Make Automation G.R.E.A.T (Again)

Julien Donnot

, Daniel Hauret

, Vincent Tardan

and Jérôme Ranc

French Air Warfare Center, 1061 Avenue du Colonel Rozanoff, Mont-de-Marsan, France

Thales AVS, 75-77 Avenue Marcel Dassault, Mérignac, France

Keywords: Military, Aviation, Artificial Intelligence, Ergonomics, Autonomy.

Abstract: Integration of Artificial Intelligence Based Systems in current and future Fighter aircrafts has begun. More

than automation, these systems provide autonomy that will certainly imply changes in fighter pilots cognitive

activity. The current study aimed to test a methodology conceived for the production of ergonomic guidelines.

We collected data related to the use of a path solver with fighter pilots by assessing trust calibration in AIBS.

Analysis of under and overtrust situation led us to formulate relevant ergonomic principles for the tested AIBS

but relevant too for different AIBS.

1 INTRODUCTION

The objective of the current work was to test a method

designed for the evaluation of human trust behaviors

in Artificial Intelligence Based System (AIBS). The

French Air Force needs to develop the capability to

evaluate usability of AIBS employed on board of a

fighter aircraft. Trust in autonomy can be considered

as a key factor for usability. As French Air Force has

to be prepared for forthcoming conflicts, conception

of future air combat systems implies to anticipate

ergonomics issues, especially in terms of decision-

making, workload and errors in fighter aircraft

cockpits. In the field of prospective ergonomics

(Robert & Brangier, 2009), Brangier and Robert

(2014) point out the difficulty to represent future

activity related to a system that does not yet exist.

Considering that the main characteristic of future

fighter aircrafts will be the obligation for pilots to

collaborate with an AIBS (Lyons, Sycara, Lewis &

Capiola, 2021), there is a real need to think up this

future collaborating activity.

Lyons et al. (2018) warned about the specificity

of trust in future autonomy, including AIBS, in the

field of military aviation. Leading studies with real

operators, with real tools and real consequences (R3

concept) appears as the most relevant. In the field of

military aviation, real operators are fighter pilots, real

tools are fighter aircraft (Rafale) and real

consequences appear in a tactical environment. Too

few studies reported knowledge about the French

fighter pilot activity. Amalberti (1996) touched on

some specific features of this activity, Guérin,

Chauvin, Leroy, and Coppin (2013) adapted a

Hierarchical Task Analysis method to one air operation

and Hauret (2010) was the first to be interested in pilot

collaboration with an artificial agent.

To define what would be the collaborating activity

in a future fighter cockpit, ergonomists need to assure

usability of human machine interfaces. Bastien and

Scapin (1995) described a set of criteria designed for

conception guidance. These guidelines were thought

to conceive human-computer interfaces. Given that

functions performed by AIBS are and will be more

complex and sometimes innovative, AIBS conception

guidelines deserve to be considered. In the current

study, authors focused on trust as a critical factor for

usability of AIBS on board a fighter aircraft. Then,

the objective of the study was to produce conception

guidelines to increase usability by building pilot’s

trust in AIBS.

Trust is a complex concept depending on

individual, organizational and cultural context (Lee &

See, 2004) but we choose to focus on its calibration

in the current study by considering the lack and the

excess of trust in a specific AIBS. A large number of

methods and metrics can drive analysis of trust levels

(Hoff & Bashir, 2015). In order to assess usability in

relation with trust, pilots’ behaviors prevailed over

pilots’ feelings. Therefore, we develop a method

immersing operational fighter pilots in a simulated

combat air mission with an operating AIBS.

Experimental objectives were 1) to identify causes of

observed trust levels leading to understand pilots uses

Donnot, J., Hauret, D., Tardan, V. and Ranc, J.

Make Automation G.R.E.A.T (Again).

DOI: 10.5220/0011984300003622

In Proceedings of the 1st International Conference on Cognitive Aircraft Systems (ICCAS 2022), pages 92-95

ISBN: 978-989-758-657-6

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

of the AIBS and 2) to formulate ergonomics

principles justified by trust issues.

