Star Wars, AI and quantum computing

Basics of interceptor design by the Federation of American Scientists
Coyle report on National Missile Defense
Boost-phase issues cited by FAS
Garwin's case for 'boost-phase' vs. N. Korea
MIT's Postol shoots down midcourse interception
NMD technical challenges discussed by Center for Defense Information
Australian parliament report on technical and political issues
Quantum factoring with Shor's algorithm
Starwars feud behind Jason's shootdown?
American Physical Society report skeptical of boost-phase interception (search APS site)

---

The National Missile Defense proposal looks a bit iffy. Given missile-to-missile problems, are laser weapons really an inferior idea? Will advances in artificial intelligence spare NMD from failure? Can quantum computation play a role in NMD? Essayist Paul R. Conant first published ca. 2000 explores these questions.

---

Laser beams or missile-to-missile collision?

As proposed, the National Missile Defense has, at best, little bang for the buck. Every phase of interception, using direct collision as the kill mechanism, has serious drawbacks.

However, a system combining boost-phase interceptors with high-powered midcourse laser beams is not implausible, though the expense of an effective system may turn out to be unacceptable. Such a system has been found unpromising in a July 2003 report from a 12-member panel of the American Physical Society.

NMD is not designed to be effective against Russia, which can fire enough ICBMs to simply overwhelm the battery of interceptors. Though some have argued that the system might work against Chinese missiles, this seems unlikely, since the Chinese are well able to deploy effective decoys and other 'penetration aids.' The assumption that more primitive states will be unable to deploy such countermeasures is highly unconvincing.

A test intercept, rated successful by the Pentagon, was widely criticized for using a decoy heat source that would not be used in a real-world situation.

NMD is not capable of preventing delivery of biowar toxins, such as smallpox and anthrax, because the warhead, when entering space, can divide into numerous small bomblets invisible to interceptor detectors.

NMD's apparent purpose is to prevent fanatical elements of 'rogue' states -- Iraq, Iran and North Korea -- from successful nuclear warhead launches. North Korea has a missile that can reach Alaska and Hawaii and is presumed to be developing one that might reach the lower 48. It is known to have diverted nuclear materials for bomb research. Neither Iraq nor Iran has the capability to hit American cities, though an unstated purpose of development of NMD might be, as a corollary, to develop weaponry for defense of Israel, as part of a theater missile defense program. Despite George Bush Senior's enthusiastic statements about the success of the Patriots against Iraq's scuds, the kill rate was apparently very low.

NMD might also be justified as a means of countering terrorists who gain control of a missile or two. But, as the events of September 11 show, such a catastrophe needn't happen only in a distrusted nation; it might happen anywhere, raising very hard questions as to where to position interceptors.

The Air Force is reported to have overcome optical turbulence problems with its long-range (250-mile) laser heat ray, which, well clear of a foe's air defenses, would incinerate missiles, either during boost-phase or during free fall. However, the experts with the American Physical Society have noted continuing technical difficulties with an airborne laser.

There has been much debate as to whether spaceborne or airborne lasers are practical because land-based weapons are easier to defend. On the other hand, nothing beats the speed of light and missiles sent to kill a laser can be zapped by laser beam. The question then becomes, how many lasers are required?

An important advantage of space-based lasers is low maintenance. Another is that they can fire during a foe's boost phase as well as during midcourse free fall and even during the end course.

Also, the heat ray needn't incinerate every missile instantly. Rather, the missile would be heated above its air friction threshhold and would burn up like a meteorite. During boost phase, it might be advantageous to hit the missile in the lower atmosphere, though the APS panel said that a laser beam would be ineffective during boost phase against solid-fuel rockets, which are more heat resistant than liquid-fuel rockets. If the intruder is past boost phase, a laser intercept might be triggered sometime after atmosphere re-entry.

Multiple beams would hit decoys, which will likely be balloons, that will either explode harmlessly or overheat and burn up on re-entry.

A penetration aid that could render rocket kill vehicles useless would be a device to jam the KV's end-game detector, which is operated by an onboard computer. That is, the warhead carries a device that sends out a high-amplitude pulse of radio waves to create a power surge in the KV's onboard computer. The jammer might be set to go off some specific time after a surveillance signal (x-band radar) is detected.

The laser's computer will tend to be much farther away from the foe's warhead, and, of course, jamming ability diminishes by the inverse square of the distance.

Space-based lasers would be vulnerable to similar lasers fired from the surface or above it, though lower atmosphere optical turbulence can impede strength and accuracy of pulses.

Still, laser satellites can be cloaked with stealth technology; that is, a satellite can be shaped to promote the bending of surveillance waves around it and coated with a skin designed to absorb radar and perhaps other surveillance waves. In addition, small rocket-motor firings can be programmed to occur pseudorandomly in order to make minor orbit changes, thus increasing difficulty of detection.

