I wrote a version of this article a while ago, but for reasons I can only partially explain, it vanished. So here’s the rewrite.
People are bad at driving cars.
Statistically, everyone who gets into a car will, at some time in their lives, be in a car crash. Car crashes kill more people than just about every other thing in the world, including guns and bombs. (The woman I’m sitting with now while I write this just told me 95% of all car crashes are caused by user error.)
There’s a great TEDTalk, (I don’t remember which one) where the presenter talked about how unsafe our driving practices are, and I’m not even talking about texting and driving, which is catastrophically selfish and dangerous.
No. The kind of dangerous she was talking about is the safe way of driving. When we want to change lanes, we turn our heads away from the direction we’re driving. When we’re just driving straight, we should be constantly checking our mirrors to make sure someone else isn’t doing something stupid–again, taking our eyes off the road.
FLAWS IN PERCEPTION
I don’t like the term “self-driving car.” It’s long and cumbersome. And, while it’s descriptive, it’s antiquated. One amazing video (again, I can’t find the link (it was a long time ago)) said they should be called Autos. I like that. It’s perfect, and it has some durability in it. When the Auto doesn’t just stay on the road, it’s still automatic.
When I originally wrote this article, and again every few months, Autos were in the news. They usually get brought up when someone dies in one, or when there’s crash, or when someone’s afraid of an accident or death because of one.
As often as not, that last part is a question of financial responsibility: If a self-driving car kills someone, is the owner or the creator responsible?
Another question: if there’s a situation where an accident involved in death is inevitable, should the Auto kill the people outside or inside the car?
Those are valid questions, but they have what I believe to be a deeply flawed underlying assumption, which is really what this essay is about.
When most people think about something, they anthropomorphize it. They project human tendencies and perspectives onto an inhuman object or entity.
This happens A LOT with AI, usually in a way that portrays an us-against-them scenario where AI is hostile to humans. See: Terminator 2; 2001: A Space Odyssey; Ex Machina; The Matrix; etc. The result is a machine with the instincts, fears, and violent reactions that we humans have. This almost certainly won’t be the case, mostly because AI isn’t evolving in the same way we humans probably did.
But, that’s not the flaw I want to address here.
Instead, it’s applying our limited ability to see and understand our environment to Autos. When we think about an Auto, we see the AI piloting it as seeing things like we do, very limited. At best, people imagine that it can see at all directions at once and in all kinds of weather. That perspective completely neglects the connected world in which we live. And no, I’m not talking about malicious hacking, which is another issue for another essay.
The key difference between an AI pilot and a human driver is the AI’s ability to connect and communicate with other sources of information beyond itself.
Think about the navigation app Waze. People love Waze because it routes and reroutes in real-time around traffic. Some crafty people in Beverly Hills hacked Waze by making it think they’re pretty and serene street was filled with traffic, which Waze then routed people around.
When a human driver sits in a car, they see what’s immediately around them. They can’t see miles in every direction, around corners. AI can do that. It can assimilate the location of everything everywhere, all the time. An Integrated Auto would get data not just from it’s own physically connected sensors. Instead, it will be getting real-time GPS data, like Waze, but it will also get satellite image data, local police and fire information, for what might be traffic in the future. It will be getting the location, speed, and intention of every Auto around it.
And (this is where it gets fun) it will get real-time location data from the mobiles of people nearby. Most kids where I live have mobiles. That mobile knows where it is. And it could send that info to the Auto, which will then drive slowly, if, say, the child is crouching behind a car, ready to jump out. The same is true for kids playing, kids at school, etc.
This idea of an Integrated Auto has very far-reaching implications and I’ve only touched the surface of it. Try and see the world not as you see it now, but with input sources that transcend the limits of human perception. Then, recompile and have fun!
P.S. There’s a great TEDTalk on this subject called What a Driverless World Could Look Like.