Human-Centered Explainable Artificial Intelligence for Marine Autonomous Surface Vehicles

Round 1

Reviewer 1 Report

Comments for author File: Comments.pdf

Author Response

Comment 1: Extensive editing of English language and style required

The authors have gone through the language very carefully and shared the manuscript with a colleague who is a native English speaker. Together, we managed to find 43 instances where editorial changes were needed, either in the form of spelling, grammar, or style. All instances were updated with ‘Tracked Changes’ on in the Word document.

However, we still cannot be sure that the language is perfect, and we expect there may still be spelling and grammar errors. We are open to hiring editorial services prior to eventual publication.

Comment 2: Methods “must be improved”

We agree and we improved the Method considerably. In the updated manuscript, we adopt a method ‘model-based reasoning’ that was developed by Nancy Nercessian in her book Creating Scientific Concepts (MiT Press, 2008). Model-based reasoning states that concept creation consists of three interrelated processes: analogy, visualization, and mental simulation. We describe the method in detail and draw from similar methods of concept creation in the social sciences. We have also re-written Section 2.3. Methodological considerations in line with the new method.

Comment 3: Results “must be improved”

We have re-organized the structure of the Findings around the three processes described in the Method (analogy, visualization, and mental simulation) (see Comment 2). We also emphasize to readers that the examples presented are not the findings in themselves, but rather the empirical observations that support the formation for the concept of human centered XAI.

Comment 4: The concept of “control threshold” should be discussed more

This has been addressed in Section 3.2.4 (line 365-8).

Comment 5: More detailed data should be shown for a better understanding in Figure 13

We have added in a close-up photograph of the ASV Telemetron, and we hope allows for a better understanding of the Figure. (Note that this is now Figure 14 in the updated manuscript.)

Comment 6: The concept of “mental models” should be explained more deeply, which is still confused about the entire analysis

We have added a clear definition of “mental model” in the text (Section 2.3, line 213-4). We also expressly refer to this definition throughout the article to remain consistent (for example, line 242-3 and line 575-6). We also refer to “mental model” in quotations through the paper to emphasize that this concept has many interpretations, but that we are using the working definition presented in Section 2.3.

Comment 7: “…The main method is not presented vividly. Seemingly, most of the work is introduced from other research results.”

Please see the response above in Comment 2, which explains the detailed updates to the Method. Also, we would like to emphasize that examples presented are not the findings in themselves, but rather the empirical observations that support our formulation for the concept of human centered XAI. All examples shown come from either our own research or from those of our colleagues and collaborators at NTNU, and all were obtained with express permission obtained in person (see Comment 3).

Reviewer 2 Report

This paper introduced the development of the AI technology application in autonomous surface vessels and the significance of Human-centered explainable artificial intelligence for marine autonomous surface vehicles, and then, it proposed the concepts, strategies using interaction design perspectives. It is necessary to build trust among the technology’s primary users. 

However, there are not any precise definitions of the concepts of human-centered explainable artificial intelligence for marine autonomous surface vehicles in the article, authors should propose the concept clearly and clarify the difference with the present definition. 

In terms of the theory, methods and strategies should be the basis of solving problems in research or applications, but not the simple exhibitions and description of case studies. 

The authors should introduce the interaction design and methodologies of human-centered explainable artificial intelligence for marine autonomous surface vehicles using interaction design perspectives in detail, the physic verification also should be supplemented. Moreover, the format of figures and tables should be checked.

Author Response

Comment 1: Moderate English changes required.

The authors have gone through the language very carefully and shared the manuscript with a colleague who is a native English speaker. Together, we managed to find 43 instances where editorial changes were needed, either in the form of spelling, grammar, or style. All instances were updated with ‘Tracked Changes’ on in the Word document.

However, we still cannot be sure that the language is perfect, and we expect there may still be spelling and grammar errors. We are open to hiring editorial services prior to eventual publication.

Comment 2: Methods must be improved.