Light and fast

The exo-atmospheric kill vehicle is designed to hit a warhead near the top of its trajectory in space. The advantage of an EKV (or of an atmospheric kill vehicle) is its light payload, which consists principally of a compact infrared (heat) sensor, coupled with an onboard computer. If an EKV's initial velocity equals that of its target, it will move much faster.

The EKV is guided to the vicinity of the target by ground-based x-band radar. The EKV's high-sensitivity heat sensor and onboard computer then orchestrate the 'end-game' intercept.

In space, the target is following a simple, largely unperturbed trajectory, so that course projection problems are less than they would be for a conventional aircraft or for a cruise missile taking evasive actions.

A salvo of EKV's would be launched at slightly different times in order to reduce likelihood of correlated errors (a miss by one KV won't imply a miss by all).

Initially, the Pentagon plans to deploy 20 interceptors, to be followed by 100 ground-based EKV's in Alaska. However, if North Korea's reach improves, it is reported that North Dakota is better positioned for an interceptor battery.

Infrared-sensing satellites would be used to detect launches. Though geosynchronous orbits are envisioned, the 22,000-mile altitude means a 2.5-second data delay. That couple of seconds adds to some minimum time between launch A and launch B.

At lower altitudes, more satellites are required for keeping inhabited areas under surveillance.

Boost-phase concerns

Boost-phase missile-to-missile interception -- knocking out the adversary missile while its booster is still firing -- is viewed by some experts as the next step in missile defense. The booster's rocket flame is easy for a KV to detect and the missile's course is highly predictable, improving likelihood of a hit. Boost-phase interception of ICBM's also avoids the decoy problem associated with midcourse interception.

The American Physical Society analysts argued that boost-phase interceptors are unlikely to be fast enough to catch missiles using solid fuel. If they are, they would be impractically large, the scientists said. They also said that, even against liquid-fuel missiles, boost-phase KV's would be unable to reach missiles fired from Iran, which is too far from submarine-launch or air-launch areas.

A major concern of boost-phase is protection of the weapon platforms, which tend to be vulnerable to adversary attacks. The KV's must be close enough for interception, meaning ships or aircraft must patrol within a specified region continuously.

Spaceborne lasers avoid this problem, though attack by satellite killer missiles, perhaps armed with jamming devices, is a concern. A foe with ground-based lasers might destroy any airborne or spaceborne weapons, but it is assumed that such systems are beyond the capability of 'rogue' nations. It would seem that if ground-based lasers were effective, the Pentagon would not be seeking to build a battery of EKV's.

It has been estimated that a ship-borne Aegis KV, loitering in the North Pacific, could knock out a North Korean missile headed for America in one shot, though weapons expert Richard L. Garwin favors stationing ships in the Japan basin or operating a joint U.S.-Russian missile base on Russian soil near the North Korean border.

Ships in the Persian Gulf would be positioned to fire interceptors at missiles boosting from Iraq or Iran, though distances to Iran's hinterlands could pose problems. Though such ships would be well-defended by batteries of short-range missiles, attack is still plausible. If Saddam rained enough missiles on the ships, they would run out of defensive rockets, thus leaving him free to fire at other targets.

Even if the United States placed boost-phase interceptors on submarines -- a formidable task that would require a new class of subs -- an adversary might 'smoke out' the subs by firing missiles in various directions, tracking the origins of KV launches, raining missiles down on the subs, and then firing a new volley at other targets. These scenarios are very unlikely because of the costs involved, but cannot be totally discounted.

An important use of boost-phase interceptors might be to deter an Indo-Pak nuclear war. India and Pakistan have missiles in the 1,500-mile range capable of carrying nuclear warheads. The political, ethnic and religious passions in the region could well be sufficient to spark such a cataclysm. As we know, a Pakistan nuclear scientist assisted the Taliban and al Queda in their nuclear weapon queries.

Because such a conflagration would imperil the security of every nation and pose radioactive fallout perils for the entire globe, it seems that America keeping a boost-phase anti-missile system primed might be worthwhile.

Midcourse hassles

The biggest hurdle to interception of an ICBM near the top of its trajectory in space is the likelihood of effective decoys.

A simple countermeasure is to have the warhead deploy reflective balloons in space and set them to rotating ('tumbling'). The sunless side of a balloon is far cooler than the warhead, which has been heated by air friction. But if the balloon rotates and its reflectivity is proportionately brighter than the warhead's, it will give off the same heat signature as the warhead, which is also rotating. In space, where there is no air resistance, the balloons will travel at the same velocity as the warhead.

Hence, the KV can't obtain sufficient data to distinguish among the objects.

As MIT's Ted Postol says, faster computers and better detectors are of no use here. 'Getting better information that is irrelevant doesn't help you.'

