The Potential of Using AI in Government
““...The successful use of AI in public administration hinges on civil servants formulating the right prompts by becoming familiar with AI tools.””
OpenAI launched ChatGPT last November, sparking widespread interest in Artificial Intelligence (AI). Now, even governments are planning to explore the use of AI in administration through a "responsible experimentation approach." Here is a list of public administration activities where civil servants can utilize generative AI in their day-to-day work:
File work
Simply put, the file is a document in which decisions are made. The initial note is the starting point of the file. This note contains all the material to make a decision, such as background, legal provisions and views of other departments. The note ends with a list of options with their positive and negative consequences. The selected alternative is the decision.
The steps that lead to the generation of note by using ChatGPT are:
First, all available material is summarized. The material could include paragraphs, articles, letters, or any relevant information to be condensed. After selecting the material, a prompt is given: “Summarize the following text for me”. Once ChatGPT asks for the material, it is pasted in the dialog box.
Second, all the summarized material is combined to make a story of the issue under examination. For this, the summarized texts are logically arranged and then a prompt is given to merge. The result is a “draft” in story form, which is then simplified.
Third, options with the consequences are generated. Here, the ChatGPT can also be asked to access external information, in addition to generating options and their consequences. The output will be the initial note containing multiple options to address the issue/problem.
As the file moves through the hierarchy, intermediate levels could add missing facts or give new interpretations to the story, or frame the issue in a different way. At each stage, ChatGPT has the potential to add or justify the views recorded on the file from external sources. ChatGPT would be able to provide the underlying reasoning, once an alternative is selected as the decision.
External communications
The decision made on file is communicated to the outside world in the form of letters, memos, etc. The draft of the external communication is created and then ChatGPR is prompted to simplify, as well as convert the communication to active English.
Generally, civil servants devote a lot of attention to the language of external communications. The ChatGPT would do this work automatically for them, thus saving effort and time. Furthermore, simplification of language would make government-speak and legalese easily understandable to the common citizen.
Meetings
One major activity of civil servants is attending or holding meetings. Using AI, civil servants can convert audio recordings of meetings into text. This can be summarized, simplified, and converted to active English to produce concise summaries of meetings. Again, time and effort that junior civil servants have to spend on producing accurate accounts of meetings would be saved.
Real-time Monitoring
Today, there is a humungous quantity of data being generated by field functionaries using different types of apps. At present, the focus is on generating, updating and reporting data. Chatboxes provide an opportunity to analyze such data and provide customized actions to be undertaken by functionaries working at different levels in Governments.
For example, let us consider the monitoring of reduction of anemia in pregnant women - a SDG goal. The steps are: test all pregnant women every month, based on these tests identify the women whose anemic levels are falling or have remained unchanged during the month, find out if these women are regularly eating hot cooked meals and iron tablets in anganwadis, and so on.
All the data is available in the apps used by the ground-level functionaries of the health and women and child departments. The chatbox would use this date to provide the following information to different levels in the Government:
● Field workers - number of pregnant women regularly eating hot cooled meals and taking the supplements (iron, folic acid).
● Supervisors- number of women who were eating the food in the anganwadis and regularly taking the supplements, yet their anemia level was increasing.
● Policy-makers - determine if take home rations would serve the policy aims better as some women are unable to come to the anganwadi everyday to eat hot cooked meals everyday.
From the above instances, it is clear the AI has potential for use in the following areas in public administration:
● Summarize and merge texts (see Elicit, a free online tool).
● Convert texts into active English.
● Extract crucial information.
● Sort textual content.
● Pull specific information from unstructured text.
● Classify information into predefined categories (see literature based discovery tools).
● Answer specific questions from dynamic datasets in real-time.
Most importantly, the successful use of AI in public administration hinges on civil servants formulating the right prompts by becoming familiar with AI tools. There is no formula to put together right prompts, and trial-and-error offers a reasonable way forward.
