Extinction, resolved

Editors' Note

Don't worry: we're not extinct yet!

The past two years have been incredibly productive for the editors of and contributors to this blog. We've had a lot of informative discussions, a few great meetings, and some unanticipated opportunities. In order to make the most of that productivity, we'll be changing the publication schedule for Extinct. Instead of the weekly schedule that we've maintained since the blog launched in 2016, we're switching to a monthly schedule in 2018. Look for a new post on the first day of each month!

Even though we're publishing less frequently, we're still on lookout for guest contributors. If you have any ideas or suggestions, don't hesitate to contact the editors.

Want more philosophy of paleontology than a monthly publication schedule can provide? (We know how you feel!) Our very own Adrian Currie is releasing his guide to the historical sciences, Rock, Bone, and Ruin, later this month! Be sure to order a copy. And if you still want more, feel free to peruse our extensive archives!

With all that said, Leonard Finkelman writes...

Introduction: The Thylacine returns

Suppose I offer you a choice between two bets on the year to come. (I mean 2019, of course, given that 2018 is already here.) The first bet is that I'll write an essay about thylacines in 2019. The second bet is that thylacines won’t be rediscovered in the wild in 2019. Which bet should you take?

Your choice is likely informed by your estimation of probabilities. If the probability that I’ll write an essay about thylacines is greater than the probability that there will be a thylacine sighting, then your rational choice would be the first bet. I wrote essays about thylacines in 2016, in 2017, and now in 2018 (and on my first opportunity, to boot!), so you can imagine the probability that I’ll write another in 2019 is pretty close to 100 percent. But is it closer to 100 percent than the probability that thylacines are well and truly extinct?

Comparison between skulls of a thylacine (denoted by a red bar) and a gray wolf (denoted by a green bar). Image from Wikimedia Commons.

Comparison between skulls of a thylacine (denoted by a red bar) and a gray wolf (denoted by a green bar). Image from Wikimedia Commons.

Biologists have developed several models to estimate the probability that a given species is extinct (Solow 1993; Solow & Roberts 2003; Marshall 2010; Bradshaw, et al. 2012; Fischer & Blomberg 2012). According to these models, the probability that any thylacines remain to be discovered is between zero and three percent. The probability of your winning the second bet, then, lies between 97 and 100 percent (and it’s likely much closer to the higher number). Place your bet accordingly.

I like to write about thylacines because the species (Thylacinus cynocephalus) provides a rare opportunity to think very clearly about extinction. As I’ve argued in previous essays about thylacines, the concept of extinction is a difficult and muddled one. This essay will be different from those: drawing from the statistical work I mentioned above, I hope to offer something more constructive. I hope to resolve the problems I raised earlier. The key, I think, is to recast the debate—not in terms of what happened to thylacines themselves, but in terms of how we can study them.

Extinction: The metaphysical problem

Keep three things in mind as we consider different ways to think about extinction:

  1. The last confirmed sighting of a thylacine in the wild was in 1933 and wild populations likely disappeared by 1935.
  2. The last captive thylacine died on 7 September 1936.
  3. As of this writing (30 January 2018), there remain several well-preserved thylacine fetuses from which geneticists can cultivate and have cultivated genetic material. (I recall that the number may be eight, but I can’t find confirmation.)
Preserved thylacine pup and fetuses. Image from  National Geographic .

Preserved thylacine pup and fetuses. Image from National Geographic.

Now consider the three different ways to conceive extinction that I’ve discussed in previous essays (see also Delord 2007):

  • A species is functionally extinct if and only if the members of the species are practically incapable of perpetuating any lineages. This is what happens to (say) a sexually reproducing species when its population consists entirely of organisms belonging to a single sex or if it the population has been reduced to an endling.
  • A species is demographically extinct if and only if every member of the species is dead. This has also been called “final extinction” and (presumptuously) “true extinction."
  • A species is substantially extinct if and only if the information necessary to produce new members of the species has disappeared. Technological advances such as those promised by resurrection biology may change the standards for substantial extinction over time.

