The government will release a mid- to long-term improvement plan as early as next month to overhaul the existing system for calculating payments for AI and software (SW) projects. A key focus is to reflect new types of projects such as generative AI and AI agents, as well as changes in software development methods driven by the spread of AI development tools.

The Ministry of Science and ICT said on Wednesday that the government is preparing a roadmap that sets out the mid- to long-term direction for improving the AI and SW project payment calculation system, with the goal of releasing it in September to October.

Under the Korea Association of AI and Software Industry (KOSA) guidelines for calculating payments for SW projects, a payment calculation standard is in place that classifies AI adoption projects as a separate category. It splits AI service adoption project costs into service fees, customisation work costs, and implementation and development costs. It also reflects AI-specific tasks such as data collection, cleansing and processing, model optimisation and fine-tuning, and verification and stabilisation.

The government believes the need to further supplement the existing standards has grown as new types of projects such as generative AI and AI agents emerge and AI development tools spread.

To that end, the ministry launched an "AI and software project payment system improvement task force" in May with KOSA and experts from industry, academia and research institutes.

The task force is discussing improvement directions based on the current payment calculation system and AI and SW project cases. It is broadly reviewing differences from existing KOSA guidelines, additional project types to be reflected and methods for calculating payments.

It is also reviewing overseas cases. The government said it is difficult to apply one country’s standards 그대로 to the domestic market because public SW ordering practices and contract and payment calculation systems differ by country.

A ministry official said details such as how it will differ from existing KOSA guidelines are currently under discussion. The official added that it is reviewing improvement directions suited to the domestic environment while also looking at cases in major countries, given differences in SW ordering practices and contract and payment systems.

The industry believes it is also necessary to consider that, in AI projects, it is difficult at the contract stage to accurately gauge the scale of work required. In many cases, it only becomes clear how much data cleansing, additional training and fine-tuning are needed after running a model with actual data.

An AI and SW industry official said AI projects often require opening actual data and applying a model to confirm the amount of work needed. The official added that criteria need to be made more specific on how to reflect, at appropriate compensation levels, the various tasks that arise in the course of projects, such as data cleansing, retraining and fine-tuning.