Case 9.18: "98 per cent disposal"

Difficulty: Medium
Syllabus point: 7: AI versus conscience
Values and laws tested: Target against indicator, honest reporting, accountability to the citizen, human review; the natural-justice principle of a hearing and a reasoned reply; Conduct Rule 3(1); Nolan's "honesty"

Illustrative example"98 per cent disposal"

Pratap is the nodal officer of the district grievance cell in a district of Rajasthan. The district has installed an AI-based grievance portal from a private vendor. The system reads the words of a complaint, classifies it, sends a ready-made reply and shows the complaint as "disposed". The dashboard shows last month's disposal at 98 per cent, against 61 per cent earlier. The District Collector wants to show this proudly at the state review meeting, and the vendor says, "The system's accuracy is very high." Pratap reads a random sample of 40 disposed complaints. In 11 the reply is irrelevant, for example a complaint of a stopped old-age pension answered with "your letter has been forwarded" and no action. Some such complaints have been filed again. The meeting is in two days.

Stakeholders: citizens who complain, especially the old and poor (the weakest party), the District Collector, the vendor, Pratap, the departments that resolve complaints, state-level officers.
Issue: the indicator (98 per cent) and image against the real purpose (resolving the complaint); honest reporting against pleasing a senior. Blind trust in the AI's number against conscience in his own sample test.
Options: A. Present the 98 per cent in the meeting and hide the sample (praise; but a false picture, unheard complaints, answerability later). B. Shut the portal down completely (a clear message; but real gains are lost too, and the workload rises). C. Report honestly with the sample's findings, confine the AI to classification and drafts, close a complaint only after human review instead of showing "disposed", and change the indicator to "actual resolution, reopen rate and citizen satisfaction" (the showy number will fall; but truth and improvement both survive).
Best: C. Nolan's "honesty" and Conduct Rule 3(1) call for telling the truth. The principle of natural justice is that the citizen gets a hearing and a reply with reasons; an irrelevant ready-made reply is the opposite.

Lesson

Action: Short term: act at once on the 11 irrelevant complaints and reopen them, a written sample report to the Collector, ask the vendor to explain the error rate. Long term: human approval before disposal, regular random sample checks, new indicators, a way for the citizen to give feedback, a quality condition in the vendor's contract.

Practice Questions

Practice questionabout 100 words

(10 marks, 100 words) What should Pratap do? Solve with the five-step card.

Model answer

The weakest stakeholders are complainants, especially the old and poor; the Collector, the vendor and departments are involved too. The issue is the 98 per cent indicator against real resolution and honest reporting. I have three routes: show the figure and hide the sample, shut the portal, or report the truth and limit the AI. The first is a false picture; the second loses real gains. I choose the third, because Nolan's honesty and Rule 3(1) demand the truth. I would reopen the 11 complaints, give the Collector the sample report and close complaints only after human review. Long term: resolution-based indicators.

Practice questionabout 50 words

(5 marks, 50 words) Explain the difference between a target indicator and the real purpose in this case.

Model answer

An indicator is a tool; the purpose is resolving the complaint. When the indicator becomes the goal, show grows.

Practice questionabout 50 words

(5 marks, 50 words) Suggest long-term measures to keep conscience and accountability in AI-based grievance redress.

Model answer

Human approval, sample checks, new indicators, citizen feedback, quality terms in the contract.