In University DX, What Truly Matters Might Not Be 'AI Implementation' but 'Design of Responsibility'

As the use of generative AI spreads, I believe the work of university staff will change significantly in the future.

Currently, the focus is on individuals using ChatGPT as a convenient tool for tasks like writing, summarizing, and brainstorming. However, from now on, I think we will be required to go beyond simply 'using AI' and start integrating it into business workflows such as applications, reviews, inquiry responses, and document verification.

However, when considering the introduction of AI into university administrative tasks, it is not as simple as saying, 'If we leave it to AI, we can improve efficiency.'

Rather, I feel that what is important is designing what to entrust to AI, where humans should make judgments, and who will bear the final responsibility.

Is the introduction of generative AI closer to RPA or DX?

University staff tasks involve a lot of administrative work, such as creating and checking documents, data entry, and answering inquiries.

Therefore, the term 'generative AI utilization' brings to mind an image of reviewing entire business processes, like RPA or DX, rather than an individual typing questions into a chat screen.

For example, suppose there is a task where, after receiving an application form, one must check for missing information, organize the necessary data, and forward it to the person in charge.

Of these, formal checks and information organization are tasks that AI is highly likely to be good at. On the other hand, there remains anxiety about leaving everything to AI, including judgments that take exceptional circumstances into account or final decisions such as whether to approve an application.

When introducing AI, it is necessary to think about which parts to automate and which parts humans should handle, rather than simply replacing existing tasks with AI as they are.

I believe this is a DX initiative that involves reviewing the business design itself, rather than just introducing a tool.

Primary response by AI, final confirmation by humans

At this point, what I am imagining is a workflow where AI handles the primary response and humans perform the final confirmation of the results.

For example, AI checks for deficiencies in documents and creates draft responses to inquiries. The person in charge confirms the content, makes corrections if necessary, and then performs the final processing or response.

If we compare it to a hospital, the role of AI might be closer to an assistant providing material for judgment rather than a doctor who makes a definitive diagnosis.

AI organizes information first and points out areas where there might be oversights. Based on that, humans make the final decision while considering the situation and background.

While generative AI is very capable, it is a mechanism that generates answers based on probability and does not always provide correct answers. There is also a possibility that it will present content that differs from the facts in a plausible way.

That is precisely why it is necessary to incorporate human confirmation into the workflow, rather than adopting AI output as is.

Why do humans make the final decision?

The biggest reason why humans make the final decision lies in the location of responsibility.

Even if we can entrust judgment to AI, we cannot make AI take responsibility when a problem occurs.

If a disadvantage occurs to students or examinees due to incorrect guidance, the university or the person in charge will ultimately bear the accountability.

Universities, in particular, handle information that has a major impact on students' lives, such as entrance exams, grades, student registration, scholarships, and personal information. It is difficult to prioritize only efficiency and use AI while the judgment process remains opaque.

What is important is not just thinking about 'how far can we let AI judge?'

We must design it to include 'if a problem occurs due to that judgment, who will explain it and who will take responsibility?'

I believe that in the AI era, business design requires visualizing not only the flow of processing but also the flow of responsibility.

Literacy education is needed before tool introduction

In utilizing generative AI at universities, there are also issues with the AI literacy of staff.

When convenient tools are provided, there is a possibility that staff might input work-related documents or data without fully understanding their mechanisms or risks.

The handling of personal information and confidential information requires particular caution.

If student names, grades, contact information, entrance exam information, etc., are input into external generative AI services without checking the terms of use, it could lead to problems in information management.

We must move from the stage of 'using generative AI because it is convenient' to the stage of understanding 'which service, what information, and for what purpose it is okay to input.'

Therefore, before introducing AI tools, it is necessary to establish minimum usage rules and literacy education.

I think it is also important to show concrete examples of situations where it can be used safely and situations where confirmation is required, rather than just listing prohibited items.

The success or failure of AI implementation is determined by operational design

The performance of generative AI will continue to improve in the future.

However, no matter how high-performance an AI appears, university operations will not automatically improve.

Which tasks will it be used for? What information is okay to input? Who will check the AI's results? How will errors be corrected if they occur?

If it is introduced without deciding on these operations, it could actually lead to an increase in confirmation work or the emergence of new risks.

What is truly important in university DX is not the introduction of the latest AI itself.

It is about separating the roles of AI and humans, and clarifying the final judgment and the location of responsibility. And it is about establishing rules and education that allow staff to use it safely.

When considering the introduction of generative AI, shouldn't we think from the perspective of 'what kind of business and responsibility mechanism do we want to create,' rather than focusing only on the performance of the technology?