When did thylacines go extinct? That question is ambiguous: "extinct" might mean any one of the three concepts defined above. The answer is also ambiguous: it depends on the kind of extinction. T. cynocephalus went functionally extinct as late as 1935. The species was demographically extinct on 7 September 1936. By the standards of substantial extinction, the species isn't extinct at all. This ambiguity has the potential to generate a variety of problems, but two in particular strike me as important.

The first problem with the heterogeneity of extinction concepts is a practical one for paleontology. Given the imperfection of the fossil record, the extinction of fossil species is always ambiguous between functional and demographic extinction. This ambiguity produces the Signor-Lipps effect: since the most recent specimen in a fossil species is unlikely to be the biological population's endling, or may not even signal a decline in the population size, paleontologists should assume some lag between the fossil species' latest appearance and the biological population's extinction--which biases our reading of the fossil record against abrupt extinction events (but see below). This makes interdisciplinary work difficult: for example, an inability to compare past and recent extinction rates would complicate conservation efforts.

The second problem with the heterogeneity of extinction concepts is a more metaphysical one. Given differences between definitions for the three extinction concepts, differences between the implications of those definitions, and likely differences in their underlying processes, it's disingenuous to use a single word to describe them all. "Extinction" doesn't seem to have an essence. To say something like, "Thylacines are extinct," would therefore be (strictly speaking) nonsensical because "extinct" per se is not a property of anything at all. In other words, there is nothing in common to all and only the species we describe with that term.

That's a pretty big problem since extinction is supposed to be a concept that unifies almost all of life's history. As paleontologists sometimes point out, more than 99% of all species are extinct, after all.

Extinction: The epistemic problem

As I was writing this post, the U.S. Fish and Wildlife Service declared the eastern cougar (Puma concolor cougar) an extinct subspecies. The last sighting of an eastern cougar was in 1938, five years after humans last laid (reputable) eyes on a wild thylacine. There was also a considerable delay before T. cynocephalus officially joined the ranks of extinct species: the International Union for the Conversation of Nature didn't declare thylacines extinct until 1982.

Caution explains the long wait in both cases. Species are considered endangered, and subject to all relevant protections, before they're declared extinct. That status and those protections are stripped away once the species' extinction is recognized. Declaring a species extinct is therefore an important step with significant consequences for conservation groups. One doesn't want to stop trying to protect an irreplaceable branch of life's proverbial tree until they're sure that there isn't anything left to protect.

The IUCN declared the thylacine extinct after five decades without a sighting, but the USFWS waited an additional three decades to declare the eastern cougar extinct; why the discrepancy? The answer comes down to epistemology. By the time it went extinct the thylacine's range had been reduced to the wilds of Tasmania, and by the time it was declared extinct the species had been the subject of several concerted search efforts. By contrast, the eastern cougar ranged across the eastern seaboard of the United States--a much larger area than Tasmania--and efforts to find remnants over the last eighty years haven't been particularly extensive. By 1982, there was much less reason to hope for thylacine remnants than there was to hope for eastern cougar survivors.

Andrew Solow (1993) statistically formalized this kind of reasoning so that scientists could minimize uncertainty over when a species has gone extinct. Given a sufficiently extensive record of species sightings--that is, an ordered list of confirmed dates on which someone saw a member of the species--one can develop a statistical model that estimates the frequency of sightings and predicts when one should expect to see a member of the species again. As more time passes beyond an expected sighting date without a sighting, the probability that someone will see another member of the species decreases--and once the probability approaches zero, one can be reasonably sure the species is extinct. It's tough to quantify hope, but that's just what Solow's method does.

Using the sighting method, Solow (1993) estimated that the Caribbean monk seal (pictured above) likely went extinct before 1973. Image from Wikimedia Commons.

Using the sighting method, Solow (1993) estimated that the Caribbean monk seal (pictured above) likely went extinct before 1973. Image from Wikimedia Commons.

There are, of course, quirks and kinks to work out. How does a declining population affect the frequency of sightings? How does one factor in the difficulty or infrequency of sighting efforts? What if members of the species are camouflaged or something? Recent work has modified the modeling process to add these subtleties (see, e.g., Solow & Roberts 2003). Fischer & Blomberg (2012), for example, combined the sighting record for thylacines with ecological niche data and population size estimates to infer that the probability of finding a wild thylacine had diminished to near zero by 1935. 

