Lieutenant General (ret) Gilles Desclaux, founder and president of consultancy company GDC2, executive adviser to Thales and ThalesRaytheon Systems, designated President of the French Civil and Military Aviation Forum (RACAM), defence adviser to the Director of the Engineering University for Cognitive Sciences of Bordeaux.

LtGen. Desclaux has commanded a fighter squadron, a fighter wing and a major air force base. In his last assignment, he was the commander in chief for air defence and air operations. In this position, he commanded the French air operation over Libya in 2011.

From 2004 to 2007, as a brigadier general, he served in NATO Joint Command Lisbon, where he fulfilled the role of Chief of Staff of the Joint Operational Headquarters. Prior to assuming this position, from 2002 to 2004, as an adviser to the prime minister, he was responsible for air policing, NATO and EU affairs and arms export control.
He retired from active military duty in 2014.

LtGen. (ret) Gilles Desclaux introduced himself as head of a joint research team from academia, the Engineering University for Cognitive Sciences in Bordeaux and ThalesRaytheon Systems. The team is conducting worldwide research into the use of cognitive technologies to improve the military decision cycle. One of the central topics of this research is the symbiosis between human and machine.

In the great complexity that increasingly characterizes modern warfare, events in several areas occur simultaneously and can quickly overwhelm the decision-making cycle of commanders and their staffs.

For such situations, the research team has redesigned one of the most important C2 processes, namely the commander’s critical information requirements (CCIRs) process, and used new technologies to make prompt and relevant decisions. CCIRs are identified during the planning process and fall into one of three categories:

• those necessary for the major decisions expected;
• those that make it possible to verify assumptions; and
• those that ensure the protection of own forces and centres of gravity.

The expectation is that by collecting and processing critical information so rapidly, the commander will be able to “read the thoughts of the enemy”. The defined process therefore makes it possible to constantly answer a list of questions:

• Status of my critical vulnerabilities?
• Status of my critical capabilities? Are they threatened?
• Are my assumptions still valid?
• What critical information is required to execute the evolving plan?
• What might cause key conditions to change?
• What may hurt most?
• What are the opportunities and challenges?
• How is the enemy conducting multi-domain operations?

LtGen. Desclaux presented an academic vision of the tool for automating this CCIR process and explained how artificial intelligence is incorporated into this tool. Each level of information employs one of three phases of AI applications:

• substitutive AI for the cues;
• collaborative IA for triggers; and
• hybrid AI for the emergence of CCIRs and associated decisions.

Furthermore, he described how the conceptual solution is based on the principles of Human-Autonomy Teaming (HAT) and in particular on its four aspects: context sharing, rules, transparency and cooperation agreements.

The tool (ANTICIPE) for automating the CCIR process consists of six components and the corresponding HMIs.

ANTICIPE collects data from all available sources in the operational headquarters (documents, chat, email, voice communication system, C4I notifications, open source) and converts these data into knowledge artifacts stored in the ANTICIPE database (commented documents and weighted graphics database) for further processing. This annotation process is a team-oriented process.

Mining takes place autonomously through semantic models, ontologies and the identification of cues and triggers. Currently, an Event Appearance Database is being built to which human curation can be applied if required. It refers to the CCIR space, the type of source and different weightings. The part of the Decision-Making Cognitive Assistant is at this stage simulated by an If-Then-Else model based on teaching and planning products. The research team is already working on decision support tools using war-gaming techniques. AI, machine learning and decision support are implemented step by step in each building block, taking into account that this phase relies on a robust architecture that allows a fast and relevant prediction of the occurrence of critical information.

LtGen. Desclaux mentioned User Experience and design research as another very important part. The academic environment in which his team works, research fellows and students are very helpful in this area.

He then presented screenshots showing what the tool proposes and how it is presented in, for example, the CAOC room. The main view shows three types of information:

• the state of the CCIRs according to their colour and the quantified evaluation of their degree of emergence;
• the ongoing actions in relation to each CCIR, summarizing the decision to prevent or mitigate the effects of these CCIRs; and
• a historical view of the active CCIRs evolving between the different thresholds.

For each active CCIR, a trend bar presents an assessment of the future development of the CCIR, based on the cue appearance rhythm, cue weighting and other factors.

A large number of sub-menus enable in-depth situational awareness.

Thus a semantic tree assigned to the CCIR or the source of the recognized information appears. Each document is evaluated according to the nature of its origin and other factors. The effects on the CCIR dashboard show a change of state of CCIRs that have reached a new threshold. According to the warning criteria, the system reaches the nominee autonomously: COM, DCOM, COS or CJOC Director. Sense making in this case is not enough, it requires deciding on the best option to solve the problem. An interface then synthetically represents the essential elements that make a human decision possible:

• Proposal of actions related to “triggers” in a synchronization matrix;
• Presentation of the state of risk resulting from the proposed measures;
• Displays the resource status of the proposed actions.

If one of the proposals for actions is deselected, the impact on resources and risks becomes immediately apparent.

The first idea is to allow the commander to deal with multiple dilemmas simultaneously while informing all staff of his intentions and decisions. Individual operators will be able to use this HMI module to implement decisions.

Additional specific features, such as multidimensional representations of the opponent’s actions, are under development. A multi-domain view makes it possible to evaluate the opponent’s strategy, especially between two datasets. The frequency of the appearance of the cues is a determining element of the opponent’s will. A flexible configuration of the graph (domain distribution) will enable the commander to visualize information according to the context and his needs.

Key take-away

ANTICIPE aims to enable the anticipation and reactivity necessary for multi-domain operations and to focus the staff on the commander’s strategic intentions while avoiding cognitive and organizational biases.

Download Powerpoint Presentation: AI in support of complex decision making (ANTICIPE) – LtGen. (Ret) Gilles Desclaux