What Happens When Everything Speaks
In a 1970 lecture, the anthropologist Gregory Bateson famously explained a bit, the unit of information, as “a difference that makes a difference”—an admirable play on words.
The first “difference” is the difference between on and off, 1 and zero, or any other binary relationship. The beauty of bits, and why they became the fundamental unit of information, is that they can work with whatever type of information a computer is processing. A bit can represent whether a stick of gum is sugar-free or not as well as whether a patient is alive or not. Because a bit means nothing in itself, it can carry anything. That’s what enabled computers to become general-purpose machines.
Bateson’s second use of difference—“makes a difference”—was a brilliant choice because of the tension between the computer-based meaning of the phrase and its resonance with “make a difference” as something meaningful you should do with your life. In that second sense, “difference” is all about meaning and values: The maximally reductive bits result in a computer process that makes a difference to us, something with meaning and value, such as flashing a warning light at us or beating us at chess.
Reduce and Conquer
The lesson is that if we reduce the world to the most primitive of primitives, we get a machine that can do more than any other machine has ever been able to: reduce and conquer.
Now, we have large language models (LLMs) that are the most general-purpose computer programs ever. We can ask them to explain a technical paper in virtually any field, instruct us on how to care for a baby pigeon, or write a limerick as a Norwegian goat herder praising our good looks. Underneath, they are, of course, yet another careful collation of bits. But the bits create a structure that expresses the relationship among words, our most common and flexible carriers of meaning. There’s much to say about this, but let’s confine ourselves to one outcome that I think is quite likely and quite significant to how we think about ourselves and our world:
LLMs could become the user interface to most of our interactions, not just with computers, but with the everyday tools and utilities we choose to link to one.
The least of this would be that we’d no longer need to speak the language of each of our tools, a minor blessing when you don’t know the difference between your dishwasher’s “Dry,” “Ready Dry,” “Crystal Dry,” and “Polish Dry” cycles. You’ll instead say to it, through the ubiquitous LLM interface, “We’re having guests, so I’d like the glasses to be super clean, but I’ll let it all air dry to cut down on energy usage.” But language is a two-way street, so prepare for your dishwasher to respond, “No can do. If you want shiny glasses, I need to dry them on high heat. It’s up to you.” Score one for the dishwasher.