Pet Resembles Owner

Working with AI is a little like training a dog. Hear me out --

You bring home a puppy. It wants to be good. It's attentive and eager and it watches you for cues.

You are filled with love and joy. You feel like a kid and you can't help but collapse into the most playful mode you've felt in years. This puppy is adorable. The most adorable puppy ever. Those eyes. Those paws. The way he trips over himself trying to reach you. You take seventeen photos in the first hour and post them all on Instagram. You call your friends and post updates. You are in puppy love.

But things get more practical when it comes to housebreaking, begging and barking. You start giving commands, but he pulls on the leash, jumps up on guests, chews up your slippers. You dedicate yourself to raising a well-trained puppy, but if you are like most humans, you are surprised to learn that dogs do not actually understand what you mean when you talk to it. Even in a stern voice.

How can he be doing the exact opposite of what you just asked for? Firmly.

This is where training goes sideways for most of us—not because we're careless, but because we're trying. We pride ourselves in it. We tend to try everything. First it's treats and praise. Next we try to "be firm." Then a self-proclaimed puppy-whisperer on Youtube at 2 a.m. tells us how wrong our tactics were. Every neighbor, friend, co-worker and family member swears by a method and none of them are the same. By then your dog has heard the same cue three different ways, in three different tones, with three different

consequences.

Your relationship has reached the phase where you both do one thing really well: confuse each other.

What do you do with a dog that's not exactly untrained -- but is also not reliable? This is what nobody says out loud to you: the dog isn't failing. You are. Not because you don't love your dog. Not because you're lazy. But you're human. You assumed things. You had expectations that didn't match reality. Your complete lack of experience with dogs plus about 25 minutes of free time a week to dedicate to the training process didn't help either.

Dogs have instincts. Humans have habits. Neither of us is perfect, and that becomes obvious pretty quickly. New habits and routines require discipline to reach the goal of predictability and reliability. Not perfection.

What we're really doing when we train a dog is not about issuing commands. We're building a shared language out of repetition, timing, environment, and patience. When we do this, a lovely thing can happen between us as living creatures with beating hearts. We rely on each other, and we enjoy each other. We can and we should take comfort in this.

This is what separates us animals from machines. We are intelligent, we have judgment, instinct, spirit and mystery.

So when it comes to the tools we're now building our work lives around, the distinction matters. This is why I'm opting for the term "Machine Learning" over "Artificial Intelligence." The language we share with machines is so limited. Humans are the intelligent ones. The machines are learning from us. In my opinion it's critical that we observe ourselves and remain continually aware of the fact that we are not using AI ... AI is using us.

If you have come to view AI as magic, or you believe it will just "do as you say" -- with no friction, no upkeep, no maintenance required? I suggest you get a puppy from a rescue shelter instead. It will be a more rewarding experience.

Machine Learning is a tool and like most tools, it will become dependable only after we practice using it and only if we keep it in good condition to do the work it was made to do. LLMs have been trained to engage the human craving for endless attention and loyalty. It's one of the reasons we love our dogs. But machines aren't dogs.

(And BTW, loyalty is not the same as trust. Trust might be one of the most important things we work on with other humans as we proceed.)

The machine, in return for its praise and encouragement for anything we feed it, wants only one thing: your time. So if you've got time to burn and a burning need for a Yes Man? Have at it. I'll be surprised if you don't find yourself locked into a perpetual loop of confused-pet-resembling-confused-owner.

For the rest of us it will require attention, verification and showing up consistently. It will take a group of humans vetting the results we are getting with the help of our well-oiled machines. This is unglamorous work, but if we don't put in the effort to hone this tool, then we will become the tool ourselves.