Share this Post
Can We Recreate the Battle of Midway’s “Luck”?
In H. Ha
This past June marked the 72nd anniversary of the Battle of Midway, recognized as the turning point in the Pacific during World War II. June 4th, 1942 is dramatized in books aptly titled Miracle at Midway, Incredible Victory, and No Right to Win. Because at 1020, happenstance brought together three U.S. Navy dive-bomber squadrons, launched at disparate times from different decks, unopposed over four Japanese aircraft-carriers. This simultaneous aggregation at the critical point is always desired but was unexpected. In the next five minutes, these fifty dive-bombers would cripple three carriers. Through the remains of that day, each side would lose a carrier and the balance of power in the Pacific was irrevocably shifted.
Can this “luck” be recreated in order to engender future opportunities? If this “luck” can be programmed as an Artificial Intelligence algorithm, then unmanned aviation will truly enter the robotic age. Military forces based on unmanned autonomous systems will profoundly change how we fight and equip for war and defense of the United States and its partners. But technology alone will not suffice, it must be paired with yet unforeseen creative applications of that technology. For as Max Boot writes, “The way to gain a military advantage, therefore, is not necessarily to be the first to produce a new tool or weapon. Often it is to figure out better than anyone else how to utilize a widely available tool or weapon.”
Battle of Midway
As dawn approached on June 4th, the Japanese “Mobile Force,” centered around four carriers, had sailed south-easterly along a line-of-bearing 200 miles from Midway. At 0430, this mobile strike force commanded by Nagumo, launched 108 aircraft at Midway in preparation for amphibious assault. He did not expect U.S. carrier action. The prevailing thought was the Americans were in respite and would require the capture of Midway to lure the U.S. carriers out for a climatic at-sea Mahanian engagement.
Unbeknownst to the Japanese, American cryptanalysts on Hawaii had broken their codes and were cognizant of their major intentions. Nimitz’s centerpieces were carriers Enterprise and Hornet in Task Force 16 commanded by Spruance and Yorktown in TF17 commanded by Fletcher and in overall command. Each carrier group would operate independently but close enough for mutual support. Meanwhile, Midway hastily received additional Marines and air assets (including Army B-26 and B-17 planes) to bolster their defenses.
To guide American forces, Nimitz promulgated, “Hold Midway and inflict maximum damage on the enemy by strong attrition tactics,” but not to accept “decisive action as would be likely to incur heavy losses in our carriers and cruisers.” Direct guidance was provided in a letter of instruction, to be “governed by the principle of calculated risk,” which he defined as “the avoidance of exposure of your force to attack by superior enemy forces without good prospect of inflicting, as the result of such exposure, greater damage to the enemy.” Target priorities were assigned to carriers, battleships, transports, cruisers and auxiliaries in that order. Both TF’s were to rendezvous northeast of Midway with intent to ambush the enemy vectoring to Midway from the northwest.
Doctrine in an Algorithm?
Nimitz’s issuance of clear intent, guidance, and priorities from his headquarters in Hawaii is well known. These mission-command type orders are necessary for effective command-and-control of distant forces in sparse communications to achieve operational goals while not stifling tactical initiative. This has similarities to designing a distributed artificial intelligence system where the overall system end-state is a sum of a multi-agent behaviors as commented in Parunak’s abstract (comments in brackets):
“Agent architectures need to organize themselves and adapt dynamically to changing circumstances without top-down control from a system operator [This describes what military doctrine and training is designed to effect and what mission-command type orders ideally achieve]. Some researchers provide this capability with complex agents that emulate human intelligence and reason explicitly about their coordination, reintroducing many of the problems of complex system design and implementation that motivated increasing software localization in the first place. Naturally occurring systems of simple agents (such as populations of insects or other animals) suggest that this retreat is not necessary [Likewise, complex exquisite weapon systems requirements may not be necessary]. This paper summarizes several studies of such systems, and derives from them a set of general principles that artificial multi-agent systems can use to support overall system behavior significantly more complex than the behavior of the individuals agents [Likewise, a campaign is a series of individual battles designed to achieve operational goals].”
