Man versus Machine at Combat Tactical Level Decision Making

The human ability to collect information, make sense of a situation, optimise action, and learn while executing has been challenged recently in games, simulators, diagnoses, and real-time analysis. How may this development reflect to future tactical combat level decision-making? Is the machine going to win the man in combat?

"This requires not only substantial investment in resources but also an open-minded and exploratory approach, in contrast to the common but sometimes exaggerated perception of military organisations as conservative entities." Meir Finkel (Finkel 2023)

"Fifth-generation warfare shifts the focus from kinetic force in physical dimension to the impact information dimension, where narratives and perceptions take centre stage, enabled by emerging technologies such as artificial intelligence, automation, and robotics." Daniel Abbott (Abbott 2010)

The article reviews some recent achievements in artificial intelligence, sets the situation for the combat technical level functions, digs deeper into decision-making under stressful conditions and illustrates a possible vision for the future state. The aim is to shake the historically conservative concepts of land battle to consider future possibilities.

Artificial Intelligence Improvements in Decision Making

A view to the evolution of machine learning improvements in various strategic-tactical games and competitions in Table 1 shows that machines are catching up and dominating men in table, card and video games and creativity competitions. Furthermore, fast-learning general-purpose algorithms are beating dedicated algorithms in those same games. 

Table 1: A sample of improvements in Machine learning applications in gaming and creativity

Year	Confrontation	Improvement
1997	Chess: DeepMind against Garry Kasparov	It took IBM 11 years to build and use customised chips to execute parallel searches. DeepMind was able to evaluate 200 million positions per second.
2016	Go: AlphaGo against Lee Sedol	A neural network-based algorithm first learned from game data, then played against itself, and finally, improved based on made mistakes. AlphaGo was able to create an unseen move during the game.
2017	Chess: AlphaZero against Stockfish (2016 top chess engine)	General purpose reinforcement learning algorithm that learned Chess after playing 4 hrs against itself. AlphaZero was able to assess 80 000 positions per second.
	Shogi: AlphaZero against Elmo (2017 world champion Shogi engine)	The algorithm learned the game after playing 2 hrs by itself. AlphaZero was able to assess 40 000 positions per second on a board that has more options than Chess.
	Go: AlphaZero against AlphaGo Lee (advanced Go engine)	Deep neural network with tabula rasa reinforcement learning algorithm. The algorithm learned the game within three days while playing itself.
	Poker: Liberatus against four champion poker players	The algorithm used a game theoretic approach for reasoning in an imperfect information environment while playing simultaneously against four human players with the following abilities: ·        Managing the whole poker competition in advance ·        Solving each game during the contest ·        Self-improvement after each day of the three-week competition
2019	Dota 2: Open AI Five against a Team of 5 esport players	The algorithm used proximal policy optimisation. The algorithm used 800 petaflops/s to gain about 45 000 years of experience within ten months. The short-term average decision time was 80ms.
2020	AlphaFold2 doubled the score of human competitors in Critical Assessment of Structure Prediction.[1]	The algorithm predicted 3D structures based on complicated rules faster and more holistic than a human.
2022	AI model that uses tens of terabytes of Earth system data and can predict the next two weeks of weather tens of thousands of times faster and more accurately than contemporary forecasting methods.[2]	With enormous amounts of data, ML algorithms can create forecasts of very complex phenomena.

---

[1] https://www.technologyreview.com/2022/02/23/1045016/ai-deepmind-demis-hassabis-alphafold/

[2] https://www.technologyreview.com/2023/07/05/1075865/eric-schmidt-ai-will-transform-science

-----------------------------------------

In conclusion and, in theory, a machine combined with the above features could:

1. Starts from zero knowledge and trains within months to master given battle scenario's technical, tactical, and possibly operational level features for victory.
2. Anticipates adversary moves ahead, creates picture of potential scenarios, and predicts adversary manoeuvring in 3-D space better than humans.
3. Makes short-term decisions within 80 milliseconds and optimises decisions simultaneously at technical and tactical levels.
4. Identifies lessons from the events and gains 150 years of theoretical combat experience teaching itself overnight.

 

Technical Level of Ground Combat is a Complex Military Decision-Making Environment

Probability and chance are well-recognised (Clausewitz 1984) (Fuller 2012) (Oliviero 2021) factors of battle environment. Tactics-technical level combat capability is a sum of surprise, manoeuvre, mass, firepower, and tempo (to name some essential tenets) orchestrated in variety of combinations with Command and Control to disrupt the adversary's socio-technical military system and exhaust its fighting ability. (Friedman 2017) The tactical tenets are in transformation to address the foreseeable changes on the battlefield. First, let's review the most likely changes in land warfare and, second, see what they will require from tactical tenets.

RUSI Land Warfare Conference (RUSI 2023) promoted the following tendencies of change in land warfare, which will challenge the contemporary tactics:

1. Transparent battlefield

• Civilian and military LEO satellite-based sensors provide a continuous feed of information from above the battlefield. The data can be acquired from commercial sources and fused with algorithms trained to identify especially military action on the ground.
• Unattended ground sensors improve details and add reliability to real-time event pictures.
• Cover and concealment become harder since sensors can fuse detection data from different parts of spectrum.
• Adversary will know the location and movement of blue forces as quickly as the information flows in the blue battle management system.

