My recent conversation with Nassim Taleb was quite fun. (Full audio coming soon). Reading through Antifragile, though, I often felt like he was overcomplicating things. Taleb is rightly annoyed when people estimate “the worst thing that can happen” using “the worse thing that has happened.” But you don’t need non-normality or convexity or fat tails to expose this fallacy. Indeed, you don’t need any distributional assumptions. Why not just say this?
1. Either the worse thing ever has already happened or it hasn’t.
2. If the worst thing ever has already happened, then the badness of the worst thing ever equals the badness of the worst thing that has already happened.
3. If the worst thing ever hasn’t already happened, then the badness of the worst thing ever exceeds the badness of the worst thing that has already happened.
4. If we know with certainty that the worst thing ever has already happened, the expected badness of the worst thing ever equals the badness of the worst thing that has already happened.
5. If we don’t know with certainty that the worst thing ever has already happened, the expected badness of the worst thing ever exceeds the badness of the worst thing that has already happened.
6. We don’t know with certainty that the worst thing ever has already happened, so the expected badness of the worst thing ever exceeds the badness of the worst thing that has already happened.
Taleb is wise to warn us against using the worst thing that’s ever happened to estimate the worst thing that can happen. But his point is a lot more general than he seems to think. As long as you have enough common-sense to admit that the worst thing ever might not have happened yet, his warning holds.
READER COMMENTS
Peter Gerdes
May 3 2018 at 10:58am
Unfortunately that doesn’t quite work because when Taleb talks about estimating “the worst thing that can happen” he almost certainly means “the worst outcome that might occur in the future.” Your proof only tells us about the maximally bad thing that has or will ever happen.
To make this concrete imagine we are trying to estimate the worst thing that can happen to our economy. Suppose we are 90% confident that things will not only never be as bad as the great depression but never even be 1/9th as bad and we think there is a 10% chance they will be twice as bad. Then our expectation for the worst things will be in the future is less than 3/10’s as bad as the great depression.
Hazel Meade
May 3 2018 at 1:04pm
As long as we’re talking about Nassim Taleb, I think his mistake on the subject of GMOs is that ecology is *already* anti-fragile, especially against genetic alteration, due to billions of years of evolution to withstand, and indeed, make use of, random mutations in genetic code. What humans are doing with genetics isn’t anything that the ecology hasn’t had thrown at it a infinitude of times already.
Izuku Midoriya
May 3 2018 at 2:27pm
Please ask him, if he is all about risk taking, and actions over “tawk”,why can’t he just sue Monsanto?
Jimmy
May 3 2018 at 7:15pm
@Hazel
Selective breeding (incremental) vs Transgenics (radical). Fundamentally different risk profiles.
Mark Z
May 3 2018 at 7:37pm
Jimmy,
Organisms (not merely species, but individuals) accrue new mutations all the time completely at random. Why are we not afraid of whatever random point mutations are present in the chicken I had for lunch might be catastrophic? More over, the same process that eliminates harmful mutations in nature – death – also eliminates individuals with harmful transgenic mutations in a lab.
At best, imo, Taleb’s position on GMOs illustrates a poor understanding of biology; at worst, a poor understanding of risk.
Thomas
May 3 2018 at 10:36pm
You have taken the long way around to say something obvious, namely: However bad things have been in the past, they might be worse in the future — or they might not be. And there’s no way of knowing if the worst is behind us or in our future.
To which I say, “So what?” How does any of this help anyone decide what to do (or not do) about an uncertain future. Persons with high risk-aversion and low discount rates will want to spend a lot of money now in the hope of averting disaster. Persons with low risk-aversion and high discount rates will want to wait and see rather than spend money. The in-between cases will come out somewhere in between. And never the twain shall meet. Math won’t settle the issue.
Richard C
May 4 2018 at 4:00am
@Mark Z,
The difference is in the genetic distance of the transgenic genes, and the potential for new untested interactions between the existing genes. Evolution has tested the local genetic environment and made it robust to local mutations etc. over millions of years. Introducing complete genes from very distantly related organisms in a way that could never happen in nature is in contrast untested, and far more dangerous due to the potential for unpredictable interactions with other genes. There is the potential for devastating impacts on the wider ecosystem, and we only have one of those.