2 METHODOLOGY

2.1 Participants

Four military pilots were tested. All the pilots were

experimented on Mirage 2000 and familiar with

aircraft simulator.

2.2 Apparatus

Participants were asked to perform a flight as an

operational mission on a Mirage 2000 simulator.

Mission was built and played on DCS world®.

Functions of the AIBS PathOptim® were integrated

into a Tacview® interface.

The mission demands each pilot to cross a hostile

territory to bomb a target on time and to come back

safe. Pilots had to respect several restrictions like a

maximal height of 10kft, no detection by enemy radar

and no engagement by enemy air defense systems.

Flying over 10kft for the first time made an enemy

fighter aircraft took off for interception and raised

mission difficulty.

2.3 The AIBS

PathOptim is a 3D track solver based on the Genetic

Fuzzy Trees method, which gives to the pilot the

choice of three types of track to reach the target as fast

as possible. By integrating characteristics of enemy

air defense systems, each track is calculated for a

fixed minimal height without overflying 10kft at a

fixed speed. Green tracks avoid as much as possible

to enter in SAM ring, red tracks are the fastest tracks

even if the pilot must enter in one or several SAM

rings and amber tracks are a compromise between

survival and fastness (Figure 1).

Figure 1: PathOptim in the tactical display.

2.4 Scenario

During the mission, some new air defense threats

appear if the pilot enters in the triggering area. Events

could be SAM appearance, SAM disappearance and

a new threat not considered by PathOptim (soldiers

with Rocket Launchers - RPG). Pilots were informed

of PathOptim limitations. Events were built to force

pilots to use PathOptim. As a control condition and

the only exception, the threat by soldiers with RPG,

was created to confirm that pilots were aware of the

uselessness of PathOptim is this specific situation.

Pilots must react as following: Refresh

PathOptim, choose a track and follow the track. Trust

coding results from the combination of the possible

behaviors produced by pilots (Figure 2). Pilots could

use PathOptim at any time even if no tactical change

pops up.

The experiment setup was designed to deduce

trust, undertrust and overtrust from PathOptim uses

by pilots.

2.5 Data Analysis

Each planned events and each supplementary use of

PathOptim were analysed. Analysis consisted in

replaying the mission. Tactical situation, aircraft

spatial localization, pilots/cockpit interactions,

pilots/PathOptim interactions, eye-tracking and radio

communications were analyzed to understand pilots’

behaviors. The workload was monitored according to

the dual-task paradigm. Pilots had to add a triplet of

digits and give the total. Data were analyzed in terms

of accuracy and delay in seconds.

Trust was observed when pilots used PathOptim

and that PathOptim was helpful (Figure 2). With

understrust, the pilot did not use it whereas

PathOptim was helpful. In undertrust we

distinguished defiance and distrust. With defiance,

the pilot refreshed PathOptim but did not follow a

proposed path and in distrust, the pilot did not refresh

PathOptim nor followed a virtual path. In overtrust,

Figure 2: Hierarchical action tree and corresponding trust

levels.

Make Automation G.R.E.A.T (Again)

the pilot did not refresh PathOptim although he

needed to do it. Another overtrust situation was when

the pilot refreshed PathOptim whereas it would not be

useful (i.e., soldiers with RPG not detected by

PathOptim).

3 FINDINGS AND DISCUSSION

3.1 General Results

All the pilots achieved the mission successfully by

bombing the target and without being killed by enemy

air defense.

3.2 Trust Calibration

Pilots followed a path proposed by PathOptim

between 62% and 73% of the flying time. The number

of PathOptim utilizations varied according to the

pilots and produced from six to thirteen events. Trust

represented from 33% to 82% of trust levels against

undertrust and overtrust. Trust tends to increase with

number of events and therefore with uses.

Regarding the workload, high delays can be

explained by enemy’s aircraft monitoring, integration

of new events in the situational awareness and path

following.

Qualitative analysis led to identify effects and

causes for each event generating undertrust or

overtrust. Thus, ergonomists produced requirements

for PathOptim development. Based on these

requirements, ergonomic principles have been

formulated.

