There’s a joke in one of my favorite movies, “Real Genius,” that seems to apply directly to many of the discussions we’re having about AI today. (It’s an ’80s movie, so it’s not a scene, it’s a montage, set to the music “I’m Falling” by The Comsat Angels.)
In this film, our protagonist Mitch attends a normal math class, but during the editing, most of the class is replaced by tape recorders of various sizes. In the final shot, Mitch enters the classroom to find that a large tape recorder has replaced the teacher himself. It is simply a recording being played by all the other tape recorders.
One of the features announced for Apple Intelligence, Smart Reply, will offer quick ways to respond to direct email inquiries, asking you simple questions (“Do you like me? Check yes or no.”) and composing a response for you.
Apple isn’t the first company to suggest that in the future, your phone will write your emails for you. Gmail’s Smart Compose feature has been doing this for several years, and Apple has been offering its own version of multi-word autocomplete for nearly a year.
But with this latest round of AI announcements, I’ve once again heard a lot of people making jokes about how, pretty soon, your AI will email my AI, and humans will never need to be involved again! This is generally considered nonsense, but I think there may be more to it than that.
Let’s imagine that our AIs end up sending each other endless emails, starting pointless conversations and leading their own inner lives. In that case, it might make for an interesting science fiction story, but I’m not sure it really matters to us as humans. Think of it this way: email is just a means of communication. It was designed for humans to communicate with each other, but for years we’ve been receiving automated emails, newsletters, spam and everything else.
If you’re familiar with technology, you’ve heard of APIs, or application programming interfaces. APIs are, at their most abstract level, an agreed-upon method for software to use or communicate with other software. APIs are in the cloud, on the web, on our devices, everywhere. So why not in our emails, too?
I realize it’s absurd to assume that a free-form email message would be better than a programmed API, but email offers flexibility that other APIs don’t. Emails can be about literally anything. And often, APIs simply aren’t used well because the people who would use them are lazy, busy, uninterested, or don’t know they exist.
Let’s say you need to find a common meeting time for you and four other people. Are there any Internet calendar APIs for this? Yes! Are there any calendar applications that support this type of scheduling? Yes! Are there literally web applications that will do this work for you? Yes! (I use StrawPoll myself.) And yet, I’d bet that most people just email everyone asking if they can set a specific time and try again until they get it right. It’s not efficient, but it’s convenient.
Now imagine the same scenario, but everyone uses an AI system that reads emails and has access to each user’s calendar. The end result might be the same as an existing API or web application, but emails between AIs take care of sorting everything out. Some AIs might know exactly when their person is available; others have to ask. But instead of users interfacing with other systems and setting everything up, AIs handle most of the work and the user simply steps in when needed.
I don’t think this is an absurd scenario. (And yes, if the AIs are particularly smart, they might use an existing calendar service to solve the problem up front.) It’s the equivalent of each of these people having their own human assistant to set up the meeting, except none of them probably have the budget to hire a personal assistant.
In fact, this is where AI assistants really run into trouble. not When they talk to other AIs, but when they talk to humans. Remember when Google introduced its service that pretended to be a human and called real people to verify Google Maps data or make reservations? That’s what I really dread: being bombarded with emails, texts, and phone calls from AIs working for people and organizations that want my attention but aren’t willing to give it to me on their own.
As long as I don’t have to read a pile of emails between AIs, I don’t mind them getting them. The protocol doesn’t really matter (use iMessage or RCS, for all I care), as long as the job gets done and I don’t have to clean up the mess. Don’t interfere except to answer questions or make my own requests.
Email and text messages may be a silly way to build an interconnected network of AI software systems, but history has often shown us that sometimes the simplest solution is the one that’s available, not the most elegant one.
If you enjoy articles like this, please support us by becoming a Six Colors subscriber. Subscribers get access to an exclusive podcast, members-only stories, and a special community.