Hornet’s Flight to Nowhere
As Nagumo’s strike package was enroute, the Americans were trying to locate the unsuspected prey primarily using PBY Catalina’s, amphibious reconnaissance planes, based on Midway. At 0552, a PBY reported Nagumo’s location, “Two carriers and battleships bearing 320 distance 180 course 135 Speed 25” from Midway. Striking expeditiously was critical, so Spruance was ordered to “Proceed southwesterly and attack enemy carriers as soon as definitely located.” As for Fletcher, Yorktown (CV-3) soon had to recover their scouting dive-bombers. Spruance, meanwhile, closed the distance to the reported Japanese position 175 miles to his southwest and set 0700 as the launch time for TF-16.
However, there was no plan to coordinate between Enterprise (CV-6) and Hornet (CV-8) within TF-16. Unlike Japanese doctrine, American doctrine had each carrier air-wing operate individually, and each carrier had its own methods for ‘spotting’ planes on deck for launch. Unlike modern aircraft carriers with angled decks and ample flattops to conduct launch and recovery operations simultaneously, WWII carriers were limited to one operation due to one runway. Either spot on deck from below-deck hangers to the aft section of the flattop in preparation, launch, or recovery. Enterprise chose to spot their Combat Air Patrol (CAP) fighters and longer-ranged SBDs (dive-bombers) first on deck for group launch. Afterwards, escort-fighters and TBDs (torpedo-bombers) were elevatored up from the hanger deck for spot and launch to join the dive-bombers hovering above. The intent was a balanced attack package travelling as one air armada: SBDs at high-altitude, TBDs at low-altitude, and fighters protecting both from enemy CAP. After Enterprise’s first spot launched, problems delayed spotting the second group. During this delay, at 0740 an intercepted Japanese transmission revealing TF16’s position added urgency to resolve the issue as the element of surprise was waning. So, Spruance ordered the airborne Enterprise dive-bombers to proceed on mission independently without their fighter-escort or torpedo-bombers. As such, LCDR McClusky, CV-6’s air-group commander, led thirty-three dive-bombers of Scouting Six (VS-6) and Bombing Six (VB-6) to the southwest climbing to 19,000 ft. So at the outset, Enterprise’s strike was fragmented into two groups and further fragmented later when VF-6 could not find VT-6 whom they were supposed to escort.
Hornet completed launch at 0755 with some variance in spotting order, but inexplicably the air-group commander, LCDR Stanhope Ring, proceeding on a course nearly due west (265) rather than 240 toward the expected Japanese track toward Midway. Shortly after take-off, LCDR Waldron in direct violation to Ring’s protests, veered off to 240 with his entire VT squadron toward the enemy.
Yorktown’s launch was well executed and aloft by 0906 for what became the only coordinated strike package sent to target. Yorktown was a veteran of the recent Coral Sea Battle in the first carrier-vs-carrier battle but lessons learned were not yet disseminated to the fleet. Fletcher maintained his Scouting Squadron onboard as a reserve. Although launched cohesively, near enemy contact, this package also fragmented into two groups: TBDs with two escort-fighters, and SBDs with four escort-fighters.
The three U.S. carriers launched 151 aircraft at disparate times toward the enemy’s general direction with little coordination to arrive upon the enemy simultaneously. This air swarm decentralized into seven pulses, but “luck” would sequence their attacks and remass concurrently for critical effect. Can we encode this?
Simplicity in Distributed Artificial Intelligence
Sometimes the most complicated problems require the simplest solutions. Such was the case for “predator-prey” problem in the field of distributed artificial intelligence that vexed researchers for years. In a moose hunt, a single wolf cannot match a powerful moose, but a pack of wolves can surround the moose so that one can deliver a killing blow when the moose is distracted by the pack. Many proposed solutions assumed reasoning and communication capabilities not organic to wolves - akin to humans using radios for a coordinated attack. A simpler solution proposed by Korf in 1992 requires only rudimentary sensing and action on the part of both moose and wolves:
1. Moose: move to the neighboring cell that is farthest away from the nearest wolf.
2. Wolves: move to the neighboring cell with the highest score as determined by,
S = d(moose) – k * d(wolf)
Where d(moose) is the distance to the moose, d(wolf) is the distance to the nearest other wolf, and k is a tuning constant modeling a repulsive force between wolves. Parunak explains, “Each individual in the wolf-moose system both influences and is influenced by the entire system. Behavior of the overall system depends critically on the relative speeds of moose and wolves (since a fast moose can always escape a pack of slow wolves), and on the value of the parameter k that establishes the repulsion among wolves. When repulsion and attraction are suitably balanced, the wolves inevitably surround the moose, without any explicit communication or negotiation of strategies.”
