ChatGPT Cockpit engineering

My wife and I were watching YouTubeTV and reminiscing about the "good ol' days" when we watched TV through a cable box. The cable box software left a lot to be desired, but there were a handful of niceties that we've lost since then. For example, we could memorize our favorite channel numbers and dial them in quickly to see what's on, rather than manually scrolling up and down on the TV remote until we find the channel we're looking for. A habitual TV watcher could memorize and recall more channels than can fit on a screen, and punching the number into a remote is faster than having to move a cursor around a screen to the channel you want.

We then began dreaming about all the ways we could tweak the TV guide screen. These ideas couldn't realistically be implemented as settings in the YouTubeTV app, as they were too granular and personal. Furthermore, even if they could be implemented, trying to tweak those settings on an Apple TV using the Siri remote would be a pain in the ass.

However, these weren't technical problems, they were product and marketing problems. It's not that the setting couldn't exist; it's that the setting would be too difficult to manipulate with the tools at our disposal.

That's when it dawned on me: an AI language model could be trained to control the settings for us. Imagine being able to say "Show me every show's rotten tomatoes score in the TV guide" or "Change it to something I can have on in the background" or "Show me what's on in prime-time on the major networks over the course of the week". With the right API, an AI language model could handle all these requests, eliminating the need for complicated settings menus and allowing for a more natural and intuitive way to interact with your TV.

I call this "AI Cockpit engineering". Your app describes a structure for UI configuration, and teaches that structure to the AI (you could think of this as a config file). The AI is capable of reasoning about the user's natural language input, forming a solution, and outputting solution in your structured format. Your app accepts the configuration, and updates the UI accordingly. The config structure could be as complex as you need it to be, as long as that complexity can be explained to the AI language model.

This is what I believe the next generation of AI-powered applications is going to look like. Squishy, flexible UI components with lots of configuration points, and an AI brain making decisions about how those UI components should appear based on the individual user's tastes.

We already have "recommended" content based on your consumption habits, but these recommendations are usually 1-dimensional, and the space to present that recommendation is static. It could be a recommended TV show in your TV dashboard, a recommended app in the app store, or products "people also bought" on a product listing. An AI language model could learn what characteristics are most relevant to you, and tailor a custom experience for you.