When I worked in content and operations at bilibili, we often proposed new product requests.
Sometimes the explanation we received was simple: the recommendation system was a “black box.”
That was not an unreasonable answer. The distribution system of a mature platform is complex, and the performance visible to an operations team is only an outcome of that system. But the phrase triggered a strong curiosity in me: how did product and engineering teams understand the problem? Which constraints mattered to them, and how did they translate a business goal into choices inside a system?
If I could understand their language, could I move the work I cared about forward more effectively?
I wanted another way of thinking, not one answer
By then I had already done a great deal of operations work, but I had never thought of myself as “only operations.”
I cared less about how many pieces of content or campaigns I completed than about why the result happened. Why did a user open something? Why did the platform distribute it? Why did a request fail to make the roadmap? Why did a reasonable business goal change shape when it reached product and engineering?
When collaboration stalls, it is easy to conclude that the other side does not understand the business. But if I can describe a problem only in my own language, I may be equally blind to the cost, data, architecture and long-term effects the other side has to consider.
The “black box” did not make me want to dismantle a recommendation algorithm by myself. It made me unwilling to leave everything I did not understand for someone else to interpret.
I wanted to ask better questions. What exactly do we not know? What can the current data prove? Is something technically impossible, simply not worth doing yet, or have we failed to describe the need clearly?
Returning to school was a way to expand my room to act
Around that time, I found a new master’s program in Artificial Intelligence and Digital Media at Hong Kong Baptist University and decided to try.
I prepared my English while continuing to work. I woke at five in the morning to study, then used the time after lunch in a milk-tea shop near the office to continue.
Looking back, the important part was not a score. It was a very concrete resistance to remaining in the same position: I could already see business problems, but I did not want to stay forever at the point of merely submitting requests.
Going back to school did not turn me into an engineer, nor did it give me a universal theory that could explain every algorithm. Systems remained complex, and cross-functional work did not become easy because of a degree.
The change happened in smaller ways.
I became more comfortable breaking a vague goal into inputs, processes and observable results. I learned what data can answer and what it cannot replace. In a technical discussion, I stopped asking only whether something could be built. I also asked about cost, boundaries, iteration and the conditions under which we would know it worked.
When I later began building AI products, the different languages finally met. Users speak in the language of lived problems. Models offer probabilistic capabilities. Product work has to design a reliable, understandable and testable path between them.
Understanding a black box does not remove every unknown
A related degree is not the only way to cross a role boundary.
Some people learn technical language by working closely with engineers over time. Others build understanding through experiments, systematic reading or making prototypes themselves. A degree does not guarantee product judgment, and it cannot replace an understanding of users and the business.
For me, returning to school meant admitting that I did not know, then giving that uncertainty a serious period of time.
I did not turn the recommendation system into a transparent box. Real products, organizations and people will always contain things that cannot be fully explained.
But I no longer stop when someone says, “It is a black box.”
I can keep asking, decomposing and testing. I can also translate among content, business, product and technology. Returning to school did not give me a standard answer. It gave me the ability to enter a problem more deeply.