2. The concentration of effects vs protection

• Standoff weapons, lethal autonomous weapon systems, and precision warheads make it challenging to survive with contemporary armour. Adding armour thickness slows tactical mobility.
• Concentrated armoured units create a lucrative target for conventional artillery, attack helicopters, or massing of anti-tank UASs.
• Platforms and actors need to become more expendable and distributed but able for coordinated manoeuvres and fires.

3. Sustainment

• Logistics enables the tempo of fighting and is essential for offensive operations. Supplying distributed units require new delivery methods.
• Movement and mass of material expose logistics for continuous, wide-spectrum surveillance, so protection and endurance of logistics become a challenge.

4. Situational awareness

• An increasing amount of data and information challenges sense-making as human cognition overburdens from large amounts of information, loses focus in the stimulus-rich environment, and makes a biased conclusion.
• The organisational culture may prevent the distribution of information (need-to-know vs need-to-share; air-gap security vs zero-trust security), so situational awareness does not meet the requirements of distributed tactics. (Mansoor and Murray 2019)

5. Boundless, urban battlespaces

• People reside primarily in urban environments, and military strategies aim to "capture the will of the people and their leaders, and thereby win the trial of strength." (Smith 2005)
• Participating actors in urban battlespace may include, for example, civilians, communal authorities, law and rescue institutes, local corporates, international corporates, non-governmental organisations, insurgents, commercial military companies, interest groups, militias, criminal organisations, adversary regular forces and adversary coalition units. (Waterman 2019)
• The urban environment is more complex as these actors do not carry clear signs for identification, their intentions may transfer from day to night, and they do not follow agreements on war crimes.

In conclusion, the following Table 2 reflects the above tendencies to classical tenets of tactics and illustrates the possible impact in battle techniques and tactics and, therefore, change of tactical sense- and decision making.

Table 2: How do visible tendencies of change in land warfare affect tactical tenets of ground combat?

Tenet / Tendency	Surprise	Manoeuvre	Massing of force	Firepower	Tempo
Transparency	Surprise in land domain may be gained through other domains and dimensions.	Swarming manoeuvre of smaller, less detectable platforms.	Concentration becomes lethal, but dispersion rules.	Target acquisition is more lethal if situational awareness is achieved.	The advantage is harder to gain in a transparent battlefield.
Effect	Systems effect creates surprise and disrupts force cohesion.	A large, moving, hot, and radiating platform is an easy target.	Calls for a mass of nimble, small, and mobile warheads	The 4IR produces software-defined effectors.	Dispersed effectors will increase friction and entropy.
Sustainment	N/A	Higher mobility and wider distribution obscures logistics.	Dispersed troops increase the logistical challenge.	Smart warheads require software maintenance.	Besides live supplies, the force needs technical maintenance.
Situational awareness	Digitalised C2 creates more cognitive bottlenecks.	Becomes a core enabler and vulnerability for the swarming of distributed effectors.	Becomes a core enabler and vulnerability.	Becomes a core enabler and vulnerability.	Becomes a core enabler and vulnerability .
Urbanisation	Provides concealment in the physical dimension.	Slows manoeuvre and promotes smaller, autonomous, and agile platforms.	Constraints massing of units, but prefers small, swarming effectors.	Favours defence but constraints offence.	Slows down units and increases their entropy.

Art of Military Sense- and Decision-making

A Concept for Sense- and Decision-making

The classical military decision-making framework defined by John Boyd is simplified as Observe, Orient, Decision, and Action (OODA) (Osinga 2007). Based on this framework, Figure 2 illustrates a concept for sense- and decision-making. In this context, sense-making consisting of observation and orientation, which interprets the equivocal data. (Mattila 2016) Furthermore, decision-making is searching and selecting alternatives optimising between projected results, capabilities, and constraints. (Mattila 2016) The concept has three different situational pictures: real-time events per domain, composed operational picture, and forecasted possible/intended situations, which are referred to existing information and, finally, shared and agreed upon at the socio-cognitive level.

Figure 2: Concept for Observe, Orient and Decide at the Military tactical level

The above Command and Control (C2) concept may be established with an emphasis on creative leadership or policy compliance. These emphases are founded in the culture from which armed forces are generated. For example, German culture from 1871 – 1945 promoted officers' autonomous and aggressive action on the battlefield. (Mansoor and Murray 2019) Conversely, after the forceful manipulation of Bolshevik government, Russian culture produced obedient younger officers and relied on experienced and resourceful commanders at the operational level. (Freedman 2022)

A Team of Military Officers in Decision-making

A successful military command should be a mixture of compliance with institutional management culture and creative operational art. (Kuronen 2015) German culture before WW II reflected the war as "an art, a free and creative activity founded on scientific principles." (Condell and Zabecki 2008) The US FM 5-0 requires adaptive leaders"…who do not think linearly, but  who instead seek to understand the complexity of problems before seeking to solve them…" (Cojocar 2011) On the other hand, NATO assesses military success with five measures of merit and only one of them, measures of performance (MoP), includes some personal leadership features. (CCRP 2002) The other four enforce doctrinal and process compliance. (NATO RTO 2002) The 1/5 ratio in expectations does not indicate innovative tactical decision-making from NATO officers.