Regarding Ring’s flight to nowhere, although nonsensical, if the separate “pulses” are considered multi-agents (wolves) in a distributed artificial intelligence system then they make sense. The repulsion factor (k) between these “wolves” dispersed them individually while collectively trying to surround Nagumo.
The American torpedo plane at the time was the TBD-1 Devastator that lugged a 2000 pound torpedo externally. It was the Navy’s first all-metal carrier monoplane, but by 1942 it was obsolete. The proscribed attack profile was to fly at 80kts, 100 ft altitude on steady course to within 1000 yards of a ship. This was a vulnerable position and doctrine was to attack with high-altitude SDBs to divide the efforts of enemy fighters present and with escort-fighters to protect both units in a combined strike package. TBD squadrons attacking alone would be suicidal. But that is exactly what happened to three torpedo-squadrons from Torpedo Squadron Eight (VT-8) from Hornet (CV-8) and VT-6 and VT-3 from Enterprise and Yorktown respectfully. Of the 51 torpedo-planes that attacked in cadenced fashion, only 7 returned back! The young Devastator pilots understood the odds facing them yet pressed on their attack runs while expectedly affecting no damage.
The common perception of these acts of valor, is that low-flying Devastators drew all the Japanese fighters to low-altitude in order to decimate the slow-flying torpedo-bombers. Subsequently, this left the three squadrons of high-altitude SDBs unnoticed and unhindered to conduct their near vertical killing blows. In the words of one Midway veteran, “You have to remember that the Japanese were subjected to no less than five separate torpedo attacks [B-26s and TBFs from Midway, then the three VT squadrons] over a period of only two hours or so, and the last three all came within the final hour. Despite their intense training and combat experience, the repeated attacks down low got those Zero pilots accustomed to the need for fighting at low altitude. By the time VT-6 and VT-3 showed up they had to expect it. They weren’t disappointed. VT-3 paid a terrible price for being the final bait, but they got it done. The Zeros were down low with them, not up there with McClusky and Leslie’s SBDs.”
Although true, Parshall presents an overlooked analysis that serial torpedo-squadron attacks prevented the Japanese from launching their counter-strike at the U.S. After Nagumo recovered his Midway strike force at 0912, he was sequentially attacked by three torpedo-squadrons preventing his flight decks to spot his own strike package upon the known American position: (approximate times) 0920 VT-8 attacks, 0940 VT-6 attacks, and 1010 VT-3 attacks. Either Nagumo was launching and recovering fighters to replenish his CAP, or radically maneuvering his carriers to evade the incoming enemy. Spotting on deck a Japanese strike force nominally takes 45 minutes to bring topside, warm-up, and launch - the Japanese carriers as a system had no respite. As suicidal as the American torpedo-bombers attack runs seemed, they disrupted the enemy’s will.
Some theorists equate Force to Capability multiplied by Will. Modern technology can enable autonomous drones where Will can be absolute. Low-cost drones are expendable and can be used to distract and disrupt the enemy, moreso if used sacrificially. Autonomy currently exists for constrained straight-forward tasks like landing on an aircraft-carrier or flying designated waypoints, but much more can be achieved.
A portfolio of UAV’s should align with two promising operating concepts, Air-Sea Battle (ASB) and Naval Integrated Fire Control-Counter Air (NIFC-CA). ASB’s operational approach to A2/AD challenges is a Networked, Integrated force capable of Attack-in-Depth to Disrupt, Destroy and Defeat adversary forces (NIA/D3). All of these terms are self-explanatory, save for Attack-in-Depth which an ASB Office representative describes as: “offensive and defensive fires and includes both kinetic and non-kinetic means to attack an adversary’s critical vulnerabilities without requiring systematic destruction of the enemy’s defenses. This is a significant departure from today’s rollback methodology that relies on uncontested communications and the ability to establish air superiority, or dominance in any other domain. The attack-in-depth methodology seeks to create and exploit corridors and windows of control that are temporal in nature and limited in geography.”
