#future present

Let's say, to make one up, the plan is to judge how threatening people are by their appearance.

Can a dog do this?

Why are we asking about dogs?  That wouldn’t be cost-effective at all, right?  Well, dogs can kind of do this, but they tend to get it wrong pretty often, and are often piggybacking on human masters’ reactions to visitors even when they do get it right.

So our plan is to go from a dog, that can’t reliably do this, to a machine learning algorithm that is, what, 10,000 times less complex than a dog?  Probably even less than that?

So why should be surprised if our machine learning algorithm does something racist like assume [skin color = threatening]?  It could easily have picked ten or twenty things that are just as dumb!  It could easily have classified the pictures by whether the subject is wearing glasses, or whether it’s dark outside, or any number of weak correlates present in the training data, mish-mashed together.  

We haven’t thrown even remotely enough computational power at the problem.

Triton’s discovery raises questions about how the hackers were able to get into these critical systems. It also comes at a time when industrial facilities are embedding connectivity in all kinds of equipment—a phenomenon known as the industrial internet of things. This connectivity lets workers remotely monitor equipment and rapidly gather data so they can make operations more efficient, but it also gives hackers more potential targets.

It’s almost certainly no coincidence that the malware appeared just as hackers from countries like Russia, Iran, and North Korea stepped up their probing of “critical infrastructure” sectors vital to the smooth running of modern economies, such as oil and gas companies, electrical utilities, and transport networks.