Most deep neural networks are trained by stochastic gradient descent. Now “stochastic” is a fancy Greek word for “random”; it means that the training data are fed into the model in random order.
So what happens if the bad guys can cause the order to be not random? You guessed it – all bets are off. Suppose for example a company or a country wanted to have a credit-scoring system that’s secretly sexist, but still be able to pretend that its training was actually fair. Well, they could assemble a set of financial data that was representative of the whole population, but start the model’s training on ten rich men and ten poor women drawn from that set – then let initialisation bias do the rest of the work.
Anderson concludes “It’s time for the machine-learning community to carefully examine their assumptions about randomness.”
I think that’s tangent to the real lesson, which is this:
“This rugged, sustainable platform will operate in permissive environments and austere conditions around the world to safeguard our Special Operations Forces on the ground,” USSOCOM Commander Gen. Richard Clarke said in an emailed statement.
Just for context, I’m a lifelong Michigander (incidentally descended from Jews who fled Ukraine back in the 1920s). Cropdusters are still fairly common here in Michigan (and the terrain, I’m told, is quite Ukraine-ish).
My wife grew up in blueberry country, and there are still plenty of fields around her folks’ place, and folks spray them from cropdusters. It is absurd and terrifying how nimble these planes are. Average Joe flyers regularly bring them in below the tree line to dust a field, and then pop back out to dive into the adjacent field. In other words, they can come in over the horizon too low to detect–likely too low to see–and be on top of you before you can bring a turret around.
Given the performance we saw of Russian tanks early in the war in Ukraine, planes like these would fucking mow them down like a goddamned scythe.
“Those checks, known in the industry as rolling retests, occur at random. They require the driver to lift a hand off the wheel, pick up the device and blow — hard — into its mouthpiece for several seconds.”
So we take someone who’s a demonstrably shitty driver and purposely distract them behind the wheel‽ o_O How could that turn out poorly?
But more importantly: is this solution the best approach? We love technological solutions to social problems, because then we get to avoid conflict (“It isn’t me, man; the machine says you must be punished.” 🤷♂️ )
“The legislatures themselves have made this instruments God. Any time you put all your faith in technology, there’s a chance you’re gonna get burned.”
John Fusco, former president of the breath-test manufacturer National Patent
Important to remember that cops aren’t scientists, and lawyers aren’t scientists, and judges aren’t scientists, and juries are rarely full of scientists. Putting a “scientific tool” in their hands doesn’t make them scientists, and doesn’t guarantee precise, accurate, or “scientific” results. Even the best tool in fallible hands will fail (at least occasionally).
These are fun on your computer, and absolutely immersively astounding on your phone/tablet. The future is here, but unevenly distributed—with some portions dune-buggying around Mars, picking at rocks and wrecking up the joint.
I sorta love things like this, not because it’s the “sound of a Martian sunrise”—because it isn’t. It’s a composition humans made, using an express (and consciously expressed) scheme that’s inspired by a Martian sunrise.
No, I love this art because it sounds pretty and pleases and soothes me, and I love projects like this because artists always and forever operate based on formulae—they just usually aren’t able (or willing) to consciously and explicitly formulate those formula. I like it when we engage with our formulae outright.
Also, I really like Mars. Our relationship with that planet has changed substantially since I was a boy, and that always fills my heart with Hope.
The Imagination Station in Toledo (where I was helping folks find the Good Noise™ all last December) is hosting their very own Mini Maker Faire this September. Great folks down there, and a great location along the river. I’ll be there all day with the Loud Lab (amplified Slinkies, simple DIY synths, electric diddley bows. and more)—so mark your calendar. And, if you’re a maker sorta a person, consider applying and showing off what you do (the application deadline is fast approaching).