Moderately Skeptical of "Risks of Mirror Biology"

Link post

Last week, a report was released on the potential risks of mirror biology—that is, biological systems with reversed chirality. I had previously looked into the topic, and was skeptical of the viability of this as a global catastrophic or existential risk. This report was tremendously helpful in filling in details of a scenario where, if mirror bacteria are developed, they could cause a global catastrophe. I will assume that readers here have or will read at least a summary, if not the original Science article or perhaps the summaries in the (300 page) report itself. And before I get into details, I will say that I’m not a biologist; I have a significant familiarity with relevant topics, but there is some chance I get technical details incorrect. I have asked a few people to look at this, but if any such mistakes exist and are substantive, they are my own, and I would welcome feedback.

Defining the Scenario

To set the stage a bit more, I will note that a common problem in discussing risks is that the specifics of what different people are saying differs, or there is no substantive disagree. In 2019, I noted that AI risk “alarmists” said they were worried that in a decade, there could be dangerous AGI, whereas AI risk “skeptics” said AI was likely to be a decade or more away—so the disagreement was about policy, not the risk.

In this case, I will qualify my response—I cannot capture all of the different possible risks, but will address what I think the main issue is. Specifically, the report is discussing chirally reversed, heterotrophic bacteria.

That means the report is very specific—as they make very clear in many places. The report is not concerned about mirrored viruses, which are very unlikely to be a threat, in part because they cannot use host cells which have the “wrong” chirality to replicate—and we are not even close to being able to make reversed viruses, which are far, far simpler. The report is also clear that there is little significant marginal risk from, nor any need to stop, research on mirror proteins and mirror RNA. It is certainly not discussing drugs—I started taking an isomer of amphetamine with changed chirality, Dexedrine, to treat my ADHD 30 years ago, but it’s been used for most of a century. So future drugs with changed chirality are no more concerning than any new potential drug.

The issue at hand is specific to a narrow type of bioengineering, one that is in its infancy even compared to bioengineering generally. And notably, most of the pioneers of these methods are co-authors, and are deeply concerned. (Unfortunately, concerns which were voiced that a ban would be too restrictive seem to have missed this point.)

For potential future bioengineered mirror bacteria, the report emphasizes that both deliberate misuse and danger by default are concerning. I will not discuss the former; I fully expect that bioengineering allows dangerous organisms independent of changed chirality. The difficulty of intentionally engineering a particularly and intentionally dangerous pathogen will only be increased if the bad actor wishes to engineer a particularly and intentionally dangerous mirror pathogen^[1]. For this reason, I don’t think mirror bacteria pose a significantly increased marginal bioweapons risk, even though attempts to do so would be deeply concerning, and could plausibly succeed.

We are therefore discussing the marginal risk of a scenario where “peaceful” research into potentially mirror-organisms creates danger—and it is deeply worrying. We have seen (incidentally, in a paper I co-authored with one of the co-authors of the new report,) that the rate of accidents at even secure labs is far too high to think that research known to potentially pose large-scale risks should be allowed, especially without a very clear direct benefit. So if chirally reversed bacteria plausibly pose such risks, as the report concludes, the expected research benefit comes nowhere close to justifying the risk.

Priors and General Skepticism

I started skeptical, but upon close review, my estimate of the substantive risk was majorly update to a high-single-digit or a very-low-double-digit-probability^[2] - far higher than my previous estimate of effectively zero risk.

This is not because I disagree specifically with the report. It is because the number of flagged uncertainties all weaken the update towards the potential danger, leaving more weight on the prior. And the reference class includes at prior scares like gray goo and nanotech concerns, and the hole in the ozone layer; it seems fairly likely that we will discover either than the risk is less concerning than initially expected, or find how to cheaply mitigate it. (I discuss one possible avenue below.)

And as a general point, we should expect that newly identified risks are mostly overestimated. I recall a presentation at the Global Priorities Institute in which Toby Ord made a simple mathematical point about new risks^[3]. That is, if there is a potential future risk about which we have limited understanding, we will have some probabilistic estimate. Given that most such risks do not occur, the base rate is low. And assuming that we are calibrated well, and update correctly based on future evidence, the very fact that the current estimate of the risk is low means that most of the time our estimate should decline over time. For example, if we currently assign a 10% probability that mirror bacteria present a threat, we expect there to be approximate a 90% chance that we will update to a lower probability over time^[4]. That is, speculative new risks will, a majority of the time, not end up turning into concerns.

