Shared Mental Models as a Way of Managing Transparency in

Complex Human-Autonomy Teaming

Gabriele Scali and Robert D. Macredie

Computer Science Department, Brunel University, Kingston Lane, London, U.K.

Keywords: Human-Autonomy Teaming, Human-Agent Collaboration, Agent Transparency, Shared Mental Models,

Dynamic Environments, Time Pressure, Environment Complexity.

Abstract: This paper argues that because of the cognitive and communication limitations of human and autonomous

agents engaged in Human-Autonomy Teaming within dynamic environments, various external factors,

which can be classified collectively as environment complexity, set boundaries to the effectiveness of

strategies for agent transparency – that is, the ability of autonomous agents to make human actors aware of

their goals, actions, reasoning, and expectations of future states. Understanding the mechanisms by which

changes in environment complexity affect transparency, and the conditions in which it can be disrupted, can

help researchers to better frame the results of existing and future studies on transparency and, in turn, inform

the development of strategies to modify autonomous agents’ behaviour to maintain transparency under

different environment conditions. It is proposed that one such strategy could be the adjustment of the level

of abstraction of the shared mental model adopted by the team as the common ground for communication so

as to keep the amount of information that is exchanged manageable within human cognitive limitations.

1 INTRODUCTION

Improvements in capabilities of Artificial

Intelligence (AI) create opportunities for

autonomous agents to be deployed in increasingly

diverse real-world work environments as partners in

mixed human-agent teams (Sycara, 2002). These

situations are the subject of interdisciplinary

research into Human-Autonomy Teaming (HAT),

which investigates the challenges related to human

collaboration with autonomous agents towards the

achievement of common objectives (Christoffersen

and Woods, 2002; Hoc, 2000; How, 2016; Shively et

al., 2018).

In order to be said to participate in a team as a

true partner, an agent must be autonomous, which

means it being able to generate its own goals and

free to act on them (Luck and D’Inverno, 1995). To

be autonomous, agents must be capable of surviving

in their environment (be viable), they must not need

help in performing their tasks (be self-sufficient),

and they must set their own goals and make their

own plans (be self-directed). The above

characterisation can only be meaningful when

referred to a specific context of activity (Bradshaw

et al., 2013; Kaber, 2018). Throughout this paper,

we refer to that context as the ‘HAT environment’,

‘operational environment’ or just ‘environment’.

Agents involved in HAT are not bound by

dependence relationships, as it is the case in

supervisory control situations (Sheridan, 2012).

Therefore, to be effective team mates, they must

behave collaboratively (Bellamy, 2017; Klein et al.,

2004). A key aspect of doing so is to remain

transparent. This paper adopts the account of

transparency proposed by the Situation Awareness-

Based Agent Transparency (SAT) framework (Chen

and Barnes, 2014), which defines transparency as

the ability of an agent to make another aware of their

goals, actions, reasoning, and expectations of future

states. In order to do so, agents have to: select the

information they intend to communicate; choose an

appropriate time to communicate it; choose an

appropriate channel to communicate it; decide when

it is appropriate to repeat it; and decide when to

communicate updates and confirmations.

Transparency is therefore to be understood as a

quality of these actions and decisions, hingeing on

humans and autonomous agents being able to share

an understanding of the situation and of the

mechanisms and rules governing it.

Existing research on transparency in HAT has

focused on defining it as a construct and on

Scali, G. and Macredie, R.

Shared Mental Models as a Way of Managing Transparency in Complex Human-Autonomy Teaming.

DOI: 10.5220/0007484801530159

In Proceedings of the 14th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2019), pages 153-159

ISBN: 978-989-758-354-4

153

manipulating its level to study its effect on team

performance (Chen et al., 2017; Stowers et al., 2017;

Wohleber et al., 2017; Wright et al., 2016). Research

is lacking, however, into the factors and mechanisms

affecting the achievement of transparency itself, and

consequently how to maintain it.

