Electric Car Prediction - Whatdayathing???

[Chris Chesley]

As a car manufacturer (not me, but if I were) I'd still be pretty wary of a wholesale shift to only all-electric vehicles. When was the last study/ audit made regarding overall national electric capacity made? If EV's were to become 5,10, or 20% of the vehicle population in just a few years, do we REALLY have the capacity to charge them with our current generating infrastructure? Without more CO2 emissions as well?

Secondarily, while I would welcome small, lightweight, long(ish) range EV's, I would not have much (any?) desire to share the freeways with even half of the existing larger, heavier ICE vehicles, not even counting trucks.

Another concern I have it the exponential growth of building, maintaining and disposing of or recycling batteries may not be as wonderful as it seems when done on a larger scale. ( i.e. rare earths availability is currently constrained, energy req'd to build the batteries may exceed energy ultimately delivered by the batters (see similar issues with PV panels) and hazmat/disposal issues)

Overall, I suspect there may be a few follow on consequences of the wholesale shift towards EV's in our transportation networks. What works well on an individual basis may not really scale up so well....

[kdh]

It's amusing now, but I think the improvement will come fast. Every time I click on a captcha I feel like I'm helping refine their driving algorithms. Is that why am I having to click on the traffic lights?

If they're as bad at that as I am, watch out!

I'm skeptical mostly because of my work in artificial intelligence in the 80s when there was similar apparent promise that was unfulfilled. The current crop of people touting AI's potential weren't around in the 80s. Yes, even though "Moore's Law" (exponential growth of transistor density) is dead, computers are faster and have a shit-ton more memory than in the 80s, and we have much more data available, but machine learning, even modern "deep learning" (a neural net with hidden layers) is still nothing more than learning by example, which is hugely limiting in its potential efficacy.

We're naturally fooled by the notion that if we humans can do something it's easy to teach a computer to do it. This is basically Musk's argument against using LIDAR sensors in Teslas--they're not needed because we humans don't use LIDAR. This is laughably naive.

[Benno von Humpback]

It's amusing now, but I think the improvement will come fast. Every time I click on a captcha I feel like I'm helping refine their driving algorithms. Is that why am I having to click on the traffic lights?

If they're as bad at that as I am, watch out!

I'm skeptical mostly because of my work in artificial intelligence in the 80s when there was similar apparent promise that was unfulfilled. The current crop of people touting AI's potential weren't around in the 80s. Yes, even though "Moore's Law" (exponential growth of transistor density) is dead, computers are faster and have a shit-ton more memory than in the 80s, and we have much more data available, but machine learning, even modern "deep learning" (a neural net with hidden layers) is still nothing more than learning by example, which is hugely limiting in its potential efficacy.

We're naturally fooled by the notion that if we humans can do something it's easy to teach a computer to do it. This is basically Musk's argument against using LIDAR sensors in Teslas--they're not needed because we humans don't use LIDAR. This is laughably naive.

As someone who studies human cognition I could not agree more.

[Jamie]

It's amusing now, but I think the improvement will come fast. Every time I click on a captcha I feel like I'm helping refine their driving algorithms. Is that why am I having to click on the traffic lights?

If they're as bad at that as I am, watch out!

I'm skeptical mostly because of my work in artificial intelligence in the 80s when there was similar apparent promise that was unfulfilled. The current crop of people touting AI's potential weren't around in the 80s. Yes, even though "Moore's Law" (exponential growth of transistor density) is dead, computers are faster and have a shit-ton more memory than in the 80s, and we have much more data available, but machine learning, even modern "deep learning" (a neural net with hidden layers) is still nothing more than learning by example, which is hugely limiting in its potential efficacy.

We're naturally fooled by the notion that if we humans can do something it's easy to teach a computer to do it. This is basically Musk's argument against using LIDAR sensors in Teslas--they're not needed because we humans don't use LIDAR. This is laughably naive.

From what I understand of machine learning, no teaching or learning the way we understand it is performed at all, and we actually don’t know the underlying means by which the “taught” algorithm performs its tasks.

It only needs to be better than us, and we’re not tha great.

[BeauV]

Crash repair would probably depend on whether the battery pack was damaged.

Beau the Tesla charger could be made compatible with Porsche. I just don’t see why Tesla would do it.

