No Analogies for AI: Why Bottoms-Up Reasoning Matters in the Intelligence Age

Every transformative moment in tech is unique. The next success stories won't come from creating another iPhone or social network. Yet most entrepreneurs and investors fall into analogy-based thinking—a habit that serves routine tasks but limits creative endeavors like startups. Using bottoms-up reasoning, on the other hand, will force you to think about the core concept that drive a technology and build from there. This approach feels (and should absolutely feel) uncomfortable, but it's a good sign you're entering the high-risk, high-reward game. Remember, you cannot get big returns without taking big risks. If you're thinking in analogies, you're guaranteeing no innovation, no risk, and by that, no big rewards.

The AI shift has energized startups, but most are disappointingly confined to vertical industry use cases—a symptom of groupthink. In the 'information age,' we built software line by line, making it impossible for companies to scale across diverse tasks due to limited resources. Thus, startups had to find a vertical niche and dominate it. For example, it's impossible to conceive that Intuit could create TurboTax, a PlayStation 5 video game, a video streaming service, and a mobile operating system. The company would collapse under it's own weight. Thus, companies in the information age naturally had to focus in order to succeed.

Now in the 'intelligence age,' many are startups are repeating this pattern, applying AI only within narrow fields. I argue that this is a mistake. AI is fundamentally different. It's not a piece of software written by hand. It's grown, or nurtured, like a plant; with a hint of fine-tuning and a lot of data. It's not like some code where you can put breakpoints in a debugger and analyze the variables. Predicting its behavior or debugging it like traditional code isn't feasible. It's important to understand that we've never had a technology like AI. Unlike deterministic programming of the information age, AI is capable of handling unbounded states without pre-coded instructions. When the CPU was invented, it was impossible to use analogies to create great new businesses on top of it. There wasn't a "CPU" before the CPU. It's hard to understate how fundamentally different AI is from the software that came before it. It's likely all the successes will be so unpredictable that there will be no analogies of those companies in the information age. After all, who would have thought that the most popular app today, ChatGPT, a chat app, isn't even used to chat with other humans?

Silicon chips and the internet introduced true paradigm shifts; however, mobile and cloud computing, while impactful, simply repackaged existing architectures. This is why analogies can often work, when the technology architecture is similar. For example, programming a 1990s desktop app is similar to almost identical to programming a 2024 cloud or mobile app.

If you're building a startup in the intelligence age, ask yourself if it's just a carryover from the information age. With AI capable of handling unbounded inputs and outputs, is it wise to limit it to a narrow domain? Is it smart to use a shotgun to give haircuts?