On one hand it is far superior to humans. On the other hand it needs our guidance. Both of these are true at the same time.
When when I got my CS degree, I revisited this thought, and reinforced it.
I thought, this is not how AI and Statistics work. We would need to have an ensemble of thousands of specialized models to run a ship.
Self Driving has been working since 2009? But is still extremely limited in scale. Given the amount of money to be made, and time saved by eliminating the need for human drivers, What is the hold up?
It has only now become somewhat reliable. They still don't have the new version the Robotaxis run. This version should roll out to consumers later this year or in Q1.
Then the next generation of the onboard AI hardware is expected Mid-2027, supposedly featuring 10x the TOPS and 5x the effective inference performance.
I have no idea about the timeline for an unsupervised, globally working self-driving feature.
My uneducated guess would be that perhaps the next major version of the model, on the new hardware, could be reliable enough.
This is the bar. If I have to be awake and at the wheel, ready to take over it’s just a gimmick. Automating the only enjoyable part of the driving experience and keeping the worst part (constant alertness) is only a gimmick.
In a perfect world, we could broadly implement self-driving the moment it would result in a net-increase in safety. Alas.
I presume you're not accounting how they perform in these conditions. Alas.
Culture shock to me first encountering it, seemed insane, but now I see how bad traffic would be here if people didn't do that.
What do self-driving cars do when encountering that? Get honked at and dangerously double-passed themselves? Do they shift right up onto the line when waiting to turn left, or do they stay in the middle of the lane?
How many other weird, sketchy things are going on with driving behavior, that's in fact happening for pretty good reasons, in other places? How many do I do but never even notice because they're just "normal"? Do the self-driving cars know to do that in (at least parts of) New England without even touching the brake, but to be far more hesitant about that kind of passing in, say, the Midwest?
Edge cases, fear, regulation, capital expense, physical movement of goods, etc etc?
The deployment is obviously accelerating like crazy.
However, this isn't a binary scenario anyway and the limited availability today is exactly why I think that this is the next thing. Similar to the first version of ChatGPT that was also quite limited in what it could do and not widely used yet.
The most unrealistic thing about it is that it would often openly reply “unknown” and then stop talking.
It's first and foremost a revolution in computer UI, as I see it. Or at least it will be as soon as I can afford the hardware to run the damn things on my computers...
I also liked how Star Trek (probably without planning to) created an entire spectrum of AIs with different abilities, starting with the Computer, to the semi-sentient Holodeck characters, to the EMH and Data and finally to the Borg as an AI species.
Maybe we should use this scale to make different concepts of AI more concrete instead of vague concepts like AGI and ASI.
I recall thinking at the time I first watched them that it was a bit weird that the computer seemed capable of answering general questions and doing things like data analysis, but that seemed to be all it was ever used to do.
And I think this is exactly the kind of AI we now have today, you can ask it a general question which it can somewhat mostly answer correctly, but requires human guidance to be applied effectively at real world problems.