In this case, the report provides a reason to update, but it’s certainly not a final word. I think it is still unlikely that by default chirally reversed bacteria would cause global catastrophic risk, but I expect this estimate to rise over time if no additional substantive concerns about the research are found, if uncertainties are resolved in ways that point to higher risks, and if greater consensus about the plausibility of the risk is achieved. But updates in the opposite direction are not only possible, but per the paragraph, are more likely than not.

Of course, we should not be prematurely confident new risks will not emerge; low probability does not equal impossibility, and it is usually worth finding out more when such risks are proposed. Obviously, some proportion of risks that are initially judged to be improbable will end up being realized—and we expect to update towards higher estimates over time in that case. But early mitigation is often safer and more inexpensive than waiting to determine with certainty whether the risk is real.

In the meantime, I agree with the report authors that the risk is plausible, and unless and until the risk is either resolved or rebutted, they are exactly right about the prescriptions; this risk is worth discussing and potentially stopping or significantly mitigating—both because the downsides are so large, and because the benefits of genetically engineering chirally reversed organisms seem very limited.

Can we just solve the problem?

As noted, mirror viruses aren’t a threat. This was an initial reason for skepticism of the risk of mirror biology, since the most worrying infectious diseases are viral. But as far as people are from creating synthetic viruses, we’re far further from making synthetic bacteria.

However, traditional bacteria are less concerning in large part because antibiotics work. And as the report reviews, a subset of antibiotics are achiral or racemic, so they would plausibly work against mirror organisms as-is. For the remainder, it seems very plausible that mirror-antibiotics of other types would be safe for humans, and it seems possible to test that safety without doing any work on the actually concerning mirror bacteria. (Safety in the presence of an actual infection is critical as well, but cannot be safely checked.) However, previous research into chirally reversed drugs has shown that it’s not necessarily safe—the tragic case of Thalidomide was due to the fact that the S-enantiomer caused birth defects, even though the R-enantiomer is safe.

However, the safety of mirror-antibiotics seems like a plausible path for effectively mitigating the risk; showing that safety seems like it could address many of the concerns without requiring the type of research the report explains is unwise. (As a counterpoint suggested by one early reader, we do not currently have capacity for an Apollo program for antibiotics, and one would be needed. But the capacity to massively increase production of classes of new molecules, or even extant drugs, has tremendous value more broadly, as we saw during COVID-19.)

Environmental Risks

One critical drawback of antibiotics is that bacteria would invade multiple organisms; most bacterial infectious can infect many mammals. We would need to protect all host species, not just humans—and once an infectious disease starts spreading widely, especially to animal hosts, stopping it is very, very difficult.

On the other hand, the ability for human immune systems to fight off mirror bacteria seems very highly correlated with the ability of other mammalian hosts to do so. That is, the risk of large scale disaster is very high conditional on it being a risk at all. For that reason, I think there is very little difference in the probability that the disease kills tens of thousands, and that it kills tens of millions—and any difference which does exist relies on containment of spread much more effective that the containment needed for COVID-19, where we failed spectacularly^[5].

Furthermore, while some bacterial pathogens only survive hours outside a host, others can last for weeks or months. So if spread began, we would not only need to isolate or kill wild animals, domestic pets, and livestock, but containment would need to continue for quite a long time.

The report notes a tension between the likely higher difficulty and likely lower speed of replication, and the rate at which these bacteria would be destroyed or fought off by an immune system, which is also plausibly lower. I notice that I find myself trying to argue that this would, on balance, more likely favor lower risk^[6], but I have far too little reliable intuition or understanding to be comfortable claiming anything other than agreeing with the authors that it’s very uncertain. This seems important, but I defer to their assessment, while noting that there are several seemingly uncorrelated uncertainties which all need to go the same direction for there to be high risk.

What do we do now?

We do not have a safe way to deal with the risk today, and given that we are unsure about safety of mirror-antibiotics, we may never have a good pathway to mitigate the risk. It seems worth checking this, and any other avenues which might be suggested to mitigate the substantive risk these organisms might pose.

One possible response is that we need to ban the work. This has been dismissed as impractical, but I think that’s premature on two fronts. First, as the original paper in Science says, the first step is to get feedback on the technical proposal and have a series of dialogs about the work. Until there is greater consensus, no formal restriction is being suggested; as the report says, the risk is a decade away with investment, and if and as consensus emerges about the risk, we should expect the investment on behalf of scientific research funders to slow or stop. Policies on research risks already address this somewhat, but greater awareness of the danger seems likely to put a substantial additional break on this work. Unless such work continues and accelerates, we have the time we should need to decide what to do.