Solow & Roberts (2006) recognized that this method could also be useful in paleontology. Substitute "dated occurrences in the fossil record" for "record of species sightings" and paleontologists can use similar models to estimate true extinction dates for fossil species. After accounting for preservation bias and other taphonomic features, paleontologists can make reasonable inferences about the lag between a fossil species' last appearance in the fossil record and the biological population's extinction, thereby minimizing the Signor-Lipps effect (see also Marshall 2010; Bradshaw, et al. 2012).

Two points are important here. The first point is that the same basic logic governs all estimation of extinction dates, whether those extinctions are ancient or recent. That logic should be familiar to philosophers: it is, essentially, enumerative induction. The second point is that all extinct species do have something in common. It turns out that common feature isn't a metaphysical property of the species; rather, it's an epistemic property of reasonable observers. A species is extinct if and only if we can't reasonably hope to see it again.

Conclusion: What really goes extinct, anyway?

Can we hope to see the thylacine again? People will certainly try. Still, the best answer to that question is one that quotes the wizard Gandalf, by way of Tolkien: "There never was much hope... Just a fool's hope." That's why thylacines are extinct.

One might call this misplaced attribution or affirming the consequent or some other horrifying dereliction of philosophical duty, but I do think there's an important lesson about extinction to be drawn here. One feature common to all extinction concepts is the improbability of observation; where the concepts differ, they differ in the degree of improbability. If one could measure such a thing as the global probability--that is, the probability of any random observer in any random place, quantified over all observers in all places--of encountering a thylacine, then that probability approached zero in 1933, decreased ever so slightly in 1936, and will bottom out if (well: when) de-extinction efforts fail. There may be different underlying processes that account for those probability shifts, but the extremely low probability of encounter is nevertheless common to all extinct species.

What this means is that we can resolve the metaphysical problem of extinction by way of resolving the epistemological problem. Extinction is problematic if conceived as a property of species per se, but it isn't problematic if conceived as a relation between a species and its observers. An extinct species is one that can't be observed. This may raise a host of questions about what constitutes observation, but that's an essay for a different blog--you know, one not named "Extinct."

This resolution suggests a sobering conclusion that's worth bearing in mind as the year 2018 kicks into gear: species aren't what really goes extinct. Our hope does.


Works cited

  • Bradshaw, C.J.A., Cooper, A., Turney, C.S.M., & Brook, B.W. 2012. Robust estimates of extinction time in the geological record. Quaternary Science Reviews 33: 14-19.
  • Fischer, D.O. & Blomberg, S.P. 2012. Inferring extinction of mammals from sighting records, threats, and biological traits. Conservation Biology 26(1): 57-67.
  • Marshall, C.R. 2010. Using confidence intervals to quantify the uncertainty in the end-points of stratigraphic ranges. The Paleontological Society Papers 16: 291-316.
  • Solow, A.R. 1993. Inferring extinction from sighting data. Ecology 74(3): 962-964.
  • Solow, A.R. & Roberts, D.L. 2003. A nonparametric test for extinction based on a sighting record. Ecology 84(5): 1329-1332.
  • Solow, A.R. & Roberts, D.L. 2006. On the Pleistocene extinction of mammoths and horses. Proceedings of the National Academy of Sciences 103(19): 7351-7353.

Four Features of Historical Counterfactuals

Derek Turner writes . . .

In 2015, the following meme flashed across social media:

Folks on the political left shared this in large numbers, and with considerable righteous indignation. People complained that in all their years of schooling in the United States, no one had ever shown them this map! Why weren't kids taught about the Native American polities that were diminished or erased by Euro-American colonization and settlement? But then someone pointed out that this map was never intended to represent any actual historical political landscape. Oops. From the very beginning, it was an exercise in counterfactual history. What political units would exist in 2015, had the epidemics, the violence, and the displacement not occurred? 

Those who didn’t already have egg on their faces quickly responded that even as an exercise in counterfactual history, the map is both far-fetched and ethically problematic. It’s ethically problematic because it still represents the erasure of many Native nations and communities that exist today.  It also builds in an assumption that indigenous polities would develop into units that look like European nation states. And don't call the Puebloans of the southwest the "Anasazi." Here, though, I want to focus more on the epistemological than on the ethical problems. What the map represents seems far-fetched and speculative. I’m very confident in saying that. But what exactly are the grounds for making such an assessment?


