Whatever evils are put upon the world by misuse of machine learning, this makes up for all of it.
The original black artists. Elvis stole from them. And then the machine learning model stole from Sir Mix-a-lot and also from Elvis.
The layers of theft.
Very smart and almost always correct person Bruce Schneier and other person I don’t know, Barath Raghavan, are espousing an idea that I have espoused in the past.
They are saying that if some ML model uses data that you made, you should receive some sort of actual financial dividend or royalty. Or since it’s essentially all of our data that is being used, we should all get paid, just like Alaskans get paid for the oil that is drilled up.
When I made the same argument it was from a communist perspective in regards to labor automation in general.
Imagine your job is replaced by a robot. In our current capitalist society, that robot is considered, well, a capital expense. The employer owns it and reaps all the benefits. Labor just gets fucked, except for the person who has the job of maintaining the robot.
If we seize the means of production, the laborers would “own” the robot. The robot is doing your job, but it’s still your job. You get to chill at home, the robot does your work for you, and the employer still has to pay you for that work. So if we invent self driving trucks, all the truck drivers own those trucks individually and still get paid, without ever having to drive.
At least that’s the principle. The reality is you probably implement it the way they suggest in the article. You put some sort of automation tax on all companies. In effect, you prevent the reduction of labor costs. Companies can fire people to replace with robots, if that’s more efficient. Companies can hire people. But if they are using machines to replace people, they pay a tax on that automation, so they are still effectively paying the labor costs of the people that would have had jobs. Then those labor expenses are split up evenly amongst everyone who isn’t working.
Yeah. Statutory royalties and moral rights proportional to the weight (or something equivalent) of your data from the training data.
However… there is a big business that has been thriving for a long time on data laundering. Basically, use tainted data to get the model you want. Then use another separate system with public data to generate an equivalent model.
Bam. You made your model from clear and free data. No royalties required. No way to prove it short of an insanely complex investigation and likely court orders.
This concept is how the US “leaked” all the info about Russia’s planned annexation invasion of Ukraine without revealing the real sources.
Automation tax?
Or just an actually fair income/revenue tax on anything that generates money. Robot generates money for you? Still paying that tax. Cap profits and cap individual income, and let the impacts of that ripple through the rest of the economy.
I feel like they have an emergency playlist on deck that consists of just Killswitch Engage in case they ever get the call
My favorite term for this is EPIS in CANDU reactors.
In CANDU reactors this is achieved by injecting a neutron poison into the reactor core itself via the EPIS or emergency poison injection system .
Farewell, cab drivers.
Happy to see it.
I think they consider that a 90% “success” rate for all intents and purposes.
Two big stories today.
Hey, some good news for once!
I’m Shocked. Shocked, I say.