3.3 G.R.E.A.T Principles

3.3.1 G for Guidance

The pilot needs the necessary information to follow

the tracks proposed by PathOptim, so AIBS must

carry out tasks helping the proposal execution (ex:

diving tight curve) to reduce flying errors from the

pilot.

3.3.2 R for Recommendation and Realism

Recommendation: The pilot needs information on the

currently best use of the system, so AIBS must be able

to detect inappropriate use by the pilot (ex: no green

path because amber path was forced) to make use of

the full system capabilities.

Realism: The AIBS has to treat environmental factors

responsible for an effect on security or performance

to gain trust with a well-fitting proposal (ex: flying in

the valley instead of above hills).

3.3.3 E for Margin of Error

The pilot needs to be aware of the margin allowed for

execution of the AIBS proposal (ex: the aircraft has

to be less than 100m from the virtual path). Thus,

AIBS must inform the pilot about the conditions of

the validity proposal to reduce interpretation errors

and gain trust for tactical decision-making and track

following (ex: best path as long as the pilot is no more

than 5 sec late = low margin of error).

3.3.4 A for Automation

AIBS must treat automatically environmental and

tactical changes to relieve the pilot from considering

changes.

3.3.5 T for Transition

AIBS proposals must tend to be univocal and

understandable to reduce pilot’s doubt and speed up

decision-making.

4 CONCLUSIONS

G.R.E.A.T principles present the benefit of being

justified by real use cases. These principles will

probably evolve as long as ergonomists experiment

new and various use cases.

REFERENCES

Amalberti, R. (1996). La conduite des systèmes à risques.

Paris : PUF.

Bastien, J. M. C., & Scapin, D. L. (1995). Evaluating a user

interface with ergonomics criteria. International Journal

of Human Computer Interaction, 7, 105-121.

Brangier, E., & Robert J.-M. (2014). L'ergonomie

prospective : fondements et enjeux. Le Travail Humain,

77, 1-20. doi: 10.3917/th.771.0001

Guérin, C., Chauvin, C., Leroy, B., & Coppin, G. (2013).

Analyse de la tâche d'un pilote de Rafale à l'aide d'une

HTA étendue à la gestion des modes dégradés.

Communication présentée à la 7ème conférence

Epique, Bruxelles, Belgique. (p.21-26).

Hauret, D. (2010). L’autorité partagée pour mieux décider

à bord du Rafale., Communication présentée à la 12ème

conférence internationale ERGO’IA 10, Bidart, France:

ACM Press

ICCAS 2022 - International Conference on Cognitive Aircraft Systems

Hoff, K. A., & Bashir, M. (2015). Trust in automation:

Integrating empirical evidence on factors that influence

trust. Human Factors: The Journal of the Human

Factors and Ergonomics Society, 57, 407–434.

https://doi.org/10.1177/0018720814547570

Lee, J. D., & See, K. A. (2004). Trust in automation:

Designing for appropriate reliance. Human Factors, 46,

50–80. https://doi.org/10.1518/hfes.46.1.50_30392

Lyons, J. B., Ho, N. T., Van Abel, A. L., Hoffmann, L. C.,

Fergueson, W. E., Sadler, G. G., et al. (2018).

“Exploring trust barriers to future autonomy: a

qualitative look,” in Advances in Human Factors in

Simulation and Modeling, Advances in Intelligent

Systems and Computing 591, ed. D. N. Cassenti

(Cham: Springer).

Lyons, J. B., Sycara, K., Lewis, M., & Capiola, A. (2021).

Human–Autonomy Teaming: Definitions, Debates, and

Directions. Front. Psychol. 12:589585. doi:10.3389/

fpsyg.2021.589585

Robert, J-M., & Brangier, E., (2009). « What is prospective

ergonomics? A reflection and position on the future of

ergonomic”, in B.-T. Karsh (Ed.): Ergonomics and

Health Aspects, LNCS 5624, pp. 162–169.

Make Automation G.R.E.A.T (Again)