Abstract:
The UK government's Department for Science, Innovation and Technology has released a new set of guidance on the use and procurement of artificial intelligence for public sector employees and procurement staff. The core recommendations of the guidelines have aroused heated discussions in the technology community: Before deploying any artificial intelligence tools, the government requires officials to seriously ask themselves whether they really need to use AI; if it is indeed necessary, they should actively choose "the weakest and smallest model that can complete the task" instead of blindly pursuing the most powerful and advanced cutting-edge systems.
This official guidance emphasizes that public institutions must exercise a high degree of restraint in following the trend of adopting cutting-edge AI technologies. The guidelines clearly point out that in many daily administrative processes and data analysis scenarios, traditional deterministic rule software, automated scripts and even standard spreadsheets are fully capable of solving problems efficiently. The forced introduction of generative AI will not only fail to significantly improve efficiency, but will instead introduce unnecessary logical illusions and error risks. Therefore, the first task for decision-makers when establishing a project is to self-examine the real necessity of using AI and avoid technical redundancy.
What is more interesting is the selection strategy of the criterion for model performance gradient. The British government recommends that when AI must be introduced, the procurement and technical teams should follow the principle of "use only what is sufficient" and give priority to lightweight or small-scale special-purpose models with the most streamlined parameters and the lowest computational overhead. It is strictly prohibited to directly call high-specification top-level large language models without justification. Officials explained that this strategy can not only significantly reduce expensive cloud API call costs and computing power expenditures for taxpayers, but also significantly reduce the huge energy consumption and carbon emissions caused by large-scale models in the data training and inference stages, thus complying with the government's overall green environmental protection and public finance prudence goals.
In addition, the guidelines also put forward strict red line requirements for data privacy security and technical autonomy and controllability. Since ultra-large-scale general models often rely on external third-party cloud services and involve massive data throughput, prioritizing the use of locally deployable or lightweight open source models can help prevent sensitive citizen public data from flowing into external servers and reduce the risk of supply chain interruptions and external platform technology lock-in.
Industry analysts pointed out that against the backdrop of today's frantic pursuit of ultra-large-scale parameters and super-intelligent computing power by technology giants, the anti-trend guidelines issued by the British government appear to be particularly pragmatic. Although this prudent and even slightly conservative strategy may slow down the pilot speed of some complex generative AI in public services in the short term, it provides a governance sample with great reference value for governments around the world on how to balance the dividends of AI technology, fiscal budget constraints and digital sovereignty security.
Related articles:
The Governor of the Bank of England issued a warning to the G20 Group: AI may trigger a global economic recession
OpenAI rarely supports the introduction of a mandatory AI regulatory bill in the UK and calls for legislation to be enacted as soon as possible during the "political window"
The British King convened a summit to focus on AI risks. Huang Renxun said that products that are not safe enough should be postponed
Comments