1 of 17

Agile, Antifragile, Artificial-Intelligence-Enabled, Command And Control (A3IC2)

FLTLT Jacob Simpson (presenter), Dr Rudolph Oosthuizen, Dr Sondoss El Sawah, Dr Hussein Abbass

2 of 17

Scope

Paper Summary

AI and Fragility in C2

C2 Agility and Antifragility

A3IC2 Functional Model

Question time

A3IC2

3 of 17

Paper Summary

C2 system effectiveness is directly tied to its ability to enable the command of military force to meet the strategic objectives of the nation it is defending. If a C2 system is designed in a way that negatively impacts strategic objectives, then it is a poor performing C2 system.

AI has the potential to generate rare but catastrophic events with strategic consequences.

If automated decision-making has the potential for rare but strategically disastrous decisions, then it could result in a fragile C2 system.

Agile C2, as a concept for C2 system design, should be extended to include ‘antifragility’ as a system property to prevent fragility.

A3IC2 is thus a design principle that enables AI integration into C2 systems while minimising the potential for a ‘fragility trap’.

To begin, a summary of the entire paper is necessary to frame ones mind on the problem A3IC2 is trying to solve.

For each point:

Tactics, operations, and strategy is along a spectrum. Tactics can have strategic impacts, as operational level decisions can have tactical impacts. Decisions in C2 are not divorced from the strategic level.

AI is strategic priority at the moment. There are high expectations on what it is able to achieve.

AI has weaknesses as a tool for C2 that have not been discussed thoroughly enough.

AI decision-making, if fully automated, could result in strategic mistakes with catastrophic consequences. Even if a very low chance, these ‘Black Swan’ events are of such consequence, that it negates the performance of the AI system broadly.

A3IC2 aims to reduce this ‘fragilising’ quality of AI.

4 of 17

C2, Strategy and Fragility

Military strategy, at its core, attempts to deal with the complexity of war in order to survive and win.

A C2 system is a tool to manage complexity in order to effectively allocate forces to achieve military objectives. C2 measure of performance is inseparable from the strategic context in which the system is operating.

If an AI-enabled C2 system with automated decision-making is unable to comprehend strategic consequences, then it has the potential to cause strategic harm.

The ability for an agile AI-enabled C2 system to cause strategic harm is classified as a form of ‘fragility’ to the system.

Military strategy has used heuristics for centuries to pass on lessons of war. These abstract theories are used to demonstrate that war is hyper-complex, that it is incredibly unpredictable and factors are often out of ones control.

The concept of friction alone is an excellent example of Clausewitz communicating the complexity inherent in warfare. . Boyd and his strategic theory is also a method of managing complexity through effective adaption.

Strategy theory is directly tied to the development of C2 systems. These systems are designed to reduce friction and reduce complexity for commanders. Through reducing complexity, a commander is able to better use military force to meet strategic objectives.

A system that harms the commanders ability to use force to meet strategic objectives is antithetical to the purpose of the C2 system.

5 of 17

AI and the Potential for C2 Fragility

Is C2 performance improved holistically if it is prepared to automate decisions to deliver deadly force with an AI prediction of 99% confidence?

What if the 1% chance of failure results in an irreversible strategic mistake?

For C2, the ramifications of getting it wrong can be so extreme that it could result in its own destruction.

Fragility from automated decision-making is due to the ‘platonic fold’. Hidden variables could result Black Swan events of extreme consequence.

C2SIM is unlikely to solve the fragility problem.

Highly Simplified State Space with Topology Formed from ML Clustering Algorithm

6 of 17

C2 AI Fragility Spectrum

Hyper

Complex Environment

Strategic Decision-Making

Complex Environment

Large

Complicated Environment

High Risk of Fragility

Low Risk of Fragility

Unknown Model Fidelity

High Model Fidelity

Specific sub-system

C2 Tasks

Tactical Competitive

Engagement

Theatre Level Competitive

Engagement

Medium Risk of Fragility

Low Model Fidelity

Tactical Decision-Making

Sense-Making

Synthetic Data

Necessitated

Synthetic Data

not a Requirement

Synthetic Data

Supplementation

C2 AI Fragility

Spectrum

High Consequence

of Failure

Medium Consequence

of Failure

Low Consequence

of Failure

System type

modelled

Environment

AI Task

Synthetic Data

Requirement

Fidelity

Achievable

System Fragility

Risk

Consequence

From Failure

Shown is the AI fragility spectrum. On the left we have the C2 functions that are largely the domain of sense-making. These are specific C2 sub-functions that can be automated with little risk of fragility. The consequence of failure is low due to other sense-making functions likely making up for any mistakes made by just one algorithm. A human is also likely acting as a quality controller for any output as the decision-maker.

On the other side is the domain of automated strategic decision-making. Automating these C2 functions comes with a high risk of fragility due to synthetic data necessity, unknown model fidelity, and a high consequence of failure.