However, General Ronald Kadish, head of the Pentagon's missile defense program, thinks improvements in complexity theory might answer this obstacle. That issue is discussed below under 'artificial intelligence.'

As for atmospheric midcourse interception, it would appear to be a useful backup option in the event of boost-phase misses. It is necessary however that the KV not also be near the top of its potential arc, other wise relative velocity might be too low, resulting in a light tap, rather than a destructive hit. Apparently, Patriot intercepts in the Persian Gulf war suffered from low impact collisions, though some scuds may have been gently steered away from their original target. Patriot warheads, however, explosively fragmented near interception in order to increase probability of a hit. Hence, greater overall relative velocity was needed.

A not-foolproof countermeasure for short and medium range missiles is use of thrusts higher than anticipated at correspondingly changed launch angles. That is, say, if the Israelis are geared to aim near the top (height H) of scud trajectory A, Saddam might redesign his scuds to go to H+x, where x varies.

Because the warhead falls from a higher altitude, the increased acceleration and associated buffeting may increase course projection error, though this may not be a serious issue. Arms expert Andrew Sessler is persuasive in arguing that buffeting-related error is liable to be insignificant for endcourse ICBM's.

The Israelis, assuming they have been aiming for what they believe is the top of the scud trajectory, would then have to be prepared for a range of maximum altitudes. Their options:

*Store enough fuel in each interceptor for any plausible altitude, possibly complicating design and rendering a lot of costly equipment obsolete.

*Add interceptors, each of which varies in thrust, to the defensive system.

*Aim low, intercepting any comer at a bit below the lowest possible maximum altitude, and tolerate disadvantages, if any.

Three-phase interception not in cards

If a missile is missed in boost phase, try again in midcourse. If missed there, try again at endcourse, on the downward arc. That seems reasonable, but it's not in the cards.

The military is unenthusiastic about the endcourse option, perhaps because of costs associated with missile-to-missile systems. However, a laser system should be able to fire at enemy missiles during any phase of flight with little difference in cost.

Artificial intelligence and course-plotting

Gen. Kadish has argued that real-world tests, along with advances in complexity theory, would likely overcome the the midcourse decoy scenarios offered by Postol and others.

As this reporter understands it, complexity theory focuses on special cases of negative entropy: conditions that yield perceived order out of seemingly random inputs. Perhaps the general has been reading reports from Los Alamos National Laboratory, which has close collegial ties with Murray Gell-Mann's Santa Fe Instititute, a private think tank devoted to the study of complexity theory. Yet it seems unlikely that Gell-Mann and his associates have come up with revolutionary (and presumably classified) mathematical theorems that would allow the government such capability. Revolutionary theorems are hard to come by -- and they surface independently of government research grants.

It may be that Kadish was referring to advances in artificial intelligence, a research area no doubt entwined with government secrecy and military funding.

AI -- or quasi-AI (which is a less-controversial notion) -- is related to complexity theory because a top issue of complexity theory is the discovery of how life forms, including how the perceptive apparatus called the human mind, might evolve from self-replicating parts. So, if you can devise a program of self-contained automata to interact and build larger systems, you might be able to obtain a program that thinks, or quasi-thinks.

Development in World War II of radar-guided anti-aircraft fire contributed importantly to the development of computers (and inspired Norbert Weiner in his philosophy of cybernetics). Aircraft course-prediction systems use various weighting methods (more weight to more recent data, for example), filtration of signal noise and smoothing algorithms (methods of approximating an output curve closely through occasional input values). These calculations are judgmental in nature, the computer using various criteria to guess the next move (and perhaps countermove). We might think of the Deep Blue program, which defeated world chess champion Gary Kasparov, as a variant of an advanced radar detection program.

Quasi-AI, like human intelligence, is useful for anticipating a position based on imperfect data. If we consider 'random' to imply 'no datum on which to base a judgment,' then neither a human mind nor a computer program can make a useful prediction. Of course, we usually mean 'random within constraints,' in which case useful prediction is limited by the constraints.

We may view 'chaotic' or 'pseudorandom' to mean outputs that cannot be ascertained without knowing previous output values, as with a recursion function of the X-next type.

So if the next move is random within constraints or pseudorandom, we may face an exponential rise in computing work to either curtail the constraints or implement a recursion function.

In the case of the midcourse decoy scenarios, pattern recognition would have to go beyond course and brightness clues.

Pattern recognition can be, like the traveling salesman problem, a computational quagmire, though not necessarily so. If a pattern is composed of independent elements, computational work rises, essentially, by n. But if a pattern is a set of interdependent elements, then computational work can rise exponentially by number of elements. That is, a detector identifying two co-dependent elements has 2! (=2) units of work to do; a detector coping with 10 interdependent elements might have in the vicinity of 3.6 million (10!) units of work to do.