I'm pretty sure Tesla won't do it. They are the leader in this market and have seriously upstaged companies like Porsche, BMW and MBZ. It's up to those guys to figure out a way to utilize the existing infrastructure. I don't think this is a real tech issue, it's just arrogance on the part of the old car companies. Obviously, Tesla will charge for power and make a profit.

Folks don't realize that as of January (10 months ago) there were over 12,000 superchargers deployed. There will be another 12,000 deployed this year alone. Ignoring that is idiotic and arrogant. The low power independent chargers were just de-installed at a few local parking lots, they weren't used enough to make money, the Tesla chargers often have lines, they're run at a profit.

The traditional car manufacturers are very late to the party and once a tech-based industry establishes a de facto standard, the followers simply don't have an option. Conform or die. Porsche doesn't appear to understand this; neither does Toyota with its dual strategy of Hydrogen fuel cells and gasoline hybrids.

I may be overstating it, but auto companies are terrible at technology transitions. Good grief, some still have their map database on a CD-ROM while the download over cellular is both cheaper and more reliable.

[BeauV]

As a car manufacturer (not me, but if I were) I'd still be pretty wary of a wholesale shift to only all-electric vehicles. When was the last study/ audit made regarding overall national electric capacity made? If EV's were to become 5,10, or 20% of the vehicle population in just a few years, do we REALLY have the capacity to charge them with our current generating infrastructure? Without more CO2 emissions as well?

Secondarily, while I would welcome small, lightweight, long(ish) range EV's, I would not have much (any?) desire to share the freeways with even half of the existing larger, heavier ICE vehicles, not even counting trucks.

Another concern I have it the exponential growth of building, maintaining and disposing of or recycling batteries may not be as wonderful as it seems when done on a larger scale. ( i.e. rare earths availability is currently constrained, energy req'd to build the batteries may exceed energy ultimately delivered by the batters (see similar issues with PV panels) and hazmat/disposal issues)

Overall, I suspect there may be a few follow on consequences of the wholesale shift towards EV's in our transportation networks. What works well on an individual basis may not really scale up so well....

You are absolutely correct. There most certainly will be serious follow-on negative effects. (Rant Coming) That won't slow this at all. Since when did the Auto industry ever pay for the massive down-stream costs of what they built? They didn't pay for the roads and freeways, they sold cars and let the customers demand that the infrastructure get built. In our "free market" and "no responsibility" market place none of those long-range downstream costs are ever born by the builder of the device. This obviously holds for commercial aircraft and a host of other innovations. It clearly holds for companies like Facebook.

There is no real mechanism in our economy to charge the builder of a truely bad device or service for the damage they do if customers want it.

I submit the simple fact that the Ford F-150 pickup truck has been the best selling vehicle in the US for almost ever, and it is clearly an idiotic choice for a massive number of customers. It has all sorts of things wrong with it, but the buyers don't care. Its primary typical use is as a commuter vehicle. Which is absurd.

We have a deeply mistaken belief in many of the supposed benefits of what we laughingly call a "free market". An inability to deal with long term consequences of what we do is just one of them. But, you're absolutely correct, it's a terrible aspect of our stupidity as a people. (Rant Off)

[kdh]

It's amusing now, but I think the improvement will come fast. Every time I click on a captcha I feel like I'm helping refine their driving algorithms. Is that why am I having to click on the traffic lights?

If they're as bad at that as I am, watch out!

I'm skeptical mostly because of my work in artificial intelligence in the 80s when there was similar apparent promise that was unfulfilled. The current crop of people touting AI's potential weren't around in the 80s. Yes, even though "Moore's Law" (exponential growth of transistor density) is dead, computers are faster and have a shit-ton more memory than in the 80s, and we have much more data available, but machine learning, even modern "deep learning" (a neural net with hidden layers) is still nothing more than learning by example, which is hugely limiting in its potential efficacy.

We're naturally fooled by the notion that if we humans can do something it's easy to teach a computer to do it. This is basically Musk's argument against using LIDAR sensors in Teslas--they're not needed because we humans don't use LIDAR. This is laughably naive.

From what I understand of machine learning, no teaching or learning the way we understand it is performed at all, and we actually don’t know the underlying means by which the “taught” algorithm performs its tasks.

It only needs to be better than us, and we’re not tha great.

Yes, this is the way it's characterized by most who write about it, and as such is a common view.