In the centre is the tactical competitive environment. Failure would likely result in the loss of life, and if the battle is crucial, it could cascade into strategic consequences.

7 of 17

Limits to Growth / Performance

8 of 17

What is Antifragility?

J. Johnson and A. V. Gheorghe, “Antifragilty analysis and measurementframework for systems of systems,”Engineering Management andSystems Engineering Department, vol. 4, pp. 159 – 168, 2013

Concave

Convex

An antifragile system has the following elements:

Convexity: potential losses are limited but potential gains are exponential

The ability to learn from shocks and harm. System requires memory and feedback.

Use of overcompensation

Optionality

Redundancy

Small-scale experimentation

9 of 17

System Types	Definition	Effect on performance from change?
Resilient	A system that absorbs the impacts of stress or shocks from the environment and reorganises itself after.	Maintains a minimum level of performance and returns to normal functioning over time.
Robust	A C2 system whereby no significant change is observed when subjected to a variety of shocks from the operational environment.	Systems behaviour shows a satisfactory response to seemingly extreme conditions and is insensitive to change.
Agile	The ability for a C2 system successfully effect, cope with, and/or exploit changes in circumstances.	Loses performance, but rapidly changes proactively or reactively to the environment and returns to an optimal functioning.
Adaptive	The ability for a C2 system to successful cope with changing circumstances through changing its own properties.	Maintains a minimum level of performance and returns to normal functioning over time.
Antifragile	A C2 system that thrives from continually exploiting, and purposefully seeking, feedback from the environment to improve itself from volatility	System behaviour shows a satisfactory response to seemingly extreme conditions and returns to a level of functioning beyond previous performance.

C2 System Types

C2 system definitions often overlap and can even be complimentary.

Resilience and robustness are very similar in definition and performance.

Agility and the ability to adapt are complimentary.

Antifragility distinguishes itself through an emphasis on feedback.

10 of 17

Fragility

Resilience/

Robustness

Antifragility

Agility

Adaptability

Immutability

Ability to enact change to survive

Ability to survive from volatility

Two C2 System Requirements

11 of 17

System Elements	Definition
Adaptability	The ability to change ones own system, organisation and/or structure to become better suited for the challenge.
Responsiveness	A Systems ability to respond, proactive or reactive, to a change in circumstances, be it a stress or opportunity.
Flexibility (Optionality)	Ability to adapt a response to change in more than one way to accomplish a task. System is convex in design, enabling positive exploitation from shocks, while minimising any downside (‘bar bell strategy).
Innovativeness (overcompensation)	Permits and entity to generate or develop a new tactic or way of accomplishing something. Risk taking and invention incorporates the antifragile qualities of overcompensation and small scale experimentation.
Memory/Feedback	Ability of the C2 system to collect, store and maintain experiences and lessons from the operational environment.
Resilience	Ability to withstand interruption/degradation and return to normal operational capacity. A system that absorbs the impacts of stress or shocks and reorganises itself after.
Versatility (Robustness)	An acceptable level of performance or effectiveness in accomplishing new or significantly altered task or mission. Is reliable to expected and unexpected inputs.

A3IC2 Properties

Combining agility with antifragility produces a complimentary list of system elements for a C2 system.

The result is an agile system that grows stronger from stress over time.

The wider and more extreme the variables an A3IC2 system encounters, the more it learns to respond to an even harsher variable in the future.

Emphasis is equally weighted between responsiveness and memory/feedback.

12 of 17

Reducing C2 Fragility with Anitfragility

Two methods to produce antifragile elements within an AI-enabled C2 system:

Function allocation of AI into a C2 system to minimise the risk from catastrophic failure but maximise gains to the system.

Innovativeness combined with ‘chaos generation’ uses experimentation to discover fragility in the system; this enables overcompensation.

Brehmer and the Dynamic OODA Loop, ties Boyd’s strategy with the functions of a C2 system. A3IC2 builds upon this model.

B. Brehmer, “The Dynamic OODA Loop: Amalgamating Boyd’s OODA Loop and the Cybernetic Approach to Command and Control,” Command and Control Research Program Press, 2005.

13 of 17

A3IC2: Function Allocation

AI sense-making has less potential for catastrophic strategic failure.

AI decision-making has high potential for catastrophic failure if the C2 function it inhibits can lead to strategic impacts.

Example: Difference between hypersonic missile decision-making if one is deciding to attack or defend.

Through function allocation that lowers strategic risk, an AI-enabled system can be Convex in design.

14 of 17

A3IC2: Chaos Generation

The Chaos Generation function purposely injects system stress into the C2 system in order to expose fragility.

It explores the circumstance space in order to understand as many areas of acceptable performance from as many generated circumstances as possible.

This could be machine generated or a human team.

Chaos generation combined with innovativeness is the key to enabling overcompensation.

15 of 17

A3IC2 Functional Model

16 of 17

Antifragility as a System of Systems

17 of 17

Q&A