A Multidimensional Epistemology for Historical Counterfactuals

In general, historical counterfactuals take the following form:

If some upstream condition A had been different, then downstream outcome C would have happened instead of what actually happened.

In an earlier post, I tried to make a little headway by drawing a distinction between preventing and enabling conditions. Some counterfactual claims are about upstream conditions that prevent downstream outcomes from happening, while other counterfactuals are about upstream conditions that make it possible for downstream outcomes to occur.

Here I carry the analysis further by distinguishing four other dimensions along which historical counterfactuals can vary. I’ll call these upstream departure, downstream departure, historical depth and historical slack.

Variation along these dimensions can affect whether counterfactuals are reasonable or (like the one above) far-fetched. My view is that it really is possible to assess counterfactuals, even when it comes to the deep past. There is, however, no easy formula for doing so.[1] The assessment is always holistic and often messy. But identifying these dimensions of variation is a good start. 


Upstream Departure

When thinking counterfactually, we can imagine bigger or smaller deviations from what actually happened upstream. To assess the degree of upstream departure, focus on the upstream conditions alone. It’s hard to be precise about the size of the upstream deviation from the actual, but some easy examples will make it clear that in practice, we often have a pretty good idea of what counts as a bigger vs. smaller upstream departure.

(1) If Senator Bernie Sanders of Vermont had won the Democratic nomination in 2016, then Donald Trump would not be president of the United States.

(2) If Senator Chris Murphy of Connecticut had won the Democratic nomination in 2016, then Donald Trump would not be president of the United States.

I don't know whether either of these statements is true, though I’m sure a lot of people believe (1). Note, however, that (2) involves a bigger upstream departure than (1). In making this judgment, we have to draw on background knowledge—for example, that Sanders and Clinton were engaged in a competitive primary race, and that Chris Murphy was not seeking the nomination at all. There’s a sense in which the departure described in (1) could easily have happened, whereas it’s very hard to see how the departure described in (2) could have happened. Claim (2) is also, I think, a lot harder to assess than (1), for the simple reason that we have less evidence about what sort of campaign Chris Murphy might have run, or how people might have responded to him.

Downstream Departure

Historical counterfactual thinking also involves imaginative departures from what actually happens downstream, and these, too, can be bigger or smaller. Thus:

(3) If Bernie Sanders had won the Democratic nomination in 2016, then he would have won the states of Michigan, Pennsylvania, and Wisconsin, defeating Donald Trump by a narrow electoral margin.

(4) if Bernie Sanders had won the Democratic nomination in 2016, then he would have won the general election in an electoral college landslide.

I'm not sure if claim (3) is true, but these sorts of claims are worth thinking about, if only because lots of people believe them. And people argue about them all the time.

Again, in order to assess the degree of departure from what actually happened, we have to consult our background knowledge. The results of popular voting in Michigan, Pennsylvania, and Wisconsin were pretty close, so (3) would seem to involve a small departure from the actual past. But popular voting results in many states, including a number of reliably red states, would have had to come out very differently in order for Sanders to win an electoral landslide. In this case, hopefully, it’s easy to see that (4) involves a bigger downstream departure than (3).


Historical Depth and Historical Slack

Historical depth is a matter of the amount of time that has elapsed between the upstream and the downstream conditions. The political examples above are all historically shallow, which also makes them a bit easier to think about. The counterfactual map of Native American political bodies is historically deeper. The counterfactuals we might care about in paleontology—like some of the claims that Gould makes in Wonderful Life—are vastly deeper.[2]

Historical Slack has to do with the relationship between the upstream change and the downstream change.[3] Here’s the rough idea: Historical counterfactuals are tight when the upstream departures from actual and the downstream departures move in tandem. That is, a tight counterfactual asserts that a big upstream change makes for a big downstream change. Or that a small upstream change makes for a correspondingly small downstream change. Historical counterfactuals exhibit more slack when the upstream departures and the downstream departures have different magnitudes—for example, where a small upstream change is said to lead to a big downstream change, or where a big upstream change is alleged to make little difference downstream. (Note: I'm treating slack as a feature of the counterfactual claims, rather than a feature of history itself.)