This paper proposes firstly that environment

complexity affects agent transparency, as the

demands it creates test the limits of agents’ cognitive

and communication abilities. This determines

boundaries within which certain strategies for

achieving transparency work effectively. When

complexity exceeds these boundaries, it may be

necessary for the agent to make adjustments in order

to maintain transparency. Our second proposition is

that one such adjustment can be made in regards to

the level of abstraction of the Shared Mental Model

(SMM) on which communication and understanding

of the situation is grounded.

As an example let us consider a scenario in

which a human pilot is teamed with a synthetic agent

navigator, with the joint goal of visiting a certain

number of waypoints that become known to the

navigator over the length of the mission. The task of

the pilot is to drive a vehicle and to negotiate the

uncertainties of the terrain. The task of the navigator

is to interpret incoming information and to relay it to

the pilot so as to direct them to visit the waypoints in

the most efficient way. With a low number of

waypoints and a slow rate of arrival of new ones, the

navigator communicates the exact position of each

waypoint and the order in which it intends to visit

them. The shared mental model is one of points on a

map. If the number of locations and the rate of their

arrival increase, at a certain point it would become

difficult to maintain an effective communication

between navigator and pilot: transparency would

break down (our first hypothesis). The navigator

agent may then choose to switch to a more abstract

mental model, based on the density of locations to

visit on the map, along with the specific position of

only the waypoint to reach next. As long as both

mental models have been practised in advance, and

are equally familiar to the pilot, this may allow the

team to re-establish transparency in the changed,

more challenging, conditions (our second

hypothesis).

The remainder of this paper will discuss

complexity factors of dynamic environments and

how they can hinder transparency, briefly introduce

SMMs and their role in maintaining transparency

and conclude by proposing a possible approach to

mitigating this effect by adjusting the level of

abstraction of the SMM.

2 COMPLEXITY IN DYNAMIC

ENVIRONMENTS

An increasing number of useful applications of HAT

are possible in dynamic environments (Bainbridge,

1997; Hoc, 1993; Russell and Norvig, 2009),

characterised by the possibility for system changes

to occur independently of an agent’s actions, owing

to spontaneously-occurring events or to actions by

agents outside the team. The uncertainty about

future states and action outcomes, together with the

inherent variability of context, makes applications in

dynamic environments the most challenging for

HAT (Kaber, 2018).

For example, in a chemical processing plant,

machines can break down or availability of certain

resources may vary due to provisioning fluctuations;

in a Command and Control (C2) application, an

adversary may try to impede the operations, or

visibility may change; and in an Unmanned Vehicle

(UxV) scenario, interfering traffic, shifting weather

conditions and mechanical problems may all occur

independently of the vehicle’s actions, and affect

their precise outcome as well as the choice of best

course of action.

Environment complexity can vary between or

within its instantiations. For example, an Air Traffic

Control (ATC) system may go through periods of

low and high traffic (number of entities), as well as

situations when flights are on schedule and there is

no need to hurry, and others when there is a need to

recuperate delays (time pressure). This determines

that an autonomous agent operating in that

environment will be faced with maintaining

transparency under possibly very different

conditions.

While environment complexity factors,

characterised sometimes as task features and

constraints of the operational environment, have

been found in previous research to impact

interaction with automation (Mosier et al., 2013),

their effect on HAT transparency has not been

examined. Several frameworks have, though, been

proposed to categorise factors contributing to

complexity of environments and of tasks performed

within them (Ham et al., 2011; Liu and Li, 2011), an

exhaustive review of which is beyond the scope of

the present work. For this research, we focus on

three of the most commonly-cited complexity

factors and discuss how they can affect agent

transparency when present in a HAT environment.

In particular we consider: time pressure (Edland and

Svenson, 1993; Liu et al., 2016); predictability

HUCAPP 2019 - 3rd International Conference on Human Computer Interaction Theory and Applications

154

(Mosier et al., 2013); and number of entities and

possible courses of action (Park and Jung, 2007).

