Let me begin by expressing my thanks to the Hoover Institution, and in particular to Secretary George Shultz and Dr. Sidney Drell for hosting this workshop. I'm delighted to have a couple of my papers tossed in front of such a remarkable group of experts, and have been looking forward to our discussions of the "countervailing reconstitution" and "maximizing decision time" issues I raised in my work the "deterrence" conference here in November. I'd be very remiss if I did not also thank Carl Robichaud and the Carnegie Corporation, whose generous support has sustained me during my work on this subject for some time now.
I. Challenges of "Virtual" Nuclear Deterrence [morning session]
Can potential weapons provide enough deterrence that strategies based upon the purposeful maintenance merely of productive capacity could come to substitute in their entirety for strategies based upon "weapons-in-being"(i.e., actual, existing nuclear weapons)? When I began my work on this subject, I was more optimistic about the possibility of "countervailing reconstitution" (CR) as a strategy for sustaining nuclear weapons abolition than I am today. The more closely I have looked at this question, the more vexed I have become with the very serious conceptual and programmatic problems that CR raises.
I remain quite interested in CR as a strategy to facilitate nuclear reductions by making weapons possessors more comfortable with smaller arsenals of "weapons in being," and I want to stress that this is already happening in the United States and likely elsewhere. This aspect of reconstitution theory remains, I believe, very promising, and deserves more work. A post-abolition world of former possessors poised on the brink of reconstituting their arsenals, however, seems to me much more problematic.
I walk through these issues in detail in the first of my papers that you have before you, but let me flag a few of the more interesting challenges here. Some of my concerns are indeed "technical," though some are more conceptual or game-theoretical.
Verification. Let me say up front that I opted not to spend much time on issues of verification, which Sid Drell, Raymond Jeanloz, and Ed Ifft have addressed in their own papers for this workshop. I'll focus on other things here, except that I would like to say that I fear my colleagues here may be underplaying the vulnerability of an abolition regime to possible cheating scenarios undertaken with premeditation by a current nuclear weapons possessor.
Even if you could as Sid and Ray hypothesize vastly increase the fidelity of fissile materials accountability in order to preclude worrisome future "material unaccounted-for" (MUF) problems in monitoring fissile production, there isn't much anybody can do about past MUF, especially in nuclear weapons infrastructures that have been producing weapons material for decades. With thousands of weapons' worth of material that probably cannot be reliably accounted for today, how you could have much confidence that a former weapons state wasn't hiding material or weapons from disarmament monitors is anybody's guess. The old MUF is most likely just flat-out lost, of course having disappeared into the cracks and crevices of ageing and perhaps now-demolished facilities, or into the environment but how safe would it be simply to assume this? Because it seems to be understood that past production is to some non-trivial extent not fully accountable, a possessor of weapons-grade material stocks that entered a disarmament regime with ill intent might not have too much trouble gaming the system.
Crisis stability and reconstitution "racing." But on to my papers.Because they're not really "technical" issues, I won't spend too much time on my game-theoretical worries here but they're big ones. With respect to technical matters, however, I would be interested in the views of the experts here today on the implications of displacing nuclear weapons competition from traditional arms racing (i.e., the construction of new and/or "better" weapons) into competition in technologies and methodologies for maximizing players' fleetness of foot in a reconstitution "race." (A system in which nuclear weapons had been eliminated but key states retained the ability to reconstitute nuclear arsenals would put a huge premium upon the speed at which such reconstitution could take place.) I don't know quite what such a race would look like, but it would be an interesting technical challenge to assess the likely future nature and characteristics of competition focused upon reconstitutive speed.
Turning to more game-theoretical worries, such reconstitution races could be quite unstable. A power that had reason to suspect it was likely to reconstitute more slowly than a potential adversary would acquire incentives to start reconstituting sooner in any future crisis than would otherwise be the case. By the same token, any power "winning" a reconstitution race would have some incentive to use nuclear weapons in order to stop its adversary's reconstitution process before they bore fruit. There would also be powerful incentives for non-nuclear preemption against an adversary's CR-related targets. All this, of course, worries me.
