Thoughts for March 13, 2022
Good afternoon. This week’s topics include the prospect of limited war, missile defense, a mea culpa, deep learning, and F-gases.
Limited War
A few days ago, Robert Zubrin published an editorial calling for NATO to impose a no-fly zone over Ukraine. That, together with whether NATO and the US should support Polish efforts to transfer MiGs to Ukraine, is a dominant political issue at the moment.
Having talked with a few people about this, I suspect that the no-fly zone is a bit of a red herring. The Russians are already struggling to achieve air superiority, and supplying Ukraine with Stingers and Patriot missile batteries may be sufficient to achieve these goals. Theodore Roosevelt’s dictum, “speak softly and carry a big stick,” applies here; I would rather see the goal achieved than a lot of bluster.
The underlying issue is how aggressively NATO should intervene, and whether we should contemplate going to war with Russia. Zubrin, one of the more hawkish voices on the subject and the author of the piece I cited last week on Aleksandr Dugin, thinks so, and he argues that a war is not the same thing as a full nuclear war that involves the exchange of strategic weapons. Zubrin’s historical analogy is the Korean War, where UN troops, led by the United States, went to war with Russia and China behind their North Korean proxies. This did not lead to nuclear war.
The situation is already risky, and further moves by NATO, such as a no-fly zone or getting behind the MiG transfer, will also be risky. But there is no risk-free option. The risk of not acting is that Putin conquers Ukraine and is further emboldened to attack the Baltic states, to use chemical or biological weapons, to use tactical nuclear weapons, or otherwise to take an action so intolerable that the United States will have no choice but to respond with war.
Like any decent person, I don’t want a war. But I recognize that this is a serious possibility and that we must prepare accordingly.
Missile Defense
A few more resources to extend discussion of a topic that I discussed last week.
As mentioned last week, the United States missile defense posture was most recently comprehensively reviewed in 2019 (see also, the Congressional Budget Office’s estimate of the cost of implementing the recommendations of the report). Our current trajectory was set with the 2002 withdrawal from the Anti-Ballistic Missile Treaty, with MDA’s defense aimed at what President Bush then labeled as the “axis of evil”: North Korea, Iran, and Iraq. While the Trump administration, particularly National Security Advisor John Bolton, started reorienting missile defense as aimed at Russia and China, our current defenses are wholly inadequate for that purpose.
In fact, it is not clear that they are even adequate for the North Korean threat, which was the primary concern expressed in the 2019 review. This report discusses the record of poor performance under controlled tests, and without the efforts to defeat the system that will doubtlessly be present in an actual attack. In response to this and other reports questioning the reliability of the system, see DoD’s report asserting that the system is sufficiently tested and reliable.
Missile defense is generally opposed by arms control advocates; see this and this editorials that make fairly typical arguments. The arms control argument against missile defense is as follows. Since the cost of shooting down an ICBM is much greater than the cost of launching it, the logical response to missile defense is to increase the number of warheads. The result is no net increase to safety, higher defense costs from the resulting arms race, and a heightened proliferation risk. It was this logic, and in particular the development of MIRVs (multiple independently targetable reentry vehicles) which allowed as many as 14 warheads on a single ballistic missile, that led to the 1972 ABM treaty in the first place.
That’s the logic at any rate. I don’t agree with it, and I would instead align with the logic in this editorial, in which the author Loren Thompson argues that the ground-based midcourse defense system is more reliable than asserted. See also this recent editorial in which Thompson lays out a more comprehensive nuclear response.
I try to approach the issue from a fresh perspective, without being biased from conclusions on other issues, but the anti-missile defense arguments have a kind of FUD feeling that will be immediately familiar to someone who is knowledgeable of the Union of Concerned Scientists’ work with civilian nuclear power. To someone who argues that missile defense is unreliable, I would counter, what is the reliable alternative? Deterrence has worked in avoiding nuclear war so far, but there were close calls such as in the Cuban Missile Crisis and a worryingly long list of accidental close calls. Arms control is not reliable either. As John Maurer outlines in this editorial, China does not seem to be interested in an arms control treaty right now, and any negotiation with Russia looks like a dead letter for the foreseeable future. (I highly recommend Maurer’s editorial as a good, accessible overview of several key aspects of the American nuclear posture.) Even if an agreement can be reached, there is a high risk of abrogation or cheating, as North Korea and Iran are suspected to have done. And as we saw with the September 11 attacks, some enemies will not be deterred by the threat of retaliation, nor is any negotiation with them possible.
