Congressional aides use AI to write speeches and analyze budgets, experts warn of content distortion and privacy risks

◎ Su Weizheng

Generative artificial intelligence (AI) has entered legislators’ offices. According to media reports, some congressional aides use AI to write speeches, create graphics and briefings, and also use AI to analyze budget documents and draft budget reduction or freeze proposals. Similar practices are also used in bill development. I can understand that staff have a heavy workload, but when AI output becomes the basis for questioning, budget review, or law revision, and the content is distorted or the position is biased, it may affect the people's rights and interests; improper data processing may even endanger privacy and official secrets.

According to reports, some congressional aides use AI to write speeches, produce graphics and briefings, and also submit budget documents to AI for analysis and draft budget reduction or freeze proposals. Similar practices are also used in bill development. (Drawing by this newspaper) The use of AI by administrative agencies may also involve personal information and confidentiality, and the work of Congress is directly linked to legislation, supervision and resource allocation, and is highly politically sensitive. Once erroneous information enters the discussion process, it may be amplified through public power. What data can be uploaded, where the output will be used, and who is responsible for the final judgment should all be clarified first.

The risk of using public information to create a picture card or a first draft of a speech is relatively low, but you should still check the facts and confirm the permission to use the material. When collating policies and public opinion, we must guard against biased sources and distorted summaries. When analyzing bills or budgets, you should pay attention to "illusions" such as AI's fictitious laws or data sources, and also check the year, amount, and unit.

Please read on...

Particular caution must be exercised regarding undisclosed agency information, negotiation contents, and submissions. The personal information or undisclosed corporate information provided by Chen Qingren does not mean that it can be transferred to external AI for processing. Before use, you must confirm whether the data processing is legal, and review the terms of service, whether the data is used for training, data retention, third-party access and international transfer arrangements; even if the name is deleted, the person involved may still be identifiable from the details.

The AI ​​Guidelines of the International Parliamentary Union (IPU) emphasize that AI should enhance human capabilities and cannot replace democratic deliberation and decision-making. Congressional assistants should assist legislators in listening to the opinions of all parties, comparing policy implications, and forming opinions through dialogue and consultation. Issues such as human rights protection, social security, and assistance to the disadvantaged involve multiple value conflicts, and value judgments cannot be left to models. Even if AI recommendations are adopted, decision-making responsibilities cannot be relieved.

The AI ​​Guidelines of the International Congressional Union emphasize that AI should enhance human capabilities and cannot replace democratic deliberation and decision-making. (Taken from the IPU official website) The review must also be traceable. The AI ​​draft must be compared with the budget, current regulations and official statistics one by one, and records of sources and revisions must be kept. Budget proposals must evaluate policy impact, while law amendment proposals must review their constitutionality and compatibility with relevant laws and regulations, and then be reviewed by staff and confirmed by committee members. When asking questions, we should pay attention to the quality of argumentation, and do not let competition in the number of proposals shorten the verification time.

Article 19 of the "Artificial Intelligence Basic Law" requires that when the government uses AI, it should conduct risk assessments, plan risk response measures, and formulate usage specifications or internal control management mechanisms based on the nature of the business. The Legislative Yuan should integrate the legal, information and information security professions, clearly list approved tools and purposes based on data sensitivity and usage risks, clarify data storage locations, retention periods, deletion methods and responsibilities, establish auditing and leak reporting procedures, and provide regular AI literacy, auditing and personal data protection training for members and assistants.

Article 19 of the Basic Law on Artificial Intelligence. (Extracted from the National Regulations Database website)

The value of professional staff lies in understanding the context of data, identifying conflicts of interest, and understanding the situation of the people. The time saved by AI can be used to research and listen to public opinion. Political judgment and value selection still need to combine expertise, experience and public opinion. Technology can assist in judgment, but the responsibility still lies with the person. AI can enter Congress, but value cannot be outsourced.

(The author is a university lecturer and former secretary to the president of the Legislative Yuan)