Strong Thesis behaves very differently from the Strong Thesis we were initially interested in. Thus, this view at least seems to concede that ST, as originally conceptualized, turns out to be indeterminable, even if we can ‘salvage’ it into domain specific variations indexed to the various usages of ‘life’ in these fields.
Some objections
I’m going to maintain my abbreviation of the views as A, B, and C from earlier so:
A. No definition of life will be sufficient until we have developed a ‘theory of life’ (and such a theory does not seem to be forthcoming).
B. Life is not definable full stop.
C. ‘Life’ is polysemous with different meanings depending on the discipline of use.
Someone sympathetic to A might say that even though we don’t know the natural kind now, we could still make progress towards a correct definition. In that progress, we might come across some necessary condition which accords with the definition that would be spit out by the correct theory of life. Then, with that necessary condition in hand, we can examine whether or not an artificial organism could satisfy that condition. If no artificial organism could satisfy that necessary condition, then we’ve successfully falsified ST!
Response: It’s not clear why we would think that any particular condition we cook up is the correct one without the general theory of life, especially considering C&C’s reasons for doubting our coming to that theory in the first place. The N=1 problem undermines our ability to generalize from examples of life we observe on Earth to life in general. Given that objection, we have reason to doubt that any candidate necessary condition for life would actually apply to life in general.
A proponent of C has, I think, the most promising response by saying that we can generate variations of ST, each indexed to a different usage of ‘life’ depending on the relevant discipline. These various STs (italicized to notate the difference) can then be stated in the standard way, confirmed or disconfirmed with the standard methodology, and so on. But as I mentioned before, this just seems to change the subject. STs differ from ST in that they are domain specific. Settling an ST doesn’t settle any of the other ones. In contrast, if you get an answer to ST as it’s standardly put forward (where ST is defined in terms of a univocal notion of ‘life’), then you settle the issue in all the scientific domains. So it seems like, if we make this move, we concede that ST is at least indeterminable. Then, the various STs seem much less interesting given their domain specificity. That’s not to say there is nothing interesting there at all, but they are certainly of less interest than the original statement of ST. So this move, while fruitful, still constitutes a major concession.
So, no matter which way you go, I think this sort of pessimism about life definitionism should leave us deeply skeptical of our ability to confirm (or disconfirm) ST, at least for the moment.
This paper is kind of a nightmare to dig up, but you can find it in The Philosophy of Artificial Life edited by Margaret Boden, it’s different from the paper that will pop up if you just google ‘Christopher Langton Artificial Life’ that is also published in 1986.
I’m fulfilling my philosophical imperative to put a footnote to Plato here