So no, it is you who has a poor understanding, perhaps this clears it up for you.
Alan Goldhammer
May 4 2018 at 8:17am
I’ve not read the Taleb book so I cannot comment directly on what he said on this topic. I do think Professor Caplan is making things too complicated. The “worst thing happen” scenario has two components: something that impacts only and individual (including family) or an event that impacts a much larger group. Examples of the first might be a bankruptcy or death(s) while the larger event might be something such as 9/11.
Caplan’s “badness spectrum” is always going to be one of individualistic perception. Regarding the “worst thing ever”, that’s pretty simple; it’s the catastrophic destruction of life as we know it. It hasn’t happened yet.
Peter Gerdes
May 4 2018 at 11:53am
@Richard C,
>Introducing complete genes from very distantly related organisms in a way that could never happen in nature is in contrast untested, and far more dangerous due to the potential for unpredictable interactions with other genes.
No, it’s just not true that there is no way in nature to have genes introduced from completely different sources. Indeed, substantial parts of most animal genomes include genetic material from retroviruses and indeed horizontal transfer of genes appears to happen between plant species (though the exact mechanism isn’t always clear). https://en.wikipedia.org/wiki/Horizontal_gene_transfer#Horizontal_transposon_transfer
More generally, it’s false to assume “Evolution has tested the local genetic environment and made it robust to local mutations etc.” The whole reason evolution happens is that it’s possible to have *novel* mutations or combinations of mutations which *by definition* haven’t been tested.
WHENEVER we modify plants either by genetic modification or by selective breeding we are introducing gene combinations that have never been seen in nature (we know because we are selecting for traits not seen in nature). It’s not at all clear there is any more reason to suspect that this will be problematic in the GMO case than the normal plant breeding case.
It’s not that there might not be some argument to this effect but this isn’t it.
Mark Z
May 4 2018 at 6:30pm
Richard C,
There is in fact extensive empirical research finding that introducing very distant genes is not more likely to produce “devastating impacts” on the wider ecosystem. “https://www.tandfonline.com/doi/abs/10.3109/07388551.2013.823595”
The vast majority of random mutations are either silent or harmful to the organism itself, meaning if anything the real ‘problem’ of introducing extremely different genes into a given organism is inducing to be non-viable. Science fiction writers fret about transfecting random genes leading to super-organism that threaten the global ecosystem. Scientists in labs fret about finding alterations to induce that don’t just cause the organism to die. I talked with one recently who was trying to render a a plant immune to a pathogen. They tried one alteration after another. None of them inadvertently portended the extinction of humanity; rather, they just killed the plant, stunted its growth, or otherwise inhibited the plant’s survival. This is what randomly introduced mutations actually tend to do.
Many years of empirical evidence has borne out the error of the assumption that artificially introduced mutations are inherently riskier or prone to be harmful to humans than those that occur in nature.
Hazel Meade
May 7 2018 at 12:14pm
Thanks Mark C and Peter Gerdes. It occurred to me that this isn’t even as off-topic as I thought, because when we are talking about the history of evolution, we are getting as close to having observed “the worst thing that can ever happen” as is conceivable.
The other thing I realized is that we have a problem in the sense that most literature on ecology describes ecology as a fragile system which can be easily collapsed by the extinction of a single species, or via some “butterfly effect” phenomenon. But I don’t believe that has ever actually been observed. The examples of ecosystem collapse in the fossil records are dependent on much bigger driving factors such as mass climate change, asteroid collisions, and so forth. The notion that extinction of a single plant or animal species is going to butterfly into a total system collapse seems to be incorrect. In fact, if that was going to happen, it would have happened already by now. Mass extinction events have happened but not as the result of cascading system failures, only as the result of planet-wide disruptive events such as astroid strikes and climate change.
Comments are closed.