AIHello World
Every engineer’s first post should probably be called this, so I won’t pretend otherwise.
I’ve joined Kondra Systems in Gurgaon, and I’m starting this blog for a fairly simple reason: I want somewhere to think out loud about artificial intelligence while I work out how much of it I actually understand. Which right now is not a great deal.
What I’m Working On
Kondra does two kinds of work and I find the combination interesting.
On one side there’s the enterprise work, helping large clients rethink how they operate digitally. Coca-Cola is one of them, and being in the room for those conversations is an education in how differently a global brand thinks about technology compared to how engineers do.
On the other side there’s product work. The one I’ve been getting to know is PageMage, which thousands of eBay sellers use to make their listings better, including placing relevant ads alongside them. It’s a good problem. A seller has a product, a buyer has an intention, and something in the middle has to decide what’s relevant. Get it right and everyone is better off. Get it wrong and you’ve wasted a page view and annoyed someone.
I keep noticing that “decide what’s relevant” is doing an enormous amount of work in that sentence.
Meeting David Vogt
I got to meet David Vogt, who founded Kondra and runs it, and the conversation has stayed with me longer than I expected.
He gave me a piece of advice that sounded almost too simple: every engineer should read The Pragmatic Programmer, and every engineer should learn to automate. Not automate the interesting parts. Automate the parts you do repeatedly without thinking, because those are the parts where you’re being a poor substitute for a script.
I’ve started the book. The argument I keep turning over is the one about being suspicious of any task you perform by hand more than twice.
Because here’s where it leads, and I don’t think this was David’s point but it’s where I ended up. If the correct response to a repetitive task is to automate it, then the interesting question becomes: what’s the largest task we can teach a machine to do? We’ve automated builds and deployments and tests. What’s the ceiling?
“Automation used to mean writing down the steps. The interesting version is when you can’t write down the steps, and the machine has to work them out.” — Sameer Gupta
That’s the thing I want to follow, and it’s why this blog exists.
The Question I Can’t Put Down
I’ll end with the thing that’s actually been keeping me up.
The DARPA challenges a few years ago had vehicles driving themselves across a desert and then through a mock town. They were slow, they were fragile, and they worked. Robots are moving out of factory cages and into warehouses. And the military applications are further along than most people are comfortable discussing.
Suppose all of that keeps improving. Not next year. Over fifteen or twenty years.
What happens to everyone who drives for a living?
I don’t ask that rhetorically or as an argument against the technology. I ask it because in this country that is an enormous number of people, many of them supporting families on it, in an occupation that requires skill and judgement and years of experience but no formal credential. It has been, for a long time, one of the more reliable ways for a person without an education to earn a decent living.
I’ve read a fair amount of confident writing about how technology creates more jobs than it destroys. Historically that seems broadly true. It also seems like a claim that’s much easier to make in aggregate, over decades, than it is to explain to a specific person in a specific decade.
I don’t have an answer. I’m not sure anyone does yet. But it seems like the sort of question worth asking early, while it’s still theoretical, rather than late, when it isn’t.
What This Blog Will Be
Mostly notes. I’m going to read papers I’m not fully qualified to read and write down what I think they mean, and I expect to get things wrong in public and correct them later.
If you know this material better than I do, which is a low bar at the moment, please tell me when I’ve made a mistake.
Let’s see where this goes.