A neural net is a map, a function, that takes input data, driving information from sensors in this case, to an output, the steering and accelerator controls of the car. The function is parametrized, i.e., has neural net "weights" learned, or more often called "trained," from input-output pairs, such as from the Captcha data we've all been providing on identifying traffic lights, motorcycles, pedestrians, etc. The functional form of a neural net is often described as "non-linear," and indeed it's easy to prove that even without a hidden layer, associated with so-called "deep learning," a net can approximate any practical function with arbitrarily good accuracy.

In other words, to use perfectly adequate terms, the neural net represents a "statistical fit to data." A good reference to these modern statistical techniques is "Pattern Recognition and Neural Networks" by Brian Ripley if you want to cut through all the silly and unnecessary new terms and perspectives and just understand the math.

[kdh]

There is no real mechanism in our economy to charge the builder of a truely bad device or service for the damage they do if customers want it.

We have a deeply mistaken belief in many of the supposed benefits of what we laughingly call a "free market". An inability to deal with long term consequences of what we do is just one of them. But, you're absolutely correct, it's a terrible aspect of our stupidity as a people. (Rant Off)

I'll argue that our system has a mechanism--government regulation. Carbon taxes or battery-disposal taxes or generally rules that shift incentives to protect the public good. To me anyway, a pure profit motive is inadequate.

[Jamie]

It's amusing now, but I think the improvement will come fast. Every time I click on a captcha I feel like I'm helping refine their driving algorithms. Is that why am I having to click on the traffic lights?

If they're as bad at that as I am, watch out!

I'm skeptical mostly because of my work in artificial intelligence in the 80s when there was similar apparent promise that was unfulfilled. The current crop of people touting AI's potential weren't around in the 80s. Yes, even though "Moore's Law" (exponential growth of transistor density) is dead, computers are faster and have a shit-ton more memory than in the 80s, and we have much more data available, but machine learning, even modern "deep learning" (a neural net with hidden layers) is still nothing more than learning by example, which is hugely limiting in its potential efficacy.

We're naturally fooled by the notion that if we humans can do something it's easy to teach a computer to do it. This is basically Musk's argument against using LIDAR sensors in Teslas--they're not needed because we humans don't use LIDAR. This is laughably naive.

From what I understand of machine learning, no teaching or learning the way we understand it is performed at all, and we actually don’t know the underlying means by which the “taught” algorithm performs its tasks.

It only needs to be better than us, and we’re not tha great.

Yes, this is the way it's characterized by most who write about it, and as such is a common view.

A neural net is a map, a function, that takes input data, driving information from sensors in this case, to an output, the steering and accelerator controls of the car. The function is parametrized, i.e., has neural net "weights" learned, or more often called "trained," from input-output pairs, such as from the Captcha data we've all been providing on identifying traffic lights, motorcycles, pedestrians, etc. The functional form of a neural net is often described as "non-linear," and indeed it's easy to prove that even without a hidden layer, associated with so-called "deep learning," a net can approximate any practical function with arbitrarily good accuracy.

In other words, to use perfectly adequate terms, the neural net represents a "statistical fit to data." A good reference to these modern statistical techniques is "Pattern Recognition and Neural Networks" by Brian Ripley if you want to cut through all the silly and unnecessary new terms and perspectives and just understand the math.

Thanks, I'll look that up. That seems a perfectly good definition of "intelligence" ? I can catch a ball without understanding physics.

[BeauV]

There is no real mechanism in our economy to charge the builder of a truely bad device or service for the damage they do if customers want it.

We have a deeply mistaken belief in many of the supposed benefits of what we laughingly call a "free market". An inability to deal with long term consequences of what we do is just one of them. But, you're absolutely correct, it's a terrible aspect of our stupidity as a people. (Rant Off)

I'll argue that our system has a mechanism--government regulation. Carbon taxes or battery-disposal taxes or generally rules that shift incentives to protect the public good. To me anyway, a pure profit motive is inadequate.

Keith, I agree that there is supposed to be a consequence. But, one has great difficulty finding a genuine "cost" applied to the creator of the problem. Run through things like:
- dam breaks in coal slag ponds
- polluted wells and earthquakes from fracking
- river and groundwater pollution from places like hog and cattle farms
- air pollution from cars driven in one entity blowing into another state, or county or country
- foster children created by opioids
- bankrupcies created by astoundingly and usurious interest rates
- traffic impact of next day delivery by Amazon and others
- medical impact of smoking
- medical impact of lead in gasoline
- etc.....