3 HOW TIME PRESSURE,

PREDICTABILITY AND

NUMBER OF ENTITIES

AFFECT TRANSPARENCY

That agents can be autonomous does not mean that

they do not differ significantly from humans in

attitudinal capabilities. The observation that humans

generally have better soft skills and adaptability,

while agents are able to account for, and process,

more information quicker and are somewhat limited

in their capacity of action in the physical world, can

be traced back to the classic Fitt’s list (Fitts et al.,

1951). Although the list may require some

adjustments owing to technological advancements

since its inception, the basic observation remains

valid that, for the moment, machines and humans

have largely differing abilities. In particular, the

ability of AI-based systems to process much more

information than the human mind, along with their

computational advantage, is likely to determine a

divergence of intelligibility between humans and

agents. As environment complexity increases it is

not possible to expect that humans can be made

aware of everything an agent perceives, does and

reasons (Miller, 2014).

3.1 Time Pressure

One of the commonly-cited contributing factors of

complexity is time pressure, which decreases the

time available to provide and understand

explanations. Examples of highly time-pressured

scenarios include search and rescue, operating

rooms, command and control, sport competitions,

and many others.

In settings where time pressure is not a driving

issue, agents have the option to slowly relay all of

the necessary information, provide detailed

explanations, suggest possible courses of action, and

then take the backseat in decision making. Their

decisions can be vetted, understood or questioned by

human actors before any action is taken. This leads

to scenarios of classic Human-Automation

Interaction, with the agent losing its autonomy in

decision-making and working instead as an advisor.

Where, however, the scenario is governed by

time pressure, the dynamic of interaction changes:

agents, with their superior computational speed and

ability to handle many concerns at once, are able to

cope with time pressure well beyond the point where

human actors become helpless. There is, though, less

time to exchange information and to understand the

agent’s decisions in depth; as such, issues of trust

come to the fore. Time pressure thus generates a

requirement for a higher throughput in exchanging

and processing information about the agent’s state,

plans and predictions. Since the cognitive abilities of

humans are fixed, the only ways to manage this are

compression or omission of information.

3.2 Predictability

When a system is predictable by an agent but not by

a human, there is an asymmetry of information,

which in turn makes the agent’s actions less

intelligible. In addition, the communication of

expectations not corresponding to the current

perception of the human actor can generate surprise

or exacerbate issues of trust. For example, while

some of the events within dynamic environments

can be fundamentally unpredictable, others are

opaque to a human, but probabilistically

approachable for Artificial Intelligence (AI), in

particular with Machine Learning (ML). In other

words, these environments provide the opportunity

to ‘shine’ for agents that can, in real time, make

better predictions or calculate according to better

models than humans are able to. Generally, these are

hard-to-explain mathematical models, however, and

even more so in real time situations. Explainable AI

(XAI) (Adadi and Berrada, 2018) is investigating

ways for Artificial Intelligence to communicate the

‘reasoning’ behind its predictions and decisions by

using explanation interfaces using techniques

borrowed from research in recommender systems

(Pu and Chen, 2006), but doing so is generally

feasible only in offline situations, in which time is

not a factor.

3.3 Number of Entities

The number of items and relationships to account for

in an environment directly generate cognitive

demand: systems can easily become so complex that

their scale and intricacy prevent humans from fully

understanding them; this is the case in any

sufficiently advanced work of ingenuity, from

skyscrapers to microprocessors, as well as in large

socio-technical systems, like a hospital. The same

can be said for the complexity of reasoning, many

examples of which can be found in the current

literature about XAI.

Shared Mental Models as a Way of Managing Transparency in Complex Human-Autonomy Teaming

155

Accounting for more entities individually

requires a larger mental model. While this is not

generally a problem for agents, it rapidly becomes

one for humans. Once the mental models diverge,

communication breaks down, and it becomes hard

for agents to describe their state and their actions in

a way that the human actor will understand –

creating a breakdown in transparency.

Having outlined how complexity factors of

dynamic environments can contribute to the

breakdown of HAT transparency, it is important to

look at the concept of Shared Mental Models, since

it is through them that human actors understand and

communicate, or otherwise fail to, an agent’s actions

in such an environment.

