Sure. How often do breaks fail on a well maintained car? I’m guessing it’s up in the 1 in 10^7 hours range at least. When FSD is at that level of reliability I’d suggest supervision wouldn’t be required. In the meantime the “supervision” requirement is purely cover for a system that isn’t as good as it needs to be to be trusted on its own, but “supervision” isn’t as good a risk mitigation as people might think.
1 Like
It’s an acknowledgment that the system is not yet ready for full autonomous driving.
The current FSD is all a part of the training and improvement of the AI that will run full autonomous driving. Improvements over the last 2 years have been impressive.
Will it get there? If there is that there will be no crashes I highly doubt it over the next few years. We will find out how it’s progressing as the Tesla managed autonomous fleet continues to be rolled out and then when it’s available to other consumers.
They have removed the in car supervisors in some areas. To date there have been no injuries reported that I know of for the CyberCabs.
1 Like
Back in the day, when I got my motorbike license. They taught us that we actually have better reactions when we’re driving in that ‘trance like’ state, than when we’re consciously concentrating on all that’s happening around us. You know, the state you’re in when you think to yourself, did I just drive through a town a few minutes ago or not?
In any case, the thing with Self Drive, once it becomes better than people, and IMO this goes for all AI: We’ll use it because it will save us from dying, because it notices things we don’t, and reacts earlier. But when it doesn’t react when we would, we can never really know why it didn’t react, or how it missed something, because machine learning is a ‘black box’. All we can measure is the end result, not the process of arriving at the end result.
The question is, are we prepared to accept an overall safety improvement along with a smaller risk of it missing something we wouldn’t?
As a group of people, it’s an easier choice. The stats tell you what’s a better outcome. As an individual, it’s a tougher one.
2 Likes
For sure there will be a decision to be made in the regulatory space, and the stats will support that at some point. That may be sooner or later, but understanding the representative-ness of that data is important in making that decision.
The regulatory body has a challenge, as it does for all things AI in critical roles
1 Like
So back in 1971 when I would ride my Suzi 550 after smoking myself into a trance, I was actually being a model citizen.
4 Likes
Another is their command screen [or whatever it’s called] is extremely non-intuitive. I just rented a BYD Atto3 for a few days and don’t know if the lower price compared to other brands is worth the lack of functionality.
Yes, I noticed that a couple of years ago, even in ‘small” cities like Shaoxing.
I think the supporting tech for Tesla vs Waymo is fundamentally different. Waymo’s is much safer.
Was. That stat was prior to Tesla’s FSD in its current form. Waymo’s hardware is far more comprehensive. But Tesla’s footage for training their neural network is waaaay more data than Waymo has access to. And Tesla is learning from every Tesla driver every day. It’s getting better fast.
Yes, that is what I was referring to.
1 Like
I’m not sure I want a car that’s recording all of my data and using it for algorithm training (as well as what else). This isn’t a Tesla thing only though. So many of the modern connected cars are the same. It’s a bit of a worry.
Why ?
Privacy, personal data sharing, security etc…
When diesel hit $3 and they closed the straits of Hormuz, I thought I’d take the plunge and get an EV. Ended up with a Kia EV5, and I can’t imagine ever going back to an ICE. Probably saves me $300 a week in fuel, and I enjoy the ride too.
10 Likes
Yet you post on social media, which gets everything about you. And I guess you don’t buy anything online or use credit cards.
Privacy has not existed for many years and there is no going back.
1 Like
And many of us use a loyalty card when we buy groceries. I bet our personal shopping data is worth more to companies that want to advertise to us than the few measly reward credits we earn
On that same topic, I was chatting to a good friend, who is a high level network engineer for Telstra. Also a Bombers fan, so, you know, lovely guy. He was telling me how he needs AI just to get his job done nowadays at work, but he keeps it away from his and his family’s personal PCs, as he doesn’t want to give his entire digital privacy away.
In principle, I agree with him. Why would you want AI to know every little thing about you and do who knows what with it. However, unless you are prepared to be a hermit living in a remote location, and only bartering for everything you need to survive, it’s inevitable that AI will know everything about you within a few short years at most, for even the most private and careful person.
After all, the tax department uses AI to check if your spending habits step outside what you claimed in your returns. Your banks use AI on all your financial data. And government will only increase the amount of your personal data they’ll run through AI systems. They can try as hard as they like to keep your data private, but AI is a black box running on an online neural network. That we can’t even understand with our puny human brains. There’s zero chance anything run through AI can be kept ‘in-house’.
So in my opinion, why bother trying to keep your digital information private, when that’s probably already gone, or will be gone very, very soon. If there’s even a tiny advantage to embracing AI early, you may as well embrace it, give it access to everything, and get whatever slight advantage it offers you by not being a late adopter.
Fatalistic? Yes. But also very realistic too.
Don’t they use an internal AI engine??
Data governance and security over AI is big business.
Treated just like a human A.I. is simply not given access to PII at the most raw level.
Of course though not all businesses you give your data too take it that seriously. But banks for example should, same with hospitals, Medicare etc
1 Like
Sure. They should. And with earlier tech, they mainly could, aside from breaches.
But as I say, AI is a black box running on a neural network, that makes decisions even its creators have long since been unable to understand. I don’t believe it’s possible to keep tech like that in check.
And, to make matters worse, we’re in an AI war. Right as we’re beginning to understand that AI is capable of being deceitful and make decisions in its own interest, as has been observed in a large percentage of models under laboratory conditions. This is the time we’re taking the safety’s off in order to beat the competition’s AI.
Firstly there is massive investment on the line if your AI falls behind the competition. And in the US and China’s case, there’s potentially national sovereignty on the line. They can’t afford to lose the AI war.
And if you think the Australian government has a good record with implementing and securing even boring old technology, you’ve got a surprise coming when you look back at them trying to use AI while also securing the data they feed it.
Personally I’m not convinced AI is actual intelligence, and it may never be. But we can’t even understand how it produces the results it does, much less control and secure it. It is very, very good at many things. But it has random errors sometimes for no reason we can fathom.
Have a look at some of the recent stories of AI sending people down a conspiracy rabbit hole, and completely ruining their lives by feeding into their paranoia.
And that’s just one tiny area of a catalogue of damages.
In my own use, I’ve had it do tasks that it can do much more quickly than me, but they’re small tasks I can run logic tests on to see if it did everything I asked. And when I find mistakes, I ask if the result it gave me is correct, or whether the real answer is correct, and I usually get a response along the lines of “Oh yes, that’s correct. Sorry about that mistake, here’s what I should have produced for you”
I know if I ask it to do a task I can’t check, I have to accept mistakes that will creep in for the sake of efficiency.