And there's a further set of conceptual problems related to how well countervailing reconstitution (CR) capabilities would deter "breakout" from an abolition regime in the first place. Much of the discussion of weaponless deterrence seems to be predicated upon the implicit assumption that nuclear weapons only serve to deter other nuclear weapons: I will be deterred from building nuclear weapons by the prospect that you will build them if I do. The problem, however, is that some countries seem to have acquired or retained nuclear weapons for reasons that aren't just about deterring someone else's nuclear weapons and if this is the case, much of the rationale for CR-based deterrence at "zero" evaporates.
Indeed, there may be a basic incoherence built into the conventional CR argument from the outset. If you hypothesize that there is any need to deter "breakout" from abolition, you've already admitted that nuclear weapons have some utility other than in countering nuclear weapons. Yet to the extent that a country desires nuclear weapons in order to do something other than just deter someone else's nuclear weapons, the existence of a CR capability elsewhere will not deter "breakout." A country perceiving a serious enough conventional threat, for instance, might find nuclear weapons "breakout" to be more attractive than compliance with an abolition agreement, even were such "breakout" to spark follow-on reconstitution by other players around the world.
One should not condemn CR just because it cannot deter all "breakout" decisions, of course. Perhaps, on balance, it would still do more good than harm. Nevertheless, given the degree to which possessors have clearly, at various points, seemed to think nuclear weapons necessary vis-à-vis non-nuclear threats, it must at least be conceded that CR is a very incomplete recipe for deterring "breakout" from an abolition regime. It is an open question whether CR's contribution to deterrence at "zero" is significant enough to make up for its "race"-related crisis stability costs.
Indeed, in some cases, the move to a pure-CR system might make some states feel "breakout" more necessary. Almost by definition, CR offers no assurance against quick attacks, which might have achieved their objectives before a nuclear arsenal could constructed. For someone facing such threats, getting ahead of the curve with nuclear weapons might start to look appealing.
For similar reasons, precisely because the point of a CR-based regime is that it would preclude rapid nuclear weapons deployment or use, "virtual" deterrence has questionable utility as a substitute for the sort of "extended" nuclear deterrence the United States offers its closest allies today. If anything, CR-based abolition might make it harder for a CR-possessor to offer ("virtual") extended deterrence to its allies, because resuming nuclear weapons production in response to a local or regional crisis would likely have a dramatic global impact by causing multiple players around the world to return to nuclear weapons work on a crash basis. A country the "extended" deterrence protections of which were weakened by its ally's shift to "virtual" weapons might find itself more interested in weaponization than before.
Programmatic challenges. But that's enough for now about such game-theoretical challenges. Let's talk about the programmatics of reconstitution. If one assumes that a CR-based "zero" is a good idea, can one actually maintain such a capability, credibly, over time?
One major challenge, particularly in light of the "race"-related crisis-stability concerns outlined earlier, is to ensure the survivability of one's CR process. If CR is actually to deter anyone, its infrastructure must presumably be able to withstand attack and be able to function afterwards, all the way up to the point of actually pointing a live nuclear weapon at someone.
It may indeed be possible to ensure process survivability against both conventional and non-conventional threats, but it would surely not be anything like easy, or inexpensive. It is also worth pointing out that peacetime and post-Cold War incentives and constraints seem to have pushed many nuclear weapons possessors in what might well be the wrong direction for CR survivability. Britain and France, for instance, each concentrate their weapons infrastructure in a single facility, while the United States has spent two decades consolidating its sprawling weapons complex into fewer and fewer locations.