4 SHARED MENTAL MODELS

Shared Mental Models (Scheutz et al., 2017) are

knowledge structures that simplify reasoning about a

certain system (all models are simplifications that

maintain some properties and relationships while

losing others, and they exist for certain practical

purposes). In particular, a model is shared so that the

parties using it can perform the same reasoning,

simplifications and assumptions to communicate or

collaborate (language itself is dominated by models).

The adoption of a SMM (Cannon-Bowers et al.,

1993; Stubbs et al., 2007) and the careful choice of

the content and timing of their communication

(Bindewald et al., 2014; Goodman et al., 2016) are

critical mechanisms for agent transparency, as they

provide the anchoring for the information being

communicated.

An important feature of mental models is their

level of abstraction. Reduction and synthesis are two

ways to make models more abstract. In turn, a more

abstract description requires less data and is easier to

summarise. As an example, it is possible to talk

about Italy as being the shape a boot – a rather

abstract model – yet, as necessary, one may refer

instead to a map fitting the page of a book, or to an

accurate digital map, describing every street in the

country. Each is preferable to support different tasks

and contexts of use. The first one when describing to

a friend which part of the country one visited; the

second to show the administrative regions; and the

third to draw an itinerary from a hotel to a museum.

The different models are not interchangeable,

therefore it is critical to transparency that agents use

a model with the appropriate level of abstraction to

optimise mutual understanding within their current

context.

Although most people have an intuitive sense of

a model’s level of abstraction, a few formalisations

have been proposed (Hayakawa, 1949; Rasmussen,

1979; Sheridan, 2017; St-Cyr and Burns, 2001).

Rasmussen’s, in particular, neatly provides a

powerful taxonomy: Models of Physical Form;

Models of Physical Function; Models of Functional

Structure; and Models of Abstract Function, that

generalises well across domains.

In everyday interactions, people commonly use

SMM at different levels of abstraction to refer to the

same systems. For example, a car engineer may

think of an engine in terms of thermodynamics and

materials (physical form and function) when talking

about his work, but in terms of elasticity, fun or

power, when explaining the car to a friend (abstract

function). The curricula in computing have

recognised the ability of dealing with abstractions as

one of the fundamental computational skills (Grover

and Pea, 2013), and levels, or layers, of abstraction

are a fundamental concept in computing.

5 ADJUSTMENT OF THE LEVEL

OF ABSTRACTION OF THE

SHARED MENTAL MODEL TO

PRESERVE TRANSPARENCY

Given that transparency is seen as a prerequisite for

HAT effectiveness (Chen et al., 2018; Christoffersen

and Woods, 2002), it is desirable for an autonomous

agent to be aware of the current level of complexity

and to adapt its strategy to maintain it. While

adaptive strategies are not new in agent computing –

a rich tradition of research exists in regards to

adjustable autonomy (Bradshaw et al., 2003;

Johnson et al., 2011) – the focus of that research is

on adjusting the Level of Automation (LOA) in

semi-autonomous systems. We propose, instead, to

investigate how an autonomous agent may adjust the

level of abstraction of SMM to maintain

transparency while remaining fully autonomous. As

we have seen in the analysis of how complexity

affects transparency, the main disruption happens in

regards to more information having to be conveyed,

and understood, or less time to do so. In other words

the main limiting factor of transparency is

throughput. As we have seen, abstraction of a model

is a simplification by way of compression or

reduction of information. As a result, we propose

that when a dynamic environment becomes more

complex, and the information to transmit becomes

too much, thus breaking down transparency, it is

HUCAPP 2019 - 3rd International Conference on Human Computer Interaction Theory and Applications

156

possible to repair it by adjusting the level of

abstraction of the SMM so that the amount of

information that must be communicated remains

manageable for a human actor.

