This is how democracy dies
unherd.com
unherd.com
This is just sad. I like some people that are very successful on streaming and video services, but it is an extremely high risk venture. It was the privilege of the pioneers that could basically be themselves. Today the demands of users are propbably completely different.
How you even address an article with such wide ranging concerns from "smart fridges will tell me to eat less meat for better insurance rates"(invasive / creepy tech) to "driverless car fatalities will have no accountability"(intersection of tech / law) to "our inability to disagree with The Computer will lead to a global thermonuclear war" (intersection of tech / politics)
People will (hopefully) not buy the creepy fridge. At the very least, there's no fridge monopoly so there should always be a noncreepy option on the market for you personally. If there is a fridge monopoly, human lawmakers and judges have the power to break up the monopoly.
There will be a slew of laws/lawsuits about driverless car fatalities when it first becomes a big thing. How that plays out will be entirely determined by humans (lawmakers / courtrooms).
The idea of having computer simulations so good that it replaces human decision making at the executive level is ridiculously far from being reality, it's pointless to worry about it.
It's already reality, see for example video recommendation on youtube (and it has far-reaching consequences like promotion of extreme political views and conspiracy theories to the point of influencing election and interfering with vaccination).
Nobody in google decided "let's make people believe vaccines have 5g chips in them".
The idea that an AI is going to decide we should first-strike another nation and top brass say "well the AI has never been wrong so let's do it" is laughable.
Real life equivalent: "AI changed the margin on humanitarian aid in country X by 28%", the end result would be war as well but the decision wouldn't be "war or peace" it would be "a little more cost-savings vs a little higher risk of problems".
And it's not sci-fi, see 2008 crisis and its consequences. Bad model that gives short-term profits is almost always preferred over good model that reduces profits.
Or, to put it in Gulag/private prisons terms: "on the advice of AI we decreased calory intake of our prisoners by 30%" sounds much better than "kill prisoners by overworking and strvation".
The executive decision made by a human "we should optimize prisoner calorie intake" is separate from the AI advice "30% is the amount of calorie reduction that prisoners can withstand"
Right now AIs are good for optimizing and calculating things that we specify they should optimize and calculate. If we develop an AI that can independently decide to optimize human calorie intake without anyone doing work to make it an option, then we should be worried. Right now that is pure fiction.
Pre-AI equivalent: These prisoners seem like they're eating too much, so let's cut their food portions. One third reduction seems reasonable.
The gulag example is a good one because with a different limit it would change from system of extermination into a civilized prison and it doesn't require any particular person to make this decision taking the full responsibility. It's just a change of some parameter by a few percents.
> Pre-AI equivalent: These prisoners seem like they're eating too much, so let's cut their food portions. One third reduction seems reasonable.
Bureaucracy and corporations are a type of AI, and they routinely make objectively evil decisions and people tend to follow. What computerized AI introduces is even stronger network effects, even easier monopolization, and even smaller accountability.
A huge bull market in surveillance because if you get rid of that our "innovations" severally lack use cases.