The military boogie-man known as A2/AD is nothing new to military history. One example is World War I after initial sweeping maneuvers devolved to static lines with intricate trench systems, interlocking machine gun fields of fire, barbed wire, and incessant long-range artillery. Such was the situation in March 1918 when the Germans began their spring offensive, but this time, new tactics were codified in a German manual The Attack in Position Warfare. A Bavarian Captain had consolidated the army’s new concept of rapid advance and disregard for security of the flanks. Attacks were led by specialized storm-troopers to cause envelopment and disruption. The German offensives from March through July 1918 saw the largest territorial gains since trench warfare ensued years prior.
NIFC-CA will provide long-range fire-control and projection, enable coordinated and cooperative situational awareness in a contested battlespace. By sharing a common-operating picture and target-solutions between platforms, an indirect fire-support system is created. Whereby, a sensor platform (forward-observer) can call-in long-range precision guided weapons from distant platforms (shooter) for massed coordinated fires.
Going back to the Wolves-Moose system, an algorithm is used to surround the moose. Likewise autonomous UAVs can be used to surround the target at effective missile range for massed fires - the wolves inevitably surround the moose, without any explicit communication or negotiation of strategies. This is what happened on 4 June at 1020, but by happenstance, when McClusky and Leslie’s three dive-bomber squadrons appeared overhead the enemy at the same time from different vectors.
Proposed Portfolio of UAV’s
Long-Range Strike Bomber (LRS-B): long-range precision-guided munitions could be ‘on call’ for fire support or assigned deep missions itself after skirmishers have disrupted or decoyed defenses.
UCAV: similar to the promise of X-47D, full broad-band stealth strike with long endurance to penetrate A2/AD defenses as a skirmisher. Its moderate internal weapons-loads can kill targets of opportunity or use NIFC-CA to call in strikes from LRS-B.
UCLASS: cheaper, low-observable drones with modular swap-in swap-out ISR payloads and armaments on external hard-points. An expendable disruptor.
QF-16: the “Boyd” drone. Boeing is working to convert F-16s into unmanned vehicles target drones. Hundreds of retired F-16s in ‘boneyards’ can be converted. Boyd helped design the highly-maneuverable short-range fighter and they can be used to get within the enemy’s Observe-Orient-Decide-Act loop construct. Preprogrammed moves based on sensed enemy disposition can be executed. UAVs can jinx through high-G force maneuvers that a human pilot could not physiologically sustain to place the enemy aircraft out of position ala “40-second” Boyd fame.
The Acquisition Strategy for this portfolio should start with the Artificial Intelligence (AI) piece as its own separate program. Individual platforms will have their own specific control systems but open architecture and open interfaces will allow for future Technical Insertions of AI for operational employment after platform development. In the near-term, development of UCAV and QF-16 should begin. The QF-16 is near full capability and under relaxed UAV export regulation, QF-16 can immediately bolster our partners through FMS or Section 1206 security funding programs countering China’s own UCAV program.
UCAV is developed early, because the lessons-learned can feed into LRS-B. Additionally, a combat-orientated unmanned platform, versus ISR-dominate, is needed early to glean integrated strike operation lessons similar to Yorktown learning carrier-based aviation best practices at the Battle of Coral Sea leading up to Midway. An ISR-dominate UAV marginalized to flying midnight ISR missions will not force the cultural and technical transformations necessary. There should be consideration for combined acquisition efforts with the U.K.’s own UCAV program, Taranis.
To this end, the planned six UAV per carrier detachment should be an Acquisition Category Level II-D program, with a multi-year contract developing six UCAV’s annually over five years. Two dets for deployed carriers, two dets for integrated manned-unmanned Carrier Qualification periods and ready in 30-days for major-combat operations surge, and one det for training and experimentation INCONUS.