Essentially every plant is now a "bottleneck" or "showstopper" facility without which the system could not function and none of them are "hardened" or presently defended in any way that would make them meaningfully resistant to any sort of serious conventional or nuclear attack. One might imagine an entirely new CR infrastructure secreted into hardened, deeply buried, and intensely defended bunkers, but it is not enough for one's productive capacity to survive if it remains entombed or otherwise unable to get weapons onto delivery systems and, if needed, dispatch them to enemy targets. Building a credibly survivable system that could go all the way from rebuilding warheads to delivering them does not seem impossible, but doing so would require enormous and sustained investments, essentially amounting to a re-creation of one's entire infrastructure in an entirely new form.
Perhaps the greatest challenge over the long term, however, is whether one can actually preserve a credible reconstitution capability at all, even if it is not attacked. The literature on the U.S. nuclear laboratories' struggles to remain on the leading edge of their field in a no-testing environment is extensive, and many of you will know these issues much better than I. From the outside, however, it seems clear that even with substantial and ongoing investments in "stockpile stewardship," we are today struggling to maintain that edge and keep the "top-shelf" talent we need. When we have occasionally stopped doing something and later tried to re-learn it, moreover such as "pit" manufacturing for the W-88 warhead, or production of the exotic "Fogbank" material we have consistently been surprised by how hard it is to recreate past capabilities. Especially in light of these struggles just since the end of the Cold War, I have heard many experts at our national laboratories voice grave concerns about whether it would be possible to maintain a serious CR capability over many decades.
How acute is the danger of CR capability decay? One key question for reconstitution, of course, is what one would seek to reconstitute. One might perhaps slow such erosion by moving to simpler designs optimized for easy production or reassembly, conceivably from pre-produced parts in long-term storage. In no case, however, can one entirely avoid grappling with the challenge of potential capability decay.
Some disarmament advocates, of course, don't mind at all the prospect of such delay. I have even heard some argue that such gradual erosion is actually the great advantage of a CR system. They seem to imagine a sort of "stealth disarmament," by which today's possessors agree to "zero" on the assumption that productive capacity will be their strategic "hedge" against worsening security threats, but then one day wake up to find that their reconstitutive capabilities have quietly wasted away.
But I think gradual erosion could create troubling instabilities. If there is a need for productive capacity as a strategic hedge, there is danger in having it erode, especially stealthily. Nor, surely, would decay occur at the same rate or in the same way with all CR possessors. Rather than having everyone wake up to find that the problem of nuclear weaponry had miraculously disappeared, therefore, they might be more likely to discover that one of them had suddenly obtained a significant advantage over another in its ability to field nuclear weapons quickly, or at all. This returns us to the crisis-instability problems of reconstitution "racing."
All of this is a long-winded way of suggesting that despite my initial interest in the topic, I have doubts about the feasibility of "countervailing reconstitution" as a strategy for "zero." As noted, I remain enthusiastic about the use of productive capacity as a strategic "hedge" that can help make current possessors more comfortable with further reductions, and indeed I hope to do more work on this subject to develop this issue further. As for abolition based upon "virtual" nuclear deterrence, however, it is not clear that a pure-CR scheme would be a good idea, nor that one could indefinitely maintain such a capability even if it were.
II. Maximizing Nuclear Decision Time [afternoon session]
Bruce Blair has been a tireless advocate for "de-alerting" measures for many years, and is one of the most assiduous and prolific authors out there on issues of nuclear readiness posture. As I note in my second paper, I think he and Scott Sagan offer the strongest argument for reducing operational readiness of nuclear weapons to the point that they could not be launched upon warning of attack. But as you can also tell from my paper, I don't think their case is strong enough to carry the day.
My principal difficulty with the thesis is that nuclear risk reduction of the "de-alerting" variety tends to impose a zero-sum tradeoff between types of nuclear risk. In my paper, I distinguish between two types of risk: "Type A" risks revolving around incentives for deliberate nuclear use, and "Type B" risks relating to inadvertent or mistaken use.
Both types of risk are important, and my argument is that the best approach to risk reduction is to pursue remedies that address both types of concern and do not require stark trade-offs between them. I think it is the nature of a complex adaptive system such as a nuclear command-and-control architecture generally to require a dynamic balance, not taking any one consideration to an extreme at the expense of others. Because I think real positive-sum risk-reduction is possible, I do not favor "de-alerting"-style remedies that would reduce Type B risks by increasing Type A problems though, as you'll see in my paper, I do favor making changes to the current situation in order to reduce risks.