We agree and we improved the Method considerably. In the updated manuscript, we adopt a method ‘model-based reasoning’ that was developed by Nancy Nercessian in her book Creating Scientific Concepts (MiT Press, 2008). Model-based reasoning states that concept creation consists of three interrelated processes: analogy, visualization, and mental simulation. We describe the method in detail and draw from similar methods of concept creation in the social sciences. We have also re-written Section 2.3. Methodological considerations in line with the new method.

Comment 3: “… there are not any precise definitions of the concepts of human centered explainable artificial intelligence for marine autonomous surface vehicles in the article and authors should propose the concept clearly and clarify the difference with the present definition.”

In the updated manuscript, we have included a clear definition of human centered XAI (Section 4). We have put the definition in a box to make it clear to readers.

We also draw reader attention to Table 1, in which we clarify the difference between human centered XAI and the present definition, i.e., technology centered XAI.

Comment 4: “…In terms of theory, methods and strategies should be the basis of solving problems in research or applications, but not the simple exhibition and description of case studies.”

We agree. We would like to draw the reviewer’s attention to Comment 2, which describes the new method structure in the paper. Also, have emphasized in the article that the point is not to exhibit and describe simple case studies, but rather to show that current research practices are pointing to the existence of new design processes – processes that we formulate as “human centered XAI.” The case studies are, in other words, the empirical data that supports our concept formulation.

Comment 5: “The authors should introduce the interaction design and methodologies of human centered XAI for ASVs using interaction design perspectives in detail.”

Please see Comment 2, where we describe the new method we adopted in detail. We also re-wrote Section 2.3. Methodological considerations to incorporate extra background detail about our methodological perspectives.

Comment 6: “The physic verification also should be supplemented.”

We interpret “physic verification” as meaning “empirical observation that supports our hypothesis.” In this sense, the empirical observation consists of the examples we drew from current research practice about ASVs. This empirical detail supports our hypothesis that a new concept called human centered XAI exists in current ASV design practices as the field prepares for wide scale deployment and end user interaction.

Comment 7: “…the format of figure and table should be checked.”

We have checked the format against the Word template for JMSE and it appears to be in full accordance. However, we are open to editorial services (at our own cost) if there should be something in the formatting template we have misunderstood that requires that the document be updated.

Reviewer 3 Report

The Evaluation of the paper titled: “Human centered explainable artificial intelligence for marine autonomous surface vehicles: concepts, strategies, and case study using Interaction Design perspectives”. The paper proposes investigation of how human centered principles of interaction design contribute to XAI for ASVs. The AI is important field and I am giving support to the authors for investigation this topic. Overall, the paper is interesting and I suggest it for consideration for publication. The authors should consider following suggestions for the revision of the paper:

- The abstract is loosely written. It is not as informative as expected. A standard abstract must present, without leaving any doubt, the objective of the paper precisely; source of data (which is not present in your abstract) and analytical approach used; key findings and any policy implication and recommendations.

- What are scientific contributions of this paper? This should be clarified to the readers.

- The results section - Technology centered XAI compared to human centered XAI should be better explained and discussed.

- Conclusions section should focus on the results, scientific contributions, limitations of the study, and future directions for the research.

Author Response

Comment 1: “The abstract is loosely written…”

We have re-written the abstract. We believe it is now more informative because it follows the standard abstract style more closely, including the objective, sources of data, analytical approach, key findings, and conclusion.

Comment 2: “What are scientific contributions of this paper?”

We have clearly added the scientific contribution in this paper in two locations: 1) Abstract (line 21) and 2) Conclusion (line 716).

Comment 3: “The results section… should be better explained and discussed”

We agree. We have re-organized the structure of the Findings around the three processes described in the new method we have adopted (which is called "model-based reasoning").

Note that we also included two new sub-sections: (1) Section 3.3.4 Human-AI Collaboration, which describe an important observation we did not include in the original manuscript; and (2) Section 3.4 Summary, which clearly summarizes the results for readers.

Comment 4: “Conclusions section should focus on the results, scientific contributions, limitation of the study, and future direction for the research”

We implemented all your recommendations in a new version of the Conclusion.

Round 2

Reviewer 1 Report

It seems all issues have been responded to properly. 

Reviewer 2 Report

The authors have properly addressed the previous comments. The quality of the manuscript has been improved.