notes · thoughts ·
Is it data after all?
As AI makes intelligence more widely available, a service’s advantage may come from the quality of the data its model uses and the systems that sustain it.
What impressed me recently while using a health management app was less its AI analysis or recommendations than the accuracy and convenience of its food search. Whatever food I searched for, it found the result I wanted quite well. Behind that was an extensive food database and a process for cleaning and verifying it.
Thinking about it, however intelligent AI becomes, it cannot know every fact about the real world accurately. Whether a particular product’s nutritional information has changed, whether a unit in the data is wrong, or which of several sources is correct ultimately requires external evidence and verification.
Of course, AI can automate a great deal of data collection, cleaning, and error detection. But being able to automate a process and being able to trust its results are different things.
If AI model performance increasingly converges at a high level, a service’s advantage may come less from the model itself than from the quality of the data it uses and the systems that continually maintain that quality.
As AI makes intelligence more widely available, it seems increasingly important how accurately we structure the real-world information that intelligence accesses, and how we keep it trustworthy.
In an age when AI can do more and more, could deciding what it should trust when answering become, paradoxically, more important?