At the tactical commitment level, all efforts should focus on gaining the initiative and, eventually, victory over the adversary (reduction of adversary combat power by more than 30%). (Oliviero 2021, 51) In reality, this is not necessarily evident for all officers: 

• Training enforces drills and tactical forms, so officers prefer to use familiar concepts to solve battlefield challenges in decision-making. 
• Viewpoints may be constrained by their basic training and arms. An infantry officer aims to gain ground, an armoured forces officer aims to gain distance, or an artillery officer assesses ranges, amount of ammunition and supplies to impose a particular effect. 
• The Red Force doctrine, officers are training against, remains linear, predictable, and unimaginative adversary. 
• Since live exercises are expensive, officers train their tactical decision-making in war games, which often neglect friction, fog, chaos, and cognitive stress present on the battlefield.

Studies (Henaker 2022) (Scott and Bruce 1995) (Loo 2000) have concluded that there are five different decision-making styles categorising individuals when making important decisions: Rational, Intuitive, Dependent, Avoidant and Spontaneous.

1. Rational seeks information systematically and prefers logical assessment. However, rational has challenges in creativity and implementation of decided intent.
2. Intuitive recognises details from the information flow and matches patterns that feel right. Intuitive relates positively to creativity and difficulty-solving. 
3. Dependent seeks social conformance from others before decision-making. The decision-making process may be distracted and in need of social support.
4. Avoidant tries to postpone decision-making because of their low self-esteem. Still, avoidant is compliant with policies, doctrines, and orders. Avoidant is not suitable for creativity and tends to have high stress levels.
5. Spontaneous tries to accomplish decision-making as soon as possible. Spontaneous does not like conflict situations but perform well in rash decision and high-risk situations.

Human vs Machine Decision-making in Future Battlefield

The section fuses the tenets of tactical combat with visible transformations and tries to reflect these new situations in human-centric and machine-centric decision-making as featured in previous sections. Table 3 illustrates the outcome of the fusion from the view of two champions:

• Human is assumed as an average decision-making officer with 3-4 years of military education and about five years of professional experience with, possibly, one year of experience gained in live tactical action. 
• Machine is assumed to be a high-performance computer running a combination of continuously learning algorithms, expert algorithms, and pre-trained algorithms with real-life or synthetic data. Digital connectivity is supposed to be at combat cloud level . 

Table 3: Human vs Machine decision-making in transforming tactical combat environment

Transforming tenets of tactical combat feature decision-making challenges	Human	Machine
Transparency increases information and requires more computing power to make sense of collected data. Tactics prefer smaller, profoundly dispersed, manoeuvrable effectors, which swarm for effect, and retreat quickly.	Available data and information may overburden the cognitive ability to comprehend the situation.	A machine can recognise images, find patterns from large data mass, and forecast complicated, interdependent behaviour.
Effect calls systems understanding for system-wide impact. Dispersed effectors are harder to control and coordinate. Software-defined precision requires better target acquisition and configuration.	The adversary must be understood as multi-dimensional actor-network (Inglis and Thorpe 2019). Dispersed effectors require coordination of larger volume of details.	A machine can map the COA spectrum, model complicated, interdependent systems, and optimise the action of small effectors.
Sustainment of distributed, cyber-physical platforms requires more flexible and expert maintenance.	Rising complexity of critical paths on availability or sustainment may overwhelm cognitive capacity under stress.	With a digital-twin model and scenario-based simulation, a machine creates an overall logistics picture and can optimise sustainment.
Awareness is achieved by delegating sense-making to lower cooperative level or improving the information management ability of a steeper, hierarchical command structure.	Socio-cultural structures and beliefs handicap the application of the optimum C2 method.	Socio-cultural structures do not constrain a machine, and it can act even with partial information environment.
Urbanisation increases entropy, slows the tactical pace, increases casualties, raises the need for sustainment, and makes the environment and situation harder to understand.	The urban environment increases entropy and requires more innovative decision-making.	A machine makes sense of complicated situation even with partial information, recognises faster volatile behaviour, and optimises effort and sustainment.

A Company Commander Meets an Ex-Machina Battle Captain

When a Human Commander meets an Ex-Machina Captain within a tactical scenario on a future battlefield, the parties of combat may have the different abilities for decision-making. In situation with equal forces, linear doctrines, and a reasonably stable battlefield, the company commander does not have a chance against Ex-Machina. A creative human commander may gain an advantage in more chaotic conditions and with innovative tactics. Are our military institutes educating agile officers? Still, higher man-machine teaming performance indications are positive in Dota 2 strategic game, but it remains to be studied in future articles.

 

Figure 3: Man vs machine in tactical decision making