Perceived vs Research AI
I was part of a AI panel discussion in Taipei over the weekend, hosted by DT42, where Jaan alluded to the differences between applied AI and pure AI. His point was to place both states on a spectrum; 100% applied AI is where the user massively finds value but has no technological innovation, and 100% pure AI is where there’s true technological innovation, but likely there is no end user or the user finds no benefit.
I’ll coin this spectrum somewhat differently; perceived AI and research AI are the two end-points. Perceived AI takes the place for applied, and research AI takes the place of pure AI.
Very few startups now can start from the research AI endpoint. The window to raise capital to sustain the long-term horizon needed to do pure research is likely over, or close to over. The difficulty in marshalling strong technical resources is also high, especially in developed AI ecosystems like Silicon Valley, Cambridge, and Beijing. In more developing AI ecosystems, however, there may be an opportunity for marshalling those said resources.
Startups should generally start from the perceived AI endpoint. More specifically, these startups should specifically ask — what user benefit am I solving, with the benefit of AI? That user benefit should be significantly higher with a specific AI solution. For example, one metric would be that startup XYZ is solving recommendations for e-commerce companies, and their solution helps companies increase 5.4x an average basket value.
Once startups establish a foothold in the perceived AI endpoint, they can move into the research AI endpoint. This move increases a startup’s value as they are able to extrapolate more insights from the data. Going from research AI to perceived AI is a very rare, and not likely, successful move.