Even After $100 Billion, Self-Driving Cars Are Going Nowhere

Yep, in the UK, if I'm on the inside on a motorway, I'll move out to overtake, and then back in. I'll turn corners, and go around roundabouts etc. The degrees of freedom for steering seem so limited, that it does not interfere greatly with 'feeling' this binary state of whether to go faster or go slower - as you're stuck with just following the road ahead. (Again, this doesn't apply if you don't know the road/route, or are in heavy traffic).

You can see that it might be a different, but related perspective to the Ouija board...

Ouija participants know the board layout
Flow drivers know the road layout...

Ouija participants are moving with the puck
Flow drivers are moving with the car...

Ouija participants are sharing the same movement - limbs on puck
Flow drivers are sharing the same movements - limbs on steering wheel, foot on accelerator...

Ouija participants are asking for binary yes or no
Flow drivers are feeling for binary faster or slower...

Ouija participants can convince themselves that they are not in control
Flow drivers are learning to distinguish between the feeling of "what I want", from this 'other' feeling/state that 'fits in' with other drivers.

So you want to claim that the 'in the flow' situation requires nothing paranormal? Another interpretation might be that these states are accessing paranormal information.

David

So you want to claim that the 'in the flow' situation requires nothing paranormal? Another interpretation might be that these states are accessing paranormal information.

David

not too sure what your meaning here?

Feds close probe into TuSimple autonomous truck crash

Alan Adler

A driver-supervised autonomous truck took an unintended sharp left turn across a lane of westbound traffic on I-10 and struck a concrete barrier. The safety driver tried to countersteer the truck, which followed a computer-generated command that was several minutes old.

TuSimple initially said the incident was driver error. Later it acknowledged its compute system and the safety driver both bore responsibility. The company grounded its fleet of mostly Navistar International trucks and made software fixes to avoid a repeat occurrence.

So the humans in the cars will be the "fall guys" when something goes wrong, in order to keep confidence in the AI...

Feds close probe into TuSimple autonomous truck crash

Alan Adler


So the humans in the cars will be the "fall guys" when something goes wrong, in order to keep confidence in the AI...

When driverless vehicles proved to be very hard to create, a sneaky alternative was to suggest that there were various levels of automation, starting with automatic parking and the like, and the penultimate step was vehicles that would drive but there would be a competent driver ready to take over if it screamed for help! This was a horrible idea because imagine that your truck drives automatically for some time, and then suddenly demands you take over.

Taking over in an emergency like that must be nearly impossible.

David

not too sure what your meaning here?

Well I was suggesting that human drivers make some use of psi to avoid crashes. Then you introduced the idea of 'flow', and I am not sure if you see this as being a psychic process, or as just another way of using your brain.

David

When driverless vehicles proved to be very hard to create, a sneaky alternative was to suggest that there were various levels of automation, starting with automatic parking and the like, and the penultimate step was vehicles that would drive but there would be a competent driver ready to take over if it screamed for help! This was a horrible idea because imagine that your truck drives automatically for some time, and then suddenly demands you take over.

Taking over in an emergency like that must be nearly impossible.

David

Yeah the goal post moving would be funny if human lives weren't at stake.

I do believe that driverless cars may be possible in the future but using machine "learning" is not really AI, it's just hoping you get enough data to cover liability.

The kind of research that IMO is needed - bare metal higher level programming + specially designed hardware - is very much in its infancy. Likely because if one is honest about the domain space of driving we would have to admit we likely don't have the programming language nor hardware to truly express it properly at this time.

The issue with hype driven investment, a larger issue than just driverless cars sadly.

Feds close probe into TuSimple autonomous truck crash

Alan Adler


So the humans in the cars will be the "fall guys" when something goes wrong, in order to keep confidence in the AI...

Well I was suggesting that human drivers make some use of psi to avoid crashes. Then you introduced the idea of 'flow', and I am not sure if you see this as being a psychic process, or as just another way of using your brain.

David

Ok, depends on your perspective I suppose... but, neither of the two options you gave me really fit with how I understand things.

That was a sobering video, and it contained a good description of how a neural net 'decides' what something is.

David

Maybe the real problem with the neural net approach, and probably many other AI strategies, is that they only converge on the truth in the limit where every possible input is covered adequately in the training set - including images that are partially covered up - and then how does the AI know which part of the image is covered by something else?

I wonder how they perform on the "prove you aren't a robot" images where you have to click all the squares that contain a bicycle. Some of the squares only show a bit of the object in question, and you are meant to recognise those situations.

David