AI has a strategy.
And we can learn a lot from how it is going to win.
Note: this post was originally written a few days before ChatGPT came out. The new capabilities of that system drive home the points in the post even more.
Useful AI has always been around the corner, and yet it never arrives. Artificially intelligent self-driving cars did not make up 50% of all Lyft rides in 2021. AI analytics did not generate $1.7 trillion dollars of value in 2020. Most warehouse workers are not being replaced by robots powered by AI in the next two years1.
Optimistic predictions were dumb... until they weren't. AI has a strategy - or rather people who developed AI have stumbled into a set of strategies - that are going to make AI ubiquitous. But I think these strategies are only apparent if you look through the lens of research on innovation and entrepreneurship. So, that's what I want to do in this post: make a positive case for why AI, and specifically creative AI, is going to win, and what those lessons can teach us about business strategy overall.
A hopeful AI market was born
Predictions of its growth were quite strong
But the hype just faded away
As the promises went astray
What a shame that the growth was so wrong
- Limerick generated by GPT-3 AI
Reason 1: Understanding MVPs and Customer Value
AI only sort-of-works in most applications. Mostly, it seems to steer cars fine, but sometimes it rams into another vehicle. Mostly, it draws great pictures, but sometimes the images are nonsense. A car that occasionally gets into accidents is intolerable. An AI artist that draws some great pictures, but also some bad ones, is perfectly acceptable. Yet AI has mostly been aimed at problems where failure is expensive, and not the ones where failure is cheap and already accepted. This made selling products that only sort-of-work very difficult. Creative AI now focuses on use cases where many bad outcomes are fine (you can easily eliminate a bad AI essay or drawing) if you get a good one sometimes. That is an excellent market for a sort-of-works miracle product.
So the first key to why AI is better positioned to win is that it has found a niche that tolerates its particular foibles. Now it can take advantage of a key strategy that entrepreneurs use to grow fast: starting with a Minimal Viable Product (MVP) that serves a burning customer need. Building the easiest and cheapest-to-make product that could attract real customer interest allows founders to test ideas at low cost. If the tests show they are correct about what customers want, they can raise money and grow their company. But if not, they pivot and change direction, modifying the product they are selling, or the market they are approaching, based on the feedback from their MVPs.
Creative AI, based on open technologies (another key to success), makes it easy for many entrepreneurs to quickly make many different MVPs for different audiences, because experimentation in creativity is cheap & valuable and the risks are low. Over the last year, AI startups have launched that help you write ad copy, help you code software, help you come up with product names, or taglines, or PR releases, and so much more. By moving to creative areas where failure is tolerated, there is an opportunity for much more learning, competition, and growth.
Entrepreneurs their vision often share
To understand what’s needed in the air
To build a product that will show and care
To build it fast, MVP is their prayer
A prototype to test, to see what’s right
To show the world it’s marketable and bright
But often the key is to keep it light
Removing features so they can take flight
The MVP is the way to start
To prove it works, to show the art
The feedback they get, they can’t discard
To ensure it’s ready before they depart
The MVP helps entrepreneurs reach
The goals they set, their vision to teach
It’s the perfect way to measure and preach
The success of their product, to beseech
-GPT-3 writing a sonnet on how entrepreneurs use MVPs. It manages to be both amazing and also cringey. But give it another few months, because…
Reason 2: Classic Disruption
The second reason why creative AI will win is that it is riding the S-curve. The classic S-curve shows us how technologies develop, and it helps explain a lot about how and when certain technologies win. Technologies typically begin with an "Era of Ferment" where everyone knows a new tech is coming, but companies experiment with possible forms and approaches. Eventually, a winning technology emerges - we call this the “dominant design” - which defines the product. Think about how phones went through wild designs in the early 2000s, before the iPhone established the dominant design and all phones began to look the same.
Once the dominant design emerges, a technology takes off. Everyone focuses on the same type of product, and new advances come very quickly. Prices drop and capabilities increase every year. But eventually, the technology begins to mature and stagnate, either because it is too expensive to develop or inertia has set in.
At that point, the old technology become vulnerable. New technologies that begin as much less capable can go through their own take-off period, and rise in capability rapidly. This is what Clay Christensen called a disruptive technology. Something that seems low capability at first, but then rapidly rises in performance, eventually replacing complacent leaders. There is a lot of controversy in academia over how often this really happens, but creative AI seems to be a classic example of a plausible case for disruption from below. Older technologies, basically writing or art done entirely by humans, are improving (think about digital art tools or spell-checkers for writing) but not as fast as AI is improving.
This is in part because AI is advancing much, much faster than most people suspect. AI models are increasing in size and capability by an order of magnitude a year, much faster than Moore’s Law pacing.
At this speed, AI will increase in capability more quickly than anyone will have time to react to. You may laugh at something an AI generates today, but it will be significantly improved in a month. Humans don’t improve that quickly. Combine rapid experimentation (from Reason 1) with rapid development, and you have a recipe for real disruption.
Reason 3: Crossing the Chasm
Geoffrey Moore famously described a chasm between the earliest adopters of a technology and the mass market. Early adopters buy a technology to gain some radical advantage, later adopters want an easy-to-use clearly defined product that serves a clear purpose. Your early adopters will want to maintain their advantage by pushing you to make the product ever more capable and complicated, but that is not a way to cross over to the mass market. Instead, to cross the chasm, businesses often need to make their products easier to use, at some cost of their specialness and connection to early customers.
A prime example is how once internet-only brands, like Warby Parker, have opened up stores, changing their entire business model. They do this to cross the chasm - the mass market doesn’t buy glasses on the internet, they don’t want quirky styles, and they want a full service solution. Crossing the chasm requires a different approach.
Many of the AI companies seem to understand this, and the competitive pressure of being in an Era of Ferment (see Reason 2) is pushing them to consider customers from the start. As a result, many key players are purposefully making their products easier to use, at the expense of upsetting their early customers. For example, the new version 4 of MidJourney, the drawing AI, moves away from the elaborate prompts that expert users loved, and to simple “it just works” style prompting.
This will help AI achieve a mass market much more quickly, as they are aiming for non-sophisticated users early in the process. Combined with the fact that there are many competitors all learning through MVPs, and the rapid increase in capability, the stage is set for AI to win.
You need to try it to believe
I think these three strategies mean that creative AI is coming faster than most people expect because it has solved key business problems that stop adoption, not just the technical ones that have always plagued AI. I want to write more about those implications in the future, but for now, I want to ask you to try AI for yourself and see if you believe me.
Trying MidJourney for images is free. Setup is easy and anyone can get access here: https://midjourney.com. You use it by communicating with a bot, on Discord. Here is a guide to Discord. Then all you need to do is type /imagine and whatever you want. To get good results, you will need to use the new version, v4. To use v4, simple add “ --v 4” to the end of your prompt (yes, you need to include two dashes before the v, a space after the v, and then the number 4). You can write about what you experienced in the comments!
Trying AI generated text is even easier. Go to https://beta.openai.com/playground, it is free to create an account and play with the system. Set the model to da-vinci-03 (which should be the default), and try asking the system for anything. I think you will be impressed. Feel free to share your results in the comments.
Update: You should really try ChatGPT instead: https://chat.openai.com/chat
I really believe that we are about to see major changes in how many people work as a result of creative AI. And I don’t think we are all ready for what that means.