How can AI be used to improve work productivity while keeping information safe and respecting the law? These questions were discussed at the Young Civil Servants and the AI Era forum.
The forum took place on the afternoon of October 9 as part of the 10th National Informatics Contest for Young Officials and Civil Servants, organized by the Central Committee of the Ho Chi Minh Communist Youth Union. Experts and speakers from government agencies, schools and businesses attended.
AI hallucinations and the trap of convenience
One of the forum's main topics was AI hallucinations. Many young officials trust documents and reports produced by AI because the wording appears smooth and logical. Dr. Tran Tien Cong, acting head of the Artificial Intelligence Department at the Posts and Telecommunications Institute of Technology, warned about this technical risk.
Dr. Cong said current large language models work probabilistically. AI predicts the next word from the large volume of data used for training, so some error is inevitable. He put the rate at about 3% for ChatGPT.
"Three percent may sound small. But if AI is used to summarize a 1,000-page document, that equals 30 pages of incorrect information. In public administration, this is unacceptable," Dr. Cong said.
Speakers therefore recommended writing precise prompts, asking AI to provide clear sources and, most importantly, checking the information personally before issuing a document.
The red line around citizens' data
AI's convenience can sometimes push officials into breaking the law without intending to. The forum presented a hypothetical case: to meet a deadline, an official uploads an Excel file containing citizens' personal data, including names, identity-card numbers and phone numbers, to a free AI tool and asks it to create a chart.
Ngoc Minh Thuan, representing the National Population Data Centre, said sending citizens' data to third-party AI servers on the internet amounts to exposing information. He said it violates the Personal Data Protection law, Decree 13, and could also violate laws protecting state secrets.
To handle such work without breaking the law, experts recommended masking: remove all personal identifiers and keep only quantitative data before asking AI to process it. A safer option is to use AI only for suggestions about algorithms and working methods instead of handing over the raw data.
Legal responsibility for using AI
Students and young officials raised another difficult question: if AI incorrectly summarizes a case file, an official approves the wrong recipient and the state budget suffers a loss, who is responsible?
Huynh Hoang Khai, a representative of the People's Procuracy's Regional Office No. 7 in Can Tho, said from a legal perspective: "AI software is only a tool created by people. It has no legal personality and cannot bear administrative or criminal responsibility. The civil servant signs and stamps the document. If that person trusts AI too much, skips checking the original file and causes consequences, that individual bears full legal responsibility for the negligent act."
Bui Van Quyet, deputy head of the Administrative Reform Department at the Ministry of Home Affairs, said public agencies should improve young officials' awareness, teach them to separate reliable information from errors and issue temporary rules for using AI.
In the long term, Dr. Tran Tien Cong said government agencies would need to build internal AI systems for their own use. Investing in supercomputer infrastructure and training models would be very costly and complex, but he described such systems as necessary to protect digital sovereignty and national data security.
Sources: Thanh Niên
