When AI transforms from an answer machine to a work partner

        Whether it's reorganizing reports, programming, planning trips, or analyzing data, AI has long been integrated into our lives. But after interning at the Information Security Department, I realized that truly using AI is not just about knowing how to ask questions, but also about further leveraging AI to assist in completing tasks. However, the role of humans has not become unimportant as a result, but rather requires more thinking and judgment. Because how much AI can do is one thing, and how much we can make it perform is another matter.

        From learning AI to truly making agents work

                During these two months of internship, in addition to assisting in testing the answer quality of AI tools, I also attempted to apply the problem analysis skills accumulated in class to my work. Faced with dozens or even hundreds of security news articles per week, I began to think about whether it is possible to use AI to identify content that may be relevant to the company's use of products, and then provide noteworthy information to internal colleagues for reference?

        I thought that all I needed to do was' let AI help with it, 'but once it really started, challenges followed one after another. How to reorganize a whole page of news? How should information flow? What tasks should AI be responsible for? How to confirm if there are any omissions or errors in judgment after the results are generated? Some methods are theoretically feasible, but too complex to implement; Some results may seem complete, but after testing, it is discovered that they are not as expected. During the process, I worked and learned while constantly revising my ideas.

        AI can provide suggestions, but the decision-making power still lies with oneself

        When encountering problems, I naturally ask AI, and it will quickly come up with a seemingly complete method. At first, I thought, 'Since AI has already suggested this, it should be the best approach, right?'? But after actual implementation, I realized that using AI is not about handing over the problem, but about having an additional partner to think with.

        When AI makes suggestions, I should continue to ask: 'Why do we need to do this?' 'Is there a simpler way?' 'Is this method suitable for the current needs?' Even if the answer seems to be correct, we cannot simply believe it and need to confirm it again.

        Ask more clearly and make yourself think more clearly

        In fact, the quality of AI's answers is closely related to the information we provide. If you just say to it, 'Help me create a system that can capture news,' it can only guess my needs based on limited information; But if we first explain the background, problem, goal, limitations, and desired results, the proposed method will be closer to the direction we need.

        Therefore, before explaining the problem to AI, I will first ask myself: What exactly do I want to solve? Where is it stuck now? What results are considered successful? Sometimes, just reorganizing these things clearly can already solve half of the problem.

        Empower AI with controllable risks

        In addition, when using AI to improve efficiency, it is also necessary to consider data, permissions, and risks. For example, sensitive information of individuals or companies should be removed from identifiable content first, or simulated data with similar structures should be used before uploading to reduce risks; Moreover, the greater the authority of AI agents, the better. Important operations and final confirmations should still be completed by humans.

        With the promotion of the Basic Law on Artificial Intelligence, I also understand that in addition to pursuing efficiency and innovation, we should also take into account safety, transparency, responsibility, and risk management, so that the use of AI can be guided and truly realize its value.

        In the era of AI, it is more important to 'learn to learn'

        The process of working on this project was not smooth sailing. Some issues were stuck for a long time, and I thought it was almost completed. However, after testing, new problems were discovered. But I gradually learned how to find the next step from failure when a path doesn't work, and these repeated attempts also made me clearer about which tasks can be entrusted to AI, which results need to be judged and verified by myself, and how to make AI a partner who can assist in work.

        Technology is changing rapidly, and today's familiar tools may have new features tomorrow. Perhaps the most valuable thing to learn is not a specific AI tool, but 'how to learn'.

        *Image source: Magnific

        #