hype about neural networks

Posted by: @robotbuilder

↑

but they are not intelligent

IMHO 🙂

That is very thought provoking.

Just what is considered to be "intelligence" ?

Have I been conned by the shear brute force of today's number crunchers ? I started out with a 6800, 1k of EPROM and 1k of RAM. That was enough to display MICBUG on a monitor. Everything was so exciting in those days.

So back to Tesla, is there intelligence in the car driving around, seeing signs, avoiding/overtaking other cars ? What I cannot answer is, does an isolated Tesla improve it's capabilities or does it all get fed back to base and they improve it's capabilities.

 I've seen many biped robots stacking boxes and avoiding obstacles at the same time - is there intelligence there ?

@hilldweller wrote:

"Everything was so exciting in those days."

Indeed it was 🙂

I would say that intelligent behavior is goal seeking behavior at its minimum. If you can see the goal the system is trying to achieve then you would likely say it was intelligent behavior even if of the simplest kind.

"So back to Tesla, is there intelligence in the car driving around, seeing signs, avoiding/overtaking other cars ?"

You could call it intelligent behavior in that it is a goal seeking behavior. But it is not advanced intelligence like the brain of a human driver even if the software is advanced to allow it to achieve this outcome at times with a precision and speed humans might not match. The car isn't "thinking about" what it is doing.

They are not "seeing signs" the way we do. It would be better the car has the information that it has to stop at certain intersections in its memory although this data base would have to be updated constantly.  A neural net's visual recognition solutions can be fooled and fail miserably. The problem with neural net solutions is we don't actually know how the data is being processed.

https://spectrum.ieee.org/slight-street-sign-modifications-can-fool-machine-learning-algorithms

https://thenextweb.com/news/researchers-tricked-ai-ignoring-stop-signs-using-cheap-projector

With the simple blob analysis I used for signs to being used as visual beacons the stop sign would not be seen as 30km speed limit sign!!

To enlarge an image, right click image and choose Open link in new window.

I now can see that AI + cars are not a done deal. But it's not a huge problem, when "they" tell us we can't drive any more and we must use Uber or whatever, road deaths will come down due to the drop in traffic as we all enjoy our 15 min cities. 15 min cities are being pushed in the UK as we speak.

I see the bigger problem as control and distribution of food with the computer doing the logical action of "no food for NY south today, they have not met their targets". For "food" substitute gas/electric/water/meds etc. Smart meters are being pushed in the UK as we speak.

Then the Boston Dynamic machines making sure we enjoy our 15 min cities but not one inch more.

It really does not matter just how these controls are implemented all that matters is that the capability is there.

I am rambling on from a background of designing CNC systems for woodworking factories. All that mattered to the bosses was a man could be more productive, down graded or fired. Humans were the enemy. Look at a modern workplace and it's and it shows how far we have progressed.

@hilldweller wrote:

"I now can see that AI + cars are not a done deal."

My point was that deep learning network programs are not intelligent in the sense we are. Neural nets do not work like fully functioning biological brains which is the hype I was referring to when they claim they are "intelligent".

Self driving cars rely on GPS and electronic bump sensors like lidar and radar. This is sufficient for robot farm tractors that can move about with pin point accuracy using GPS.  Visual recognition isn't advanced enough to be safely relied on to work every time.

Humans can deal with unexpected events while driving on the roads. Neural nets are limited to dealing with data they have been trained on. Thus if a bus overturns it ceases to be a bus if overturned buses were not in the training data. The human visual system is many times more sophisticated and capable of creating a 3d world around us than the simple visual pattern recognition done by neural nets. Current neural nets do not model the world as we know it.

Posted by: @robotbuilder

↑

@hilldweller wrote:

Humans can deal with unexpected events while driving on the roads.

We can also screw up badly especially in the cell phone era.

I watched a YouTube video a while ago. USA. The amateur video set a challenge, his Tesla against an unmanned Uber. Simple, go from A to B in a city. It was highly impressive no what the underlying technology is. I forget how the Uber performed in real time, but it did complete the journey. The Tesla owner filmed it well, it was like watching a newly qualified driver in action. It coped with lanes closed for roadworks.

Another video that caught my eye and semi-truck system, close to going live. Their logic, a driver has limited hours, replace with a computer, it can drive for 24 hours and a bit slower and they are on a winner. They are doing the easy part, just hub to hub on main roads and then a human hitches up a manual truck for the last miles. Another human bites the dust.

I might have drifted off your topic, I hope this is acceptable.

@hilldweller In regard to

 I've seen many biped robots stacking boxes and avoiding obstacles at the same time - is there intelligence there ?

