What Are the Conditions for AI to Become a 'Colleague'?

A survey paper jointly published by Tencent and multiple Chinese universities raises important questions about the nature of AI systems. While current AI excels at answering questions, it still lacks sufficient capability to complete work tasks from start to finish. The researchers outline in this paper what is necessary for AI to truly become a 'digital colleague.'

The chatbot-type AI currently in widespread use operates on a 'question-and-answer' mechanism, where it returns answers when users pose questions. While this is convenient, once a conversation ends, the memory is reset, and the next conversation cannot maintain the previous context. Researchers point out that this mechanism alone cannot adequately handle the kind of continuous task execution that occurs when humans cooperate with colleagues in the workplace.

The key points of the solution presented in the paper can be summarized into two main aspects. The first is a 'persistent workspace'—establishing an environment where the state of work and progress can be maintained and continued. The second is 'reusable skills,' which means being able to apply procedures or knowledge learned once to other tasks. By combining these two elements, AI can transform from a mere response device into an entity capable of autonomously completing complex work, as this perspective suggests.

The heightened attention to this discussion stems from growing interest in what is called 'autonomous AI agents.' An AI agent refers to AI that performs multi-step tasks while making independent decisions toward a given goal, and many companies and research institutions have recently been advancing their practical implementation. However, in the current state, there are many cases where problems arise when errors occur during work or situations change, and ensuring reliability has been identified as a key challenge.

The reason researchers emphasize a 'persistent working environment' can be understood as directly linked to this reliability issue. When humans work, it is natural to continue on the next day by carrying over the work from the previous day. Only when AI is equipped with similar continuity does it approach a level where we can confidently entrust it with actual work. This logic is persuasive when viewed from the perspective of practical workplace experience.

The concept of 'reusable skills' also holds important practical significance. If we can apply procedures learned in the past rather than learning from scratch each time new work arises, there is potential to substantially reduce AI implementation costs and adaptation time. This is positioned as a factor that will lower barriers to integrating AI into the workplace.

This paper is a survey study that systematically organizes the current state and challenges of AI; it does not immediately result in new products or services. However, the fact that it explicitly articulates in research terms the direction from chatbot to full-fledged work partner suggests that it may influence the design philosophy of future AI agent development. The question 'How much work can we trust AI with?' is simultaneously a technological issue and a theme concerning our very way of working, and it continues to merit attention.