To do so, agents must continuously assess the

complexity of the environment by monitoring the

factors contributing to it – for example by keeping a

count of how many entities are present. Work in

adjustable autonomy can inform the present research

in regards to the strategies used by agents to detect

and model the conditions that trigger the

adjustments, this could include counting entities,

keeping track of rates of change of the environment,

and keeping track of the occurrence of unexpected

events. When an agent decides that it must switch

the level of SMM, it must do so in a way that is clear

and does not cause loss of Shared Situation

Awareness (SSA) (Grimm et al., 2018). One way of

doing this would be to make sure, prior to their

deployment, that human agents are equally familiar

with the different mental models they will encounter,

for example by use of training; other ways to prevent

loss of SSA would be to mark the switch through

explicit communication and to design the different

levels to be clearly distinguishable. Another concern

in turning to a more abstract mental model is that, by

definition, some of its ability to carry detailed

information is going to be lost, and therefore it

becomes crucial to establish what trade-off is most

advantageous between loss of transparency and loss

of information.

6 SUMMARY AND FURTHER

WORK

In this paper we have examined the relationship

between the complexity of dynamic environments

and transparency in HAT, and have argued that the

effects of complexity occur mainly as a consequence

of the demands that complexity puts on throughput

of communication between autonomous agents and

human actors, as well as on the ability of humans to

process larger amounts of information. We have

presented the concept of SMM and put forward that

varying the level of abstraction of the SMM may

mitigate the disruptive effects of complexity on

HAT transparency, by allowing the re-establishment

of a common ground in terms that can be understood

and communicated effectively under the new

complexity conditions. Finally, we have highlighted

some ways in which this could negatively affect

SSA.

Our current research programme is directed at

testing the above hypotheses. To do so we intend to

build a virtual environment for HAT in which

factors of complexity and SMM can be manipulated,

and to investigate ways of measuring transparency

within it. The overall objective of the research is

therefore to compare transparency in situations in

which complexity is increased with the SMM

unchanged, to other situations in which the SMM is

adjusted to compensate the increase.

REFERENCES

Adadi, A. and Berrada, M. (2018). Peeking inside the

black-box: A survey on Explainable Artificial

Intelligence (XAI). IEEE Access, 6. pp. 52138 -

52160.

Bainbridge, L. (1997). The change in concepts needed to

account for human behavior in complex dynamic

tasks. IEEE Transactions on Systems, Man, and

Cybernetics Part A:Systems and Humans, 27(3). pp

351-359.

Bellamy, R. (2017). Human-Agent Collaboration: Can an

Agent be a Partner? In: Proceedings of the 2017 CHI

Conference Extended Abstracts on Human Factors in

Computing Systems - CHI EA ’17. New York, New

York, USA: ACM Press. pp. 1289–1294.

Bindewald, J., Miller, M. and Peterson, G. (2014). A

function-to-task process model for adaptive

automation system design. International Journal of

Human Computer Studies, 72(12), pp. 822–834.

Bradshaw, J., Hoffman, R., Woods, D. and Johnson, M.

(2013). The seven deadly myths of autonomous

systems. IEEE Intelligent Systems, 28(3), pp. 54–61.

Bradshaw, J., Sierhuis, M., Acquisti, A., Feltovich, P.,

Hoffman, R., Jeffers, R. and Van Hoof, R. (2003).

Adjustable Autonomy and Human-Agent Teamwork

in Practice: An Interim Report on Space Applications.

In: H. Hexmoor, C. Castelfranchi and R. Falcone,

eds., Agent Autonomy. Boston, MA: Springer US. pp.

243–280.

Cannon-Bowers, J., Salas, E. and Converse, S. (1993).

Shared Mental Models in Expert Team Decision-

Making. In: N. Castellan, ed., Individual and Group

Decision Making: Current Issues. Hillsdale, NJ:

Lawrence Erlbaum Associates.

Chen, J. and Barnes, M. (2014). Human-Agent teaming

for multirobot control: A review of human factors

issues. IEEE Transactions on Human-Machine

Systems, 44(1), 13–29.

Chen, J., Barnes, M., Selkowitz, A. and Stowers, K.