Pattern recognition of the decoy balloons tumbling through space looks to be an iffy task. Detection equipment would need to identify a set of small differences and analyze them, but possible clues are so negligible that it seems recognition of interdependent subsets would be required. (The total number of possible subsets of a set is 2^n, an exponential quantity.)

And suppose an 'advance in complexity theory' makes decoy recognition feasible? What countermeasure might yield another computational quagmire?

Bottom-up artificial neural networks have proved effective at 'learning' to discriminate among patterns. For example, in 'The Engine of Reason, the Seat of the Soul' (MIT Press, 1995), Paul M. Churchland cites programs in which a vector quantity is assigned to each element of a limited set of faces. The program then uses a process of error reduction until it is able to match a face with a numerical code (name) most or all the time. Such a system requires repeated trials (run by a serial computer).

The technology's appeal to weapons designers lies in the fact that the program is effective at identifying the correct pattern (target) even with degraded (noisy) data.

This technology, while fascinating, is unlikely to help much in the discrimination of Postol-type decoys from warheads. Neither human nor classical machine will do well at such a task because the differences in input data are so miniscule.

As of June 12, 2002, the Pentagon, taking advantage of the post-9/11 secrecy craze, had classified data on future tests of decoys, leading some to charge the Rumsfeld contingent of trying to shield NMD from legitimate technical criticism. Nevertheless, the Pentagon has never directly refuted the point made by Postol and others, but rather relied on the notion that technical breakthroughs will save the day. The possibility that NMD can be made to work only against inferior decoys but never against the best decoys designable is not addressed.

Does quantum computing lurk behind NMD

In December 2001, the New York Times reported that an IBM research team was about to announce the factoring of the number 15 using quantum computation, an experiment with defense repercussions. No confirmation of this initial report can be found in the usual places, such as science magazines or even at the relevant IBM web site. However, an IBM scientist referred this writer to a Nature article by the IBM researchers (Jan. 4, 2002), which discusses the potential, using lasers and beam splitters, for quantum computing using photon quantum effects. Nothing about the actual factorization of the number 15 is noted, though the IBM researcher does not deny the accuracy of the Times report.

The team reputedly used a quantum device to test and validate Shor's algorithm, which shows a way to use the quantum phenomenon of superposition to do simultaneous factoring. The team factored the number 15 into its primes of 3 and 5. Of course, this is a far cry from factoring hundred digit numbers but the validation of such a method sent shockwaves around the world.

If a classical supercomputer can crack a code by factoring numbers with 10^x digits, code-makers simply use primes that, when multiplied together, result in numbers with (10^x) + 1 digits. On a classical computer, the work of cracking such a code rises exponentially by digit place. It seems likely that decoys may present similar computational issues, particularly if the course-plotting program is already souped-up with nonlinear methods.

Is there a way out?

There remains the intriguing concept of quantum computation, which still appears an elusive quarry, despite its apparent validation in principle.

Essentially, the idea behind such a device is that though a quantum particle, once observed, appears to have taken a specific path, we can NEVER predict which path it will take with absolute certainty. For example, if a photon goes through a symmetrical interferometer, it is said to take one of, say, two paths each with a probability of perhaps 1/2, but, if left undetected enroute, the photon emerges as if it takes one path with probability 1. Some sort of interference causes the photon to have a nonprobabilistic final path.

Can this quantum weirdness be harnessed? In principle, yes, as the experimenters proved.

According to the 'many worlds' theory proposed by Hugh Everett (who, incidentally, spent his career as a Pentagon scientist), for each path the photon MIGHT take in our world, it actually DOES take in 'another world' (or 'parallel universe'). Our universe differentiates into two universes at the time a quantum particle, such as a photon, seems to make a 'random' choice. It is possible for differentiated universes to merge back into a single wave under the right conditions. If somehow these split universes could be merged back into 'our' universe, you might be able to compute classically exponentially hard problems in the blink of an eye.

Supposing the 'many worlds' idea holds, just imagine that you could somehow get nearly identical computers in a large set of universes to parallel-process a tough problem -- one universe/computer per route in the traveling salesman problem, for example.

A more usual view is that the photon, before detection, has various possible positions superposed. On detection, the wavelike nature collapses, and the photon's position can be known.

In the quantum computation experiment, factors are associated with superposed quantum states called spins.

At any rate, though quantum computation might someday prove a boon to code-crackers and AI program designers, it seems at this point unlikely that the code-busting National Security Agency would be very happy at the prospect of the Pentagon squandering such an asset on an easy-to-spy-on weapons system.

For deep insights into the worlds of quantum theory and AI, see The Emperor's New Clothes and Shadows of the Mind by Roger Penrose and The Fabric of Reality by David Deutsch. Also see Deutsch's Frontiers article.