I won't spend much time here on the Type A critique of "de-alerting," because much of what I suggested earlier about the crisis-stability problems of "zero"-based CR also apply to a "de-alerted" posture in which nuclear weapons cannot be returned to readiness before an incoming attack has landed. Just as I worry about reconstitution "races," so do I worry about the stability of "re-alerting" races.
I think it's of paramount importance to maximize the decision-making time available to national leaders in a nuclear crisis. "De-alerting," however, actually constricts that time in a very important sense. The decision-making time I prize is not just how long one has after an attack warning before having to make a launch decision. I also prize the time in a nuclear crisis before one side feels compelled to do something that would alarm the other side and worsen the crisis.
Having nuclear forces that take some time to be made ready puts pressure on decision-makers to begin taking provocative steps earlier than would otherwise be the case: in a functional sense, they are left with less time than ever in which to try to defuse things. Because neither party will want to "lose" a re-alerting race, both will feel pressure to begin re-alerting at the drop of a hat. This seems likely to make crises more unstable, not less so. Making matters worse, the "winner" of a re-alerting race might even feel incentives to use nuclear weapons if he were confident that he could use his re-alerted force before the adversary returned to readiness.
This doesn't mean that I do not take the Blair/Sagan thesis very seriously, however, or think that "launch on warning" (LOW) is not dangerously destabilizing. People with whom I have spoken with in recent years in the U.S. nuclear weapons world think Bruce exaggerates the degree to which survivability issues have forced the U.S. and Russian nuclear systems into a sort of de facto LOW posture. Nevertheless, even if he is only partly right, there is cause for concern and it surely makes sense to do more to reduce LOW incentives if we can. I myself think nuclear stability is well served in the U.S.-Russian context by retaining a core of forces postured to be able to launch immediately, but I agree it would be madness to hang our collective survival on "launch-on-warning," whether official doctrine or not.
I make several suggestions in my paper that I think might help reduce Type B risks without notable Type A costs, but let me focus upon one in particular here. I think Bruce's analysis actually does help point us to the best way to reduce nuclear risks, but this remedy isn't "de-alerting." He spends a great deal of time in his various books arguing that even if nuclear forces can themselves be made effectively immune to a first strike as most people seem to agree is the case with deployed ballistic missile submarines, for instance the vulnerability of nuclear weapon command-and-control systems is still such that the United States and Russia would face "use-or-lose" problems in a nuclear crisis or when confronting what might be no more than a false alarm.
If Bruce's analysis of these LOW incentives is right, however, the most obvious remedy isn't to "de-alert" our forces, for this would displace the "use-or-lose" problem to the "end" of a re-alerting race rather than solving it, and could heighten the crisis-stability pressures facing race participants. Instead, the most obvious remedy is to cure the command-and-controlvulnerabilities he feels have encouraged a de facto LOW posture, and to continue to improve force survivability e.g., to increase the proportion of strategic forces that might be expected to survive and attack.
(In retrospect, by the way, I think my paper should have spent more time talking about force survivability instead of only command-and-control survivability. I don't want too quickly to dismiss Bruce's analysis of the "damage expectancy' pressures that he says give U.S. and Russian planners incentives to launch early because nearly all of their forces are needed to address mandated wartime target sets, thus making significant attrition unacceptable and "ride-out" thus unattractive. Both types of survivability are vital.)
By Bruce's own logic, ensuring a more credible option of nuclear attack "ride out" would go a long way toward reducing "use-or-lose" pressures, allow our de facto posture to more closely track our longstanding official position against launch-on-warning, and probably strengthen deterrence to boot. Unlike Bruce's preferred "de-alerting" remedies, this is a "win-win" proposition to Type A and Type B thinkers alike.
Thank you. I look forward to our discussions.