(2017). Effects of Agent Transparency on human-

autonomy teaming effectiveness. In: 2016 IEEE

International Conference on Systems, Man, and

Cybernetics, SMC 2016 - Conference Proceedings.

IEEE. pp. 1838–1843.

Chen, J., Lakhmani, S., Stowers, K., Selkowitz, A.,

Shared Mental Models as a Way of Managing Transparency in Complex Human-Autonomy Teaming

157

Wright, J. and Barnes, M. (2018). Situation

awareness-based agent transparency and human-

autonomy teaming effectiveness. Theoretical Issues in

Ergonomics Science, 19(3), pp. 259–282.

Christoffersen, K. and Woods, D. (2002). How to Make

Automated Systems Team Players. Advances in

Human Performance and Cognitive Engineering

Research, 2, pp. 1–13.

Edland, A. and Svenson, O. (1993). Judgment and

Decision Making Under Time Pressure. In: Time

Pressure and Stress in Human Judgment and Decision

Making. Boston, MA: Springer US. pp. 27-40.

Fitts, P., Viteles, M., Barr, N., Brimhall, D., Finch, G.,

Gardner, E. and Stevens, S. (1951). Human

Engineering for an Effective Air-Navigation and

Traffic-Control System. Oxford, England: National

Research Council, Div. of.

Goodman, T., Miller, M., Rusnock, C. and Bindewald, J.

(2016). Timing within human-agent interaction and its

effects on team performance and human behavior. In:

2016 IEEE International Multi-Disciplinary

Conference on Cognitive Methods in Situation

Awareness and Decision Support (CogSIMA). IEEE.

pp. 35–41.

Grimm, D., Demir, M., Gorman, J. and Cooke, N. (2018).

Team Situation Awareness in Human-Autonomy

Teaming: A Systems Level Approach. In: Proceedings

of the Human Factors and Ergonomics Society Annual

Meeting, 62(1), pp. 149–149.

Grover, S. and Pea, R. (2013). Computational Thinking in

K-12: A Review of the State of the Field. Educational

Researcher, 42(1), pp. 38–43.

Ham, D., Park, J. and Jung, W. (2011). A framework-

based approach to identifying and organizing the

complexity factors of human-system interaction. IEEE

Systems Journal, 5(2), pp. 213–222.

Hayakawa, S. (1949). Language in Thought and Action.

Oxford, England: Harcourt, Brace.

Hoc, J. (1993). Some dimensions of a cognitive typology

of process control situations. Ergonomics, 36(11), pp.

1445–1455.

Hoc, J. (2000). From human-machine interaction to

human-machine cooperation. Ergonomics, 43(7), pp.

833–843.

How, J. (2016). Human-autonomy teaming. IEEE Control

Systems, 36(2), pp. 3–4.

Johnson, M., Bradshaw, J., Feltovich, P., Jonker, C., Van

Riemsdijk, M. and Sierhuis, M. (2011). The

fundamental principle of coactive design:

Interdependence must shape autonomy. Lecture Notes

in Computer Science, 6541 LNAI. Springer, Berlin,

Heidelberg. pp. 172–191.

Kaber, D. (2018). A conceptual framework of autonomous

and automated agents. Theoretical Issues in

Ergonomics Science, 19(4), pp. 406–430.

Klein, G., Woods, D., Bradshaw, J., Hoffman, R. and

Feltovich, P. (2004). Ten challenges for making

automation a team player in joint human-agent

activity. IEEE Intelligent Systems, 19(6), pp. 91-95.

Liu, D., Peterson, T., Vincenzi, D. and Doherty, S. (2016).

Effect of time pressure and target uncertainty on

human operator performance and workload for

autonomous unmanned aerial system. International

Journal of Industrial Ergonomics, 51, pp. 52–58.

Liu, P. and Li, Z. (2011). Toward understanding the

relationship between task complexity and task

performance. Lecture Notes in Computer Science,

6775 LNCS. pp. 192–200.

Luck, M. and D’Inverno, M. (1995). A Formal Framework

for Agency and Autonomy. In: First International

Conference on Multiagent Systems. AAAI Press MIT

Press. pp. 1–7.