The technology behind missile defense continues to advance, and the ultimate goal would be a satellite-based, direct energy defense using lasers or microwaves, as imagined in Reagan’s SDI vision. Directed energy is for missile defense what fusion power is for energy. Direct energy defense would not only be more reliable than SAM missiles, it would be much cheaper than the missiles they defense against. This would give such a decisive advantage to the defense over the attacker that ICBMs would become functionally obsolete, freeing the world from this terrible existential danger. But we are still far from developing a functional directed energy system, and there are all sorts of questions about how it would work, or if it could even work at all. Nevertheless, the appeal is so great that this is a vision worth working toward.
My view is that missile defense should be a high priority, and that we should also explicitly orient our defenses around protecting against Russia and China, not just rogue states. Yes, this will trigger an arms race. An arms race against Russia is easily winnable, as Russia has less than 10% (probably around 5% after the sanctions kick in) of the GDP of the United States and a dysfunctional, kleptocratic economy. An arms race against China is a more difficult and more long-term prospect, but we’re in it whether we choose to acknowledge it or not.
Mea Culpa
Last week I criticized several elements of the political system that continue to take a pro-Putin, pro-invasion stance, and so in fairness I should take ownership of things I have said in the past.
In particular, I have read and promoted several pieces over the years from Palladium Magazine. The last thing I read from them was an anti-Ukraine piece written on the eve of the invasion. Since then, most of the crew associated with that magazine have taken pro-Russian positions. There were many other red flags about the magazine’s crypto-fascist leanings of which I was not aware, but should have been. I will not be reading or promoting their material again.
I continue to take the view that the United States, and the Western world as a whole, suffers from a dangerous malaise that manifests as high debt levels, sub-replacement birth rates, economic stagnation, and regulatory sclerosis. These problems are obvious to anyone who takes an honest view of the situation. But there is a temptation to, in response, gravitate towards illiberal philosophies that promise to cut through the barriers. It is my view that any real solution must be consistent with the rule of law and respect for human rights and freedom. Otherwise, it is a false promise.
Deep Learning
Now for some completely different topics. A few days ago, Gary Marcus wrote an editorial entitled “Deep Learning Is Hitting a Wall”. There are quite a few interesting ideas here.
Marcus argues that a neural network based approach to machine learning is good for perceptual tasks where rough-and-ready results are sufficient. Tasks like image recognition and translation. But it doesn’t work very well for precise tasks or for tasks that require a high level understanding, such as MRI or CT diagnosis or autonomous driving. He also points out that behind the impressive GPT-3 demos from a couple years ago is a great deal of brittleness.
Marcus argues that the way forward for artificial intelligence is a revitalization of a symbolic approaches, which have been maligned under the deep learning revolution.
I made similar arguments a few years ago, and while my view (and I think Marcus’ as well) of deep learning was overly pessimistic, I would agree that symbolic methods are needed as well.
F-gases
Believe it or not, I have been doing a bit of actual work lately too, and my present task is to figure out how to mitigate SF6 (sulfur hexafluoride) emissions in electrical equipment.
In transformers, SF6 is commonly used as an insulating gas to prevent destructive electric arcs. But SF6 is a potent greenhouse gas, and its leakage from transformers is a problem. It should be noted that even with the status quo, SF6 leakage causes far less global warming than is saved by renewable energy deployment or electrification, so it is not really an argument against those things, but it is a problem that has to be dealt with. Overall, SF6 leakage from electrical equipment is less than 1% of total world greenhouse gas emissions.
This review from China finds argues that controlling leakage and end-of-life management of equipment is the best strategy for mitigation for now. Replacement of SF6 by low-global warming potential gases is where we ultimately want to get. This can be partially done now by mixing SF6 with nitrogen, but we’re not there yet on a full substitution.
Some papers that discuss alternatives include this one and this one and this one and this one and this one and this one. I have more work to do to understand these alternatives, but it doesn’t seem that there is much consensus on what the best SF6 replacement will look like, nor does it appear that there will be a widespread replacement soon.