I know what you're thinking, "yet another one of those".
Well, yes, but no. I really just want to set the record straight. In recent days I read yet another article about how self-driving vehicles may put truck drivers and taxi drivers out of their jobs.
And while a large group of people loosing their livelihood or having their income severely impacted is nothing to sneeze at, and is very sad, I don't think that the real risk lays there. Through all technology advances, we have seen this play out in the same way. Some occupations get impacted, and then the people just shift to doing something that supports the new reality, and most times, the market grows and employs more people. This is how I see the self-driving vehicle evolution play out.
However, there is one aspect to AI that we tend to overlook.
While all the robotic stuff, and steering wheel-less cars can be really sexy, AI is really making strides into a completely different type of occupation: what we call the "knowledge worker". More and more, certain activities can be done by an AI. Image creation, written copy generation, chatbots, etc. Yes, at first the results are lackluster, but stop fooling yourself, they will be, and they are getting better.
You may be thinking, But what's the difference? Isn't a "knowledge robot" just like a driverless car?
At first sight, sure. Both are a substitute for something that was done by a human before. But there is huge difference between a robot or a driverless car, and a chatbot. For making a new robot, or a new driverless car, we need raw materials, put them together, and get them to the place where they would be used. There is a lot of people (and of course, robots!) that take a role in that.
A knowledge based robot though, is a piece of code, and we only have to create a new session. As long as the underlying infrastructure supports the load (and now that that's always possible with elastic compute on the cloud), it is possible to replicate as many of them as we need. An accountant may say that the "marginal cost of each new unit approaches zero".
This is the true risk. We may end up with one company that understands customer support really well, replacing ALL 3 million agents in the US (a comparable workforce than truck drivers) with just a good piece of code, and almost no infrastructure build.
Build it once, use it many times, this is the real impact.