Miller, C. (2014). Delegation and transparency:

Coordinating interactions so information exchange is

no surprise. Lecture Notes in Computer Science, 8525

LNCS. Springer, Cham. pp. 191–202.

Mosier, K., Fischer, U., Morrow, D., Feigh, K., Durso, F.,

Sullivan, K. and Pop, V. (2013). Automation, task, and

context features: Impacts on pilots’ judgments of

human-automation interaction. Journal of Cognitive

Engineering and Decision Making, 7(4), pp. 377–399.

Park, J. and Jung, W. (2007). A study on the development

of a task complexity measure for emergency operating

procedures of nuclear power plants. Reliability

Engineering and System Safety, 92(8), pp. 1102–1116.

Pu, P. and Chen, L. (2006). Trust Building with

Explanation Interfaces. In: Proceedings of the 11th

international conference on Intelligent User Interfaces

(IUI '06). ACM, New York, NY, USA, pp. 93-100.

Rasmussen, J. (1979). On the Structure of Knowledge - a

Morphology of Mental Models in a Man- Machine

System Context. Risö-M 2192, Risö National

Laboratory, Roskilde, Denmark.

Russell, S. and Norvig, P. (2009). Artificial Intelligence:

A Modern Approach, 3rd edition. Pearson.

Scheutz, M., DeLoach, S. and Adams, J. (2017). A

Framework for Developing and Using Shared Mental

Models in Human-Agent Teams. Journal of Cognitive

Engineering and Decision Making, 11(3), pp. 203–

224.

Sheridan, T. (2012). Human supervisory control. In G.

Salvendy, ed., Handbook of human factors and

ergonomics 4th ed. Hoboken, NJ, USA. John Wiley.

pp. 990–1015.

Sheridan, T. (2017). Musings on Models and the Genius of

Jens Rasmussen. Applied Ergonomics, 59, pp. 598–

601.

Shively, R., Lachter, J., Brandt, S., Matessa, M., Battiste,

V. and Johnson, W. (2018). Why human-autonomy

teaming? In: Advances in Intelligent Systems and

Computing (Vol. 586). Springer Verlag. pp. 3–11.

St-Cyr, O. and Burns, C. (2001). Mental models and the

abstraction hierarchy: Assessing ecological

compatibility. In: Human Factors and Ergonomics

Society Annual Meeting Proceedings, 45(4), pp. 297–

301.

Stowers, K., Kasdaglis, N., Rupp, M., Chen, J., Barber, D.

and Barnes, M. (2017). Insights into human-agent

teaming: Intelligent agent transparency and

uncertainty. In: Advances in Intelligent Systems and

HUCAPP 2019 - 3rd International Conference on Human Computer Interaction Theory and Applications

158

Computing (Vol. 499). Springer, Cham. pp. 149–160.

Stubbs, K., Wettergreen, D. and Hinds, P. (2007).

Autonomy and common ground in human-robot

interaction: A field study. IEEE Intelligent Systems,

22(2), pp. 42–50.

Sycara, K. (2002). Integrating Agents into Human Teams.

In: Proceedings of the Human Factors and

Ergonomics Society Annual Meeting, 46(1), pp. 413–

417.

Wohleber, R., Stowers, K., Chen, J. and Barnes, M.

(2017). Effects of agent transparency and

communication framing on human-agent teaming. In:

2017 IEEE International Conference on Systems,

Man, and Cybernetics, SMC 2017 (Vol. 2017–

January). pp. 3427–3432.

Wright, J., Chen, J., Barnes, M. and Hancock, P. (2016).

Agent Reasoning Transparency’s Effect on Operator

Workload. In: Proceedings of the Human Factors and

Ergonomics Society Annual Meeting, 60(1), pp. 249–

253.

Shared Mental Models as a Way of Managing Transparency in Complex Human-Autonomy Teaming

159