This system works very differently, though. In a way, Watson is aiming for a more intellectual goal, a kind of evidence-based cognition. And in Watson's most useful applications, it grinds through data, and spits out possible answers and conclusions for review by humans. Robo Brain doesn't care about creating human-digestible conclusions or advice. It's translating human-speak, basically, into robot action, machine-readable results that tell bots how to physically perform certain tasks.

Yeah, I've made it my mission to try to tamp down the general hysteria, when it comes to coverage of really interesting robotics projects. But I spent a solid hour writing and deleting stupid SF-fueled intros to this story. It feels like a movie—not a very good one—that's on the verge of writing itself. Like all you'd have to do is give it the wrong chunk of data culled from the internet, and it would mobilize a machine army that *only* knows how to commit atrocities.

Force-feeding was a bit of a silly choice, on my part, but something about the process felt similar. It's not like they unleash Robo Brain on the internet, and let it hoover up whatever it pleases (and bully for that, given what's on the internet). They also don't let the machine filter out topics that it doesn't care for. So if we're going to anthropomorphize this system—which, of course we are, since we're a narcissistic species—it seems more like the Gluttony victim from Se7en than a willing participant.

Hopefully it'll become enamored of those hot ladies w/ guns videos, and direct all of its replicants to saunter into battle, hip-firing in their bikinis.

Awesome. It does make me wonder whether the various researchers watch any of the videos in their entirety, before feeding them into Robo Brain. I should have asked the guy I spoke to.

I agree. I think that's a serious, serious concern, or should be. It's also proof (to me) that full autonomy is only going to work when some sort of mandate requires that all cars are robotic. Pretty Draconian, since it means making driving illegal. Until that happens, it'll be region or lane-specific autonomy, if anything.

But I didn't use kill in the headline...

Yeah, pretty hackish of me. Sorry about that.

But those Google cars are extremely coddled. 1) They're always relying on pre-loaded LIDAR data, so always on routes that Google has mapped ahead of time. It's going to take a very long time to establish a fully comprehensive, national LIDAR map, and even when it's completed, it'll have to be updated constantly. 2) They aren't driving in extreme weather conditions. 3) They're driving in a very limited radius—no one's taking them on a cross-country trip, dealing with unfamiliar and poorly marked roads. So that's 700,000 miles in a highly controlled environment. 4) And 700,000 miles is pretty skimpy, isn't it, when it comes to gauging the problems that arise with cars, and with drivers (human or otherwise)? It's not as though everyone gets into a potentially fatal accident every 500,000 miles. But when you combine all vehicles and all those drivers, and consider the total number of cars on the road, that's where the number of total collisions start to get scary. I have no doubt that robotic cars will reduce crashes, and, if they really become the rule, not the exception, will save a crazy amount of lives (and money). Still, it's not like these things are going to be creeping along at 20 mph. At 65 mph, physics can kick your ass real quick.