NO, that is not what I would call intelligence. That is simple sensor-based movement. They are not that different from a modern-day elevator with sensors to find the doors and to detect if the 'rope' has broken. Most assembly lines use things like micro switches, photodetector switches etc., to control the movement of the line. A more sophisticated example is the commercial airplane auto-landing system. It looks miraculous, but it is simply sensor inputs and procedural code.

To be 100% honest, I am not sure what AI is, but the hype suggests something other than what exists today. Remember, ChatGPT IIRC has a record of 50% wrong answers, I would not call that any intelligence. And recently, even answering rather simple questions, the accuracy has dropped from the high 90s to under 10% correct. Yes, it is adapting, but in the WRONG direction!

Posted by: @zander

↑

@hilldweller In regard to

 I've seen many biped robots stacking boxes and avoiding obstacles at the same time - is there intelligence there ?

NO, that is not what I would call intelligence. That is simple sensor-based movement.  

Damned impressive though when you see them climbing onto a box and down the other side.

Of course I/we see only what they want us to see, they never show how many times if fell over or for that matter it would be possible to fake it with a hand control.

I've seen enough videos of the robot dogs to be convinced they really work. Like the UNO, the cheap Chinese copies of Boston Dynamics are out there now.

We are in Dronebot Workshop, I guess many of us have a self balancing robot to show how easy it is on a very basic level. One thing I've not spotted is a self tuning PID to get these rock solid automatically.

@hilldweller Boston Dynamics has been owned by the South Korean company Hyundai since 2020. As far as faking it, I am pretty sure the US Army would detect that when they tested the robots for various jobs in the army.

@hilldweller wrote:
"I might have drifted off your topic, I hope this is acceptable."

I have said all I wanted to say so by all means post whatever you like.

Posted by: @hilldweller

↑

What I cannot answer is, does an isolated Tesla improve it's capabilities or does it all get fed back to base and they improve it's capabilities.

I'm just starting this thread and haven't read past this post.  

That is an excellent question!  In one regard, I have heard that it is supposedly adapting to your driving habits.  BUT... does that mean that if you manually do something new (and good and useful) does your car, incorporate that directly into its learning?  OR does it only upload back to Tesla and someone (software developer/engineer/manager) or something (DOJO) evaluate it and choose to incorporate into the common brain that is later downloaded in an overnight update?

Not that I have any actual information on the subject; but I suspect it is the latter.  They are already talking about having issues with one of the Internet exposed AI's and how it peaked on its %correct and then started getting more stupid as time went by.  I also would think because of the quantity of the data points coming in on your one car, that there isn't enough compute power on the car to do the back propagation steps of ANN.  Although the car's compute engine is impressive, if it could handle the problem size, the DOJO wouldn't have really been needed.   

I personally think it is a waste of time debating whether AI, ANN, GA is "Intelligent".  First, the term is not even really defined.  There is something called human intelligence and AI is not that.  IT IS ARTIFICIAL INTELLIGENCE... by definition!  Until there is a common understanding among all speakers within the debate (or universally defined) definition for intelligence, we might as well be debating how beautiful some person/place/thing is and/or which one is more beautiful.  It makes about the same amount of sense.

Really... the only answer that can be of use is determining when something is useful ENOUGH.  When does an AI replace a human worker.  Does it make it intelligent?  I would say no.  But, I would also say, who cares?  If it is useful, it is going to happen!

1. Graphic artist are being displaced in droves.  Writers are being displaced in droves.  Entry software developers are losing entry.  Hundreds of professions are already feeling the pink slip.  I want to ask how does a young potential software developer get enough experience to jump over the AI software developer.  Is that intelligence.  Again, no in my book.  Is it useful... well at least it is to the employer and the consumer, but not the employee.
2. What happens when a Tesla robot can replace a automobile line-worker!?  Is that robot intelligent.  No, but who cares?!
3. We all think we're great drivers... you might hear yourself saying, "There is no way, I going to let a robot car drive ME around!"  At the same time, you see the stupid things people do on the road.  Drunk drivers, soccer persons (formally, Soccer Mom's) with ten screaming brats in their car, individuals texting while driving, others running red-lights... ad nauseum.  Teslas, Waymo, Uber, etc are already statistically safer than an actual person.  Can you at least be honest with yourself and say, those people should be using self-driving cars?  Then look harder at yourself!  Try actually getting on YouTube and do the search "tesla saves life".  One shows a wheel that came off a car on the opposite side of an US Interstate with the huge concrete barrier between directions.  The wheel traveling 70mph jumped over the barrier and was heading toward a Tesla going 70mph in the opposite direction.  The car saw it and dodged it.  The driver NEVER saw it.  Only after looking at the recording, did the cold chill run down their spine.  Is that intelligence?  Who cares if we say it is intelligent or not!  It is useful!