Human-Machine Teaming Use Case

Human-Machine Teaming-Predict and be Predictable

Human-Machine Teaming (HMT) refers to situations in which people and intelligent agents, or robots, work together to achieve a common goal. The benefit of HMT is that each team member can bring a unique set of skills to the task, enhancing the team's efficacy and increasing its chances of success. For example, a future view of aviation anticipates the collaboration of pilots and air traffic control personnel with several AI agents (aircraft, air traffic control, weather, or maintenance agents) to handle the increasing demand for air travel, while maintaining safe and seamless aviation. 

A key requirement for enabling human-machine teams is predictability: team members need to be able to assess how other members would react in a given context. 

The Challenge of Building Human-Machine Teams 

The most challenging aspect of creating actual HMTs is, once again, modeling. For a team to work effectively, each team member, whether human or machine, requires a model of the other members to predict how they will act or respond in a particular situation. This anticipation is essential to plan own actions and ensure effective collaboration.


Anticipating other's behaviour is required for any type of teams. For example, in sports, throughout the training, each team member develops an understanding - a model - of how other team members interact in a given setting and, during the actual game, draws on this implicit model to choose how to engage with each accordingly. 

Developing human models to provide to an AI agent - and vice versa - is a subject of research that is constantly evolving. The Social AI Platform offers a somewhat different approach to enable team members to collaborate on a shared task. The first step is to develop an event-based model - the Reference Events Model (REM) - for the scenario to be supported by the team. The team members each rely on their IDT to align with the reference model and to ensure their actions are consistent with the scenario objectives. During execution, each team member's Information Digital Twin (IDT) customizes the REM to their users' specifics and task scope. Over time, each team member becomes increasingly reliant on their own customized version of the REM to anticipate and coordinate with other team members.

The HMT Platform

Human-Machine Teaming

A hypothetical Reference Events Model (REM) example of an HMT rescue mission scenario. The model manages a specific scenario event: minimize human team member casualties. The model's "input" represents numerous mission parameters linked to team equipment, configuration and mission context (e.g., weather). The model's "output" refers to the parameters reflecting the ongoing mission's progress and the achievement of the mission's overall objectives.

The inputs and outputs are defined using historical data on mission casualties.   For example, how many casualties occurred during the last 50 flood rescue missions? (line 35). Similarly, in the last 50 missions, most casualties occurred during missions that lasted 50% longer than intended (column C). The advantage of the REM is that it allows for the use of expert assumptions in the absence of data, which are then updated with each future mission. A probability is then assigned to each parameter to indicate its significance in predicting the occurrence under consideration: team casualties.

Numerous probability calculated may be intuitive and already known to mission managers. However, an actual casualty risk is largely determined by a mix of dozens of parameters, which is why the multiple IDTs/HDTs exist: to calculate the combined probability unique to each team member and thus provide a more robust, individualized, prediction signal in real-time. 

On the other hand, as not all parameters are accessible to all team member relying on a reference, shared model, the REM, ensures that the various parameters are up to data. For example, a drone can provide the actual location of the team members and update the corresponding parameter for each member upon which the HDT of that member can update their own risk accordingly.  

As the mission progresses, each team member's IDT/HDT evaluates its user's current input and output parameters, updates their specific risk score, alerts them when they approach a set critical limit,  provides recommendations to reduce the user's risks, and continuously updates the REM with actual user parameter values. 

A general mission casualty risk level can be determined at the mission control level based on signals and updates from various team members. Furthermore, mission control can change the significance of an entire category, such as giving the power supply levels category a lesser significance than the visibility level category. The change in relevance is reflected in all IDTs/HDTs risk score assessments and subsequent team actions, resulting in improved risk assessment and fewer casualties. .