Where Do AI Products Go Next?
It’s the early stages of the AI platform shift. Today’s AI applications are remarkable. Yet, if we are indeed in the early innings, many will seem quaint in 5-10 years’ time. I have a hard time wrapping my head around this idea. How could it be so?
The first wave of AI applications have flowed from the classic inspiration paradigm of software products: solving (human) pain points around tedious, manual tasks. Knowledge workers have pain points around customer support, targeting and writing sales emails, and creating entertaining videos, to name a few. AI software startups like Sierra, Clay, and HeyGen have seen surging growth by building AI applications to address these problems, respectively.
These startups offer a dramatic leap in innovation compared to SaaS startups of just a few years ago. That said, these startups are typically solving pain points that stem from (human) processes created decades before the advent of modern AI software. In 5-10 years, they will probably seem skeuomorphic — that is, like the early 90s version of nytimes.com which consisted in the newspaper plastered on a primitive webpage.
Perhaps there’s an analogy to the first AI chess programs, where a human working with the AI program could defeat a pure AI program. Eventually, the AI-only program improved enough to easily beat the best AI + human tandem. Humans ended up slowing down the AI. Are today’s AI applications, designed to work with humans and their processes, like in those early chess AI programs?
A few years ago, a professional Starcraft player described a new AI algorithm that was regularly beating top players as employing “gameplay that’s unimaginably unusual.” I suspect that the most successful AI applications in 5-10 years will be performing tasks that seem, by today’s standards, unimaginably unusual.
I think the good news for today’s AI startups is that they are well-positioned to build AI products that can do the unimaginably unusual. The bad news is that as AI’s capabilities rapidly improve in the next few years, so will the expectations of customers. To consolidate enduring market leadership positions over the course of the next decade, AI startups will have to pursue lines of innovation that are literally inconceivable to humans today.
The rise of AI has ushered the tech industry into a stage of AI & human symbiosis, like the early period of AI chess algorithms. At some point in the next decade or two, as the AI platform shift approaches maturity, it seems probable that AI’s development capabilities will outgrow their human sidekicks.
What will the tech industry look like at that point? I think it will be much larger, capable of producing remarkable levels of innovation. How should today’s startups try to navigate the path from here to there? I would focus on building product, using AI tooling wherever possible, and compounding value creation for users in tandem with AI’s advances. At a certain point, not too far away, it will be more sensible to ask an AI these questions than a human.