3 ms·

The machine recognizing you was the harder problem. That's the "Minority Report" stuff that people thought was in the _far far_ future that we don't have to wor

The machine recognizing you was the harder problem. That's the "Minority Report" stuff that people thought was in the _far far_ future that we don't have to worry about constantly. Storing and using that model across various devices is trivial. So if we've done the hard part, the easy part is _definitely_ going to be done. There's no doubt about that. States, companies that can convince the state that they somehow need the data (cough insurance companies) will, I repeat, WILL utilize it. And now, suddenly your premiums for this month have gone up cause in that ski trip you took last weekend, a camera noticed the lack of headgear for about 5 mins. That's the kind of risk the insurance company cannot take. Pay up.

bayesian_horse 7y ago
"Minority Report" was about something that Humans couldn't really do, meaning to look into the future, and they even needed psychics to do it... Good luck trying to advance science by avoiding to invent anything that can recognize image!
- skohan 7y ago
  Taken metaphorically we’re not far away. Imagine if we come to trust ML to make decisions for us in everyday life, and we build a system which can predict with high confidence that someone will commit a mass shooting. Is it so hard to believe that might be acted on?
  - bayesian_horse 7y ago
    To avoid these issues it is imperative to hold natural and legal persons responsible for the software they use, just as it has always been, regardless if it is using anything artificial intelligence-ish. Under that rule there will always be a way to handle problems arising from AI. Without following that rule, pretty much nothing else matters. In the case of the mass-shooting prediction: Good luck proving that it is accurate enough to legitimize punitive consequences for the suspected perpetrator, and without violating a half-decent democratic constitution. Some of the predictors are in themselves crimes, and picking those up would justify correctional measures, no problem. Schools and police also need to get a lot better at picking up and handling mental health problems, be it through algorithms or smarter laws and procedures. Make it easier for the mentally ill to get medical help than to buy an assault rifle, for starters.

bayesian_horse 7y ago

"Minority Report" was about something that Humans couldn't really do, meaning to look into the future, and they even needed psychics to do it... Good luck trying to advance science by avoiding to invent anything that can recognize image!

skohan 7y ago
Taken metaphorically we’re not far away. Imagine if we come to trust ML to make decisions for us in everyday life, and we build a system which can predict with high confidence that someone will commit a mass shooting. Is it so hard to believe that might be acted on?
- bayesian_horse 7y ago
  To avoid these issues it is imperative to hold natural and legal persons responsible for the software they use, just as it has always been, regardless if it is using anything artificial intelligence-ish. Under that rule there will always be a way to handle problems arising from AI. Without following that rule, pretty much nothing else matters. In the case of the mass-shooting prediction: Good luck proving that it is accurate enough to legitimize punitive consequences for the suspected perpetrator, and without violating a half-decent democratic constitution. Some of the predictors are in themselves crimes, and picking those up would justify correctional measures, no problem. Schools and police also need to get a lot better at picking up and handling mental health problems, be it through algorithms or smarter laws and procedures. Make it easier for the mentally ill to get medical help than to buy an assault rifle, for starters.