In the mid-term, LRS-B will start development as the AI will be nearing completion and ready for Technical Insertion into UCAV airframes. Far-term is UCLASS development as our present inventory of hundreds of armed-capable drones near end-of-life. At this point-in-time, additive manufacturing advances may allow for the impromptu printing of thousands of cheap UAV frames to be assembled with higher-density power sources and ISR payloads.
Final Act of Valor
John Keegan writes that “Nature argued for flight, for cowardice, for self-interest; nature made for Cossacking, whereby a, man fought if he chose and not otherwise, and might turn to commerce on the battlefield if that suited his ends- this was ‘real war’ at its worst. The best-observed ideals of regimental culture, however – total obedience, single-minded courage, self-sacrifice, honour – most nearly approached that ‘true war’ which Clausewitz convinced himself a professional soldier should make his end.” As war more nearly approximated ‘true war’ the better it served a state’s political needs.
Indeed the crews of three Devastator squadrons emulated the virtues of true war as defined by Clausewitz. But ‘true war’ is unbearable. Whereas, the psyche of a human warrior cannot walk the fine line between real and true war due to their incompatible values, unmanned systems can maintain ‘true war’ as written in their code and mimic ‘real war’ in constrained situations when advantageous.
In the future, a young cyber-warrior charged by mission-command orders and informed by doctrine will generate the application for the robotic age - an “app” to implement operating concepts as the situation dictates. This tactical adaptability is warranted, because (in pugilistic terms) the resources applied in three balanced attacks in recurring ‘jab, right-cross’ combinations, may have more efficacy if restructured in ‘jab, jab, jab, right-hook, left-hook’ fashion. At Midway, this occurred serendipitously, although enabled by training, doggedness and bravery. The panacea will be AI - capturing valor in an algorithm - and the opportunities that such a genie will provide in a combined-arms approach to warfare. Such valor as demonstrated by three torpedo-squadrons and the 99 of 128 crewmembers that bore the ultimate sacrifice for their country.
 Good primer addressing the coming military-technical revolution, its opportunities and challenges. Robert O. Work and Shawn Brimley, 20YY Preparing for War in the Robotic Age (Center for a New America Security, Jan 2014).
 Max Boot, War Made New (NY: Gotham Books 2006) p.459.
 Robert J. Cressman et al., A Glorious Page in our History (Missoula, MT: Pictorial Histories Publishing 1990) p.39.
 H.V.D. Parunak, “Go to the Ant: Engineering Principles from Natural Multi-Agent Systems,” Annals of Operations Research, 75:69-101, 1997.
 Jon Parshall and Tony Tully, Shattered Sword: The Untold Story of the Battle of Midway (DC: Potomac Books 2007) p.134-135.
 Jon Parshall and Tony Tully, Shattered Sword: The Untold Story of the Battle of Midway (DC: Potomac Books 2007) p.172-173.
 To this day, the issue whether Ring departed on 265 or 240 is debated with hints of cover-ups. Alvin Kernan, The Unknown Battle of Midway (New Haven: Yale University 2005) p.128-136. Ronald W. Russell, No Right to Win (NY: iUniverse 2006) p.127-146.
 H.V.D. Parunak, “Go to the Ant: Engineering Principles from Natural Multi-Agent Systems,” Annals of Operations Research, 75:69-101, 1997.
 Ronald W. Russell, No Right to Win (NY: iUniverse Inc 2006) p. 195.
 CDR John Callaway, “The Operational Art of Air-Sea Battle,” Center for International Maritime Security, http://cimsec.org/operational-art-air-sea-battle/11913
 John Keegan, The First World War (NY: Alfred A. Knopf 1998) p.394-410.
 Matthew Bell, “Boeing touts operational QF-16 UAV,” HIS Jane’s Defense Weekly, 07 May 2014, http://www.janes.com/article/37617/boeing-touts-operational-qf-16-uav?from_rss=1.
 John Keegan, A History of Warfare (NY: Vintage Books 1993) p.16-22.
 LT Robert Bebber, “Developing a Strategic Cadre in the Information Dominance Corps”, Center for International Maritime Security, 29August 2014 (http://cimsec.org/developing-strategic-cadre-information-dominance-corps/12742).