AI shouldn’t make ‘life-or-death’ decisions, says OpenAI’s Sam Altman
AI shouldn’t make ‘life-or-death’ decisions, says OpenAI’s Sam Altman
Human beings will continue to decide “what should happen in the world” regardless of the rise of artificial intelligence, OpenAI’s Sam Altmann said Thursday at the World Economic Forum in Switzerland.
Sam Altman, CEO of OpenAI, speaks at the meeting of the World Economic Forum in Davos, Switzerland. (Denis Balibouse/Reuters)
You're viewing a single thread.
But it should drive cars? Operate strike drones? Manage infrastructure like power grids and the water supply? Forecast tsunamis?
Too little too late, Sam. 
But it should drive cars?
Oh, definitely. Humans are shit at that. Get bored when we have to concentrate for 10 minutes.
Yes on everything but drone strikes.
A computer would be better than humans in those scenarios. Especially driving cars, which humans are absolutely awful at.
So if it looks like it’s going to crash, should it automatically turn off and go “Lol good luck” to the driver now suddenly in charge of the life-and-death situation?
I'm not sure why you think that's how they would work.
Well it's simple, who do you think should make the life or death decision?
The computer, of course.
A properly designed autonomous vehicle would be polling data from hundreds of sensors hundreds/thousands of times per second. A human's reaction speed is 0.2 seconds, which is a hell of a long time in a crash scenario.
It has a way better chance of a 'life' outcome than a human who's either unaware of the potential crash, or is in fight or flight mode and making (likely wrong) reactions based on instinct.
Again, humans are absolutely terrible at operating giant hunks of metal that go fast. If every car on the road was autonomous, then crashes would be extremely rare.
Are there any pedestrians in your perfectly flowing grid?
Again, a computer can react faster than a human can, which means the car can detect a human and start reacting before a human even notices the pedestrian.
Plus, there will be far fewer variables when humans aren't allowed to drive outside of race tracks and the like. Reason why fully AI cars are a bad idea right now is because of all the chaotic human drivers that react in nonsensical ways. e.g. Pedestrian steps out. Thing that makes sense is for the AI to stop the car. But then the driver behind them decides to swerve around and blare the horn, then see the pedestrian, freak, turn into the AI car, and an accident is caused. Without the human drivers, then all the vehicles can communicate with each other and all of them can react in appropriate ways, adjusting how they drive up to miles back
Have you seen a Tesla drive itself? Never mind ethical dilemmas, they can barely navigate the downtown without hitting pedestrians
Teslas aren't self driving cars.
According to their own website, they are
Well, yes. Elon Musk is a liar. Teslas are by no means fully autonomous vehicles.
Here's the summary for the wikipedia article you mentioned in your comment:
No true Scotsman, or appeal to purity, is an informal fallacy in which one attempts to protect their generalized statement from a falsifying counterexample by excluding the counterexample improperly. Rather than abandoning the falsified universal generalization or providing evidence that would disqualify the falsifying counterexample, a slightly modified generalization is constructed ad-hoc to definitionally exclude the undesirable specific case and similar counterexamples by appeal to rhetoric. This rhetoric takes the form of emotionally charged but nonsubstantive purity platitudes such as "true", "pure", "genuine", "authentic", "real", etc. Philosophy professor Bradley Dowden explains the fallacy as an "ad hoc rescue" of a refuted generalization attempt.
As advanced cruise control, yes. No, but in practice it doesn't change a thing as humans can bomb civilians just fine themselves. Yes and yes.
If we're not talking about LLMs which is basically computer slop made up of books and sites pretending to be a brain, using a tool for statistical analysis to analyze a shitload of data (like optical, acoustic and mechanical data to assist driving or seismic data to forecast tsunamis) is a bit of a no-brainer.