Lessons learned in using AI

(PLVN) - The recent incident where the People's Committee of a northern province issued a document to rectify the advisory work of a specialized department after discovering a series of serious errors in a report submitted to the Provincial People's Committee; suspected use of online tools and artificial intelligence (AI); serves as a lesson for everyone when using these tools.

Previously, the report submitted by the Department to the Provincial People's Committee contained numerous spelling errors, technical errors, misinterpretations of the subject matter, and fundamental data inaccuracies. These included misspellings, omissions, and distortions of words: "mở dữ liệu" (open data) became "mở dữ li" (open data li), "ươm tạo doanh nghiệp" (incubating businesses) became "xà tạo" (incubating), "cơ hội và thách thức" (opportunities and challenges) became "cơ hội và sơ thức" (opportunities and initial knowledge), "cuộc vận động" (campaign) became "cuộc đua vận động" (campaign race), "27 hội nghị" (27 conferences) became "27 hội viện" (27 parliaments)... These elementary errors appeared with considerable frequency, indicating that the review and verification of documents before signing were not carried out rigorously.

Some technical terms in the report were misused or severely distorted, completely altering the nature of the advisory content. For example, the phrase "strengthening the border belt transportation axis" was changed to "border belt reinforcement"; "building a seedling nursery" to "building similar tree saplings"; "online meeting platform" to "online exhibition platform"; and " educational ecosystem" to "biological attitude education system".

Many aspects of state management are also misinterpreted, hindering understanding and implementation. For example, the statement "detected, verified, and handled 94 cases; initiated legal proceedings in 51 cases involving 91 defendants" was altered to "detected, verified, and handled 94 tasks; initiated 51 services and 91 potential devices," completely distorting the information and posing a significant risk of misunderstanding in leadership and management.

The report also contains errors such as misspelling the names of scientific tasks, programs, and research topics; misspelling the names of historical and cultural relics and landmarks; and even distorting the names of legal documents, issuing agencies, and cited content, rendering the report legally invalid and unusable.

Many opinions suggest that the aforementioned "report" exhibits characteristics of using online tools and AI in document drafting, but without being controlled, reviewed, or evaluated by professional staff. This is a very dangerous phenomenon, because a specialized department, before issuing a consultative report, must go through several "filters," from specialists and functional departments to leaders. However, these "filters" have problems, allowing such a careless and irresponsible report to slip through.

The incident also shows that the user of the online tool, AI, that created the "report" violated Clause 2, Article 4 of the Law on Artificial Intelligence, regarding the fundamental principle in AI operations, which states: "AI serves humanity, does not replace human authority and responsibility. It ensures the maintenance of human control and the ability to intervene in all decisions and actions of the AI ​​system." In this case, it cannot be ruled out that the official completely outsourced the report-making work to the AI, resulting in the aforementioned "product."

Following the incident, the Chairman of the local People's Committee requested the Director of the relevant department to review and clarify the cause, comprehensively rectify the advisory work, and tighten the process of using AI. For everyone, this is also a serious lesson about using digital tools and AI; it is important to remember that AI in many activities can only serve to support humans; it must be checked, controlled, and accountable by humans.

Source: https://baophapluat.vn/bai-hoc-trong-su-dung-ai.html