Rolling Back History

What I’ve said so far is really just an opening sketch. Much more needs to be said about how these four features—upstream departure, downstream departure, depth, and slack—bear on the question of whether it’s rational to believe counterfactual claims. And then much more needs to be said about how this might apply to questions about evolutionary history.

However, with nothing more than these opening distinctions on the table, it’s possible to glimpse an epistemology of historical counterfactuals that’s different from one standard approach. In his book, The Philosophy of Philosophy, Timothy Williamson describes an imaginative “rolling back” method for assessing historical counterfactuals—a method that sounds very similar to Gould’s thought experiment of replaying the tape of history, the topic of John Beatty's recent post, though Williamson seems unaware of Gould.

The offline use of expectation-forming capacities to judge counterfactuals corresponds to the widespread picture of the semantic evaluation of those conditionals as “rolling back” history to shortly before the time of the antecedent and then rolling history forward again according to patterns of development as close as possible to the normal ones to test the truth of the consequent.[4]

So we rewind the tape of history in the imagination, make the upstream changes, and run the tape forward while holding as much fixed as we can.

[O]ne supposes the antecedent and develops the supposition, adding further judgments within the supposition by reasoning, offline predictive mechanisms, and other offline judgments … To a first approximation: one asserts the counterfactual conditional if and only if the development eventually leads one to add the consequent.[5]

Apply this to the case with which we began: Roll back time, say, to 1492. Then “suppose the antecedent”—suppose that Europeans do not come to North America to colonize, settle, displace, or dispossess. Then roll history forward again, while relying upon other background knowledge. Do things end up as represented in the map above?

This account of the rolling back method is fine as far as it goes, but it doesn’t take us very far, because it doesn’t say anything about the differences among counterfactual claims, and how those differences might matter epistemically. What we need to do is contrast different historical counterfactuals, while focusing on the dimensions outlined here.


So What’s (Epistemically) Wrong With the Map?

One problem with the map, as an exercise in counterfactual history, is the magnitude of the upstream departure. The map exhibits considerable historical depth (500+ years), and a fair degree of tightness, since the suggestion is that huge upstream departures would have made for correspondingly big downstream changes in political outcomes. But the upstream departure puts out of play so much of what we know about the world, that it becomes very difficult to get a fix on what the resulting downstream departure should be.

But we should be careful not to conclude that bigger upstream departures always make for less plausible historical counterfactuals. Or that smaller upstream departures generally make for greater plausibility. Consider:

(5) If time-traveling Homer Simpson had sneezed 66 million years ago, then everyone today would have forked tongues. 

(6) If no asteroid had hit the earth 66 million years ago, then humans would not exist today.

Whether someone sneezes is a much smaller change than whether an asteroid hits. Yet the counterfactual (6) with the much bigger upstream departure is vastly more plausible than (5). These examples, show, I think, that the assessment of historical counterfactuals has to be holistic and multidimensional. It won't work to focus on upstream departure alone. Much also depends, for example, on relevant background knowledge about the causal connections that might link the upstream with the downstream departures.


I shared some of these ideas with an audience at UCLA’s Cotsen Institute for Archaeology in November 2015. (This partly explains my use of a non-paleontological example here!)  I also want to give a special thanks to Ramzi Kaiss for many helpful conversations about contingency and counterfactuals last year. 


[1] But see Avi Tucker, Our Knowledge of the Past. Cambridge University Press, 2004, who tries to account for assessment of historical counterfactuals within a Bayesian framework.

[2] S.J. Gould, Wonderful Life: The Burgess Shale and the Nature of History. W.W. Norton, 1989.

[3]Tucker says that that “when historiographic counterfactual hypotheses are overstretched across many causal links, evidence is missing for determining them” (2004, p. 231). This notion of stretching across many causal links is ambiguous between historical depth and historical slack.

[4] Timothy Williamson, The Philosophy of Philosophy. Blackwell Publishing, 2007, p. 150.

[5] Williamson (2007, p. 153.)