IDK... I suspect we're pointing to the errors and lack of economies in today's technologies (actually yesterday's if not yesteryear's tools are all we're being exposed to) rather than watching the trajectory of improvement. I have been skeptical about every techno-cultural change that has occurred in my lifetime, from e-commerce to streaming to social media to crypto, and have been slow to invest in these areas. In hindsight, my error was to look at how each product was worse than the competing way of doing things, and to think those errors mattered more than other factors.
It's been a few years since I read Clayton Christensen's
The Innovator's Dilemma but I remember the following key points:
- Innovations almost always start out performing worse than the incumbent method, and appeal to tiny niches due to factors that are different than how we measure the performance of the incumbent method (i.e. price vs. convenience, speed vs. quality, different minimum order quantities or form factors)
- Successful innovations begin with worse performance than incumbent methods, but are improving at a faster pace than the incumbent method. Thus the performance of the innovation will eventually intersect with and surpass the incumbent method.
- Sales of the innovation will be minuscule compared to the incumbent method, but grow at a faster pace than the incumbent method.
So generative AI may be useless for most people today, but because each of the above appear to be true all the pieces seem to be in place for a disruption in the next few years. The question is what is possible from a technical standpoint. I.e. will the technology hit a performance ceiling that will arrest its trajectory of improvement? ChatGPT itself demonstrated a breakthrough in 2021, and that was followed by breakthroughs in graphics generation. Prior to these events, most people assumed these were the ceilings that technology would never figure out.
So far, I'm seeing only fast, incremental improvement. The third legs and sixth fingers in AI images appear to be becoming more rare. AI-generated video (full of errors, of course) are becoming common. Text is becoming more creative, with more complex or even artistic sentence structures and coordination. People in my field are talking and giving presentations about how their paid AI subscriptions are speeding up their communication, writing, and analysis functions, and how they are saving time even if they must edit the output or generate a dozen iterations to obtain the output they want.
So there you have it - ball the ingredients: paying niche customers who value something different than what the traditional technique offers, a faster trajectory of improvement than the incumbent methods are managing, and tiny but exponentially growing sales. This is why the big tech companies are investing billions. They don't want to be left in the dust by the next technological disruption, which has become obvious much more quickly than the sorts of disruptions that toppled earlier tech monopolists.
Again, this is all consistent with what Christensen noted about disruptive innovation. It's not about asking whether
today's product can replace a tax accountant or generate error-free output. It's about drawing a line from where we were in circa 2020 to where we are in 2024 and asking where that trajectory leads us in the 2030s. I think it leads to error-free output and AI tax accountants.
