The future of AI in verticals?

To be honest, looking at those claimed "omnipotent" general-purpose AI agents in the vertical field repeatedly hit the wall, but I have more confidence in the development of vertical AI. Recently in the medical field OpenEvidence is a perfect example - it is not like some general-purpose AI rhetoric, but honestly plowing into the vertical track of medicine, and as a result, but to make a truly recognized by the doctors of the intelligent tools. This reminds me of the sentiment of a chief physician of a tertiary hospital when attending a medical AI summit last year, "What we need is an AI that can read CT films, not an AI that can write poetry."

Vertical AI breakthroughs

Why is vertical AI more likely to succeed? The key lies in the word "focus". Financial risk control, for example, last year, a bank deployed an intelligent risk control system, through the deep learning of millions of transaction data, and now can identify 98% of fraudulent transactions in real time - this figure is more than one order of magnitude higher than any general-purpose AI. In the end, the business logic of a vertical is like a precision clock, where every gear has to fit tightly, and how can a general-purpose AI not go wrong when it tries to hit all the parts with a single hammer?

Data barriers are the real moat.

I remember once chatting with a friend who does legal AI, he said that it took them more than a year to organize the case data from the adjudication website into a usable knowledge graph, but it was this seemingly boring process that allowed their intelligent legal assistant to accurately cite specific laws and jurisprudence. On the contrary, some general AIs, although articulate, start babbling when they encounter specialized issues. This can't help but make people think: in the age of AI, perhaps data depth is more important than algorithmic breadth?

Vertical AI's Landing Challenges

Of course, the development of vertical AI is not always smooth. Last year, a manufacturing AI project encountered a typical problem - it is difficult to convert the operational experience of factory teachers into data. This also reminds us that AI is not a panacea, especially in those areas that rely on tacit knowledge, human-machine collaboration may be the best solution. That said, isn't this exactly where AI should be positioned? To be the human's right-hand man, not a replacement.

Looking at more and more vertical areas of AI began to emerge, I think this wave of AI has really found the right direction. After all, rather than being a "universal doctor" who knows everything but is loose in everything, it is better to become an "expert consultant" in a certain field. What do you think?