This brings us to another key point; the primary reason that we should be concerned about mirror biology rather than the generic “someone could make something scary” is that it’s far simpler to design mirror organisms than it is to make an arbitrary future organism. That is because we don’t need complete understanding in order to design a new organism, we can just “copy” functioning organisms. Imagine the difference between designing and building an entirely new engine compared to copying one, with left and right sides reversed.

But even if it is easier, this doesn’t solve the fundamental problem and the difficulty of trying to assemble an entire bacteria from scratch. (We can assemble viruses because they bootstrap themselves from DNA, and we can produce DNA.) And a number of key advances will be required to make this viable for mirror bacteria—unlike engines, we can’t use the methods used for normal bioengineering, since they rely on parts that have non-mirror chirality. Biologists do not just need to build all of the machining required to make a novel bacteria—which is not currently possible—but also make an entire set of complementary tools to build the reverse bacteria.

If the specific advances required for this mirror tooling are shunned by scientists, if work in the area is seen as irresponsible, then we should expect far fewer researchers to pursue it. We can hope that key journals wouldn’t publish such work, that biological supply companies wouldn’t stock key materials needed, that mirror tooling will be a neglected area of work. There will be little incentive for academics to pursue it, and limited ability for others to do so as well.

Of course, this does not make this class of dangerous research impossible, just slower and more expensive. But a second reason I disagree with the claim that a ban is impractical is that even if scientists agree that this is dangerous and unwise, and should be banned, negotiating a new ban vai international treaty is probably unnecessary. That is because it can likely be subsumed under existing programs; there are already agreements in place about export controls of potentially dangerous biological substances, like the Australia group, and the Stockholm convention has rules about banning some organic substances that pose ongoing threats. These could be used to prevent the creation and transfer of key material needed for this work, if an when a consensus emerges. Similarly, the biological weapons convention bans work known to be dangerous already. If there is consensus among scientists that there is significant danger from mirror bacteria, any development would be de-facto biological weapon development, which is already covered. And despite the current dysfunction^[7], diplomatic breakthroughs and changes over the course of the coming decade, as various biological risks become more obvious, seem worth pursuing, rather than any sort of new narrow mechanism.

And this isn’t limited to mirror biology. As noted above, the key reason for the focus on mirror biology is the relative ease of building it. Compared to needing to fully understand a microorganism in order to design something novel, this would be an easier pathway to building something novel. But any truly novel organism could pose risks in similar ways—the safety questions raised by bioengineering of novel organisms, like the safety concerns which should exist given almost any new technological affordance, will continue to be critical to answer along the way to developing safe and useful technologies.

Conclusion

Clearly, risk assessments should pay attention to the risk of mirror biology. And given the immense multitude of uncertainties flagged in the paper, there’s a fairly significant chance that we will conclude the risk is, in fact, overblown—but until then, we can and should be cautious. As Roots of Progress noted, “Given that the threat is relatively distant, no immediate action is needed. We have time to discuss it thoroughly, among a wider set of participants. The article released today is meant to be the beginning of that wider conversation, not a call to urgent action.”

Again, this does not mean stopping or even questioning all work on anything with different chirality; drug development and long-lasting, nonimmunogenic therapies are avenues of research that does not require bioengineering of chirally reversed organisms, and conflating them is silly. But ignoring the problem would be just as bad—as we’ve seen in the past, and see today, it sometimes takes time to build consensus about risks and how to address them . This is one where we have enough time to figure out what to do, so we shouldn’t waste it.

^
Both are already subject to international bans under the biological weapons convention.
^
I will give a numeric estimate of 10%, while noting that as discussed, I expect this number to change greatly over time.
^
He did not recall this when I asked, but says he believes he has said it before, so I’ll give him credit.
^
The actual path of updates to our estimate will be more volatile and random, but if we expected a greater than 10% chance that we will update to a higher estimate, we should already have used that higher estimate!
^
One topic the report does not address is rapid detection; I am unsure how feasible this is, but also seems worth pursuing—environmental detection seems plausible, but as noted, detection is less helpful once spread begins that it is even for traditional infectious diseases, which it has limited value unless response protocols are in place. No such protocols exist for mirror-bacteria, but this is another potential question to address.
^
For example, it seems that other organisms which eat bacteria won’t necessarily be able to distinguish mirror bacteria, even if they cannot digest them, and it seems likely that an immune system which can deal with mirror molecules
^
As I sit here in Geneva writing this instead of attending the effectively cancelled 2024 BWC Meeting of State Parties due to what is frankly an embarrassingly low willingness by state parties to get their acts together.

Moderately Skeptical of “Risks of Mirror Biology”