I completely agree that the profit motive is completely inadequate. My position is that the regulatory methodology used in the US is equally inadequate, despite some real efforts by some folks. Once profits start to fall the PR machine reves up attacking "over-regulation" and the "nanny state" and all manner of utter claptrap. When what's really going on is businesses attempting to retain a profit in a business that is not profitable if long term consequential damage is paid for.

I get genuinely pissed off when folks put a company's profit ahead of the obvious societal good. In some cases, the early returns on a dangerous activity aren't clear. But in the case of lead in gasoline, everyone knew it would be terrible. We know that because the head of R&D of General Motors told the CEO and the Board of Directors exactly what was going to happen. They ignored him. He was right. We damaged multiple generations of people so that Dupont and GM could spit the profits on tetraethyl lead. Disgusting.

[BeauV]

Regarding "deep learning" and "neural nets" Keith has it right. It is a simple matter of pattern matching. My Dad and I built a "computer" out of raisin boxes with colored M&Ms in them. It played Othello.

On the outside of each box was a drawing of a possible state in which the board of the game could find itself. All states were represented.

All legal moves were represented on the drawing and color-coded to one of the M&Ms inside the box. If there were three legal moves available, there were three M&Ms with colors corresponding to the drawings of the moves.

My sister and I started to play Othello against the "computer". My much younger brother operated the computer and programmed it. The operational part of the computer was to find the box with a picture that matched the current state of the game, extract an M&M at random, and make the move on the picture which matched the color of the M&M extracted. Then, put the M&M back into the box provided that move didn't cause the "computer" to lose the game.

If the move caused the computer to lose the game, then the computer operator (my little brother) did NOT put the M&M back into the box, he ate it. This altered the program by removing the move just prior to a loss each time the "computer" lost. The computer was being trained. Eventually, all games ended up as a win for the computer or a draw. No human could beat the computer as we had pruned the decision tree of all bad moves by eating the M&Ms.

It was a great experience for us, three kids. I think Dad devised this from some Scientific American article about doing a similar thing mathematically. I believe his innovation was using raisin boxes and colored M&Ms. That certainly made the game more fun. Dad would occasionally add M&Ms to the boxes randomly to see what would happen. Then we'd have to re-train the computer so that it could win or draw every game.

One can see that this could easily be applied to games like tic-tac-toe etc... It's not so obvious that one could apply the method to chess. However, modern computers are so fast that in modern AI they are simply eating a lot of M&Ms which represent "bad moves".

I know some guys who are trying to get a computer to play Go better than humans using a large scale version of eating M&Ms. It's beginning to work. But, in the end, it's just eating M&Ms.

[Jamie]

I don’t see the difference with “organic “ learning except that computers don’t forget and can have access to larger data sets.

“Free” market is a myth. Even chaotic systems develop “rules” over time.

[BeauV]

I don’t see the difference with “organic “ learning except that computers don’t forget and can have access to larger data sets.

“Free” market is a myth. Even chaotic systems develop “rules” over time.

Jamie, I agree that computers have massive data sets to work with and they appear to process it faster than we can. But.... I think that we are attending to the conscious mind, the one that thinks rationally and/or logically. We are ignoring, or at the least disregarding, the unconscious mind. I believe this is a flaw in our understanding of "thought" and "understanding".

From what I've read, Mozart simply wrote down the music as he knew it should be. He wasn't "composing" he was "documenting" what was flowing through his mind. I've considered this for a long time and feel that there is some other form of thought which isn't available to the linguistically and logically oriented bits of our brains. It is the "other brain" which takes over when one is skiing a really difficulty line, when driving insanely fast and the world slows down, when sitting at the piano and just letting whatever is in there come out. It's then that I feel that other metal capacity straining to get out.

It's fun to consider what capacity is held by this other part of the brain.

When faced with an insanely difficult problem, I go for a walk. I try to get lost. I try to "not think". Then, all at once, the answer appears. All at once I know what the solution is. I have no idea how that knowledge entered my mind, but it's obviously correct. It's blindingly correct.

I've spent most of my adult life trying to be able to call up this feeling on demand; mostly failing.

Perhaps this is what the zen master really knows: how to enter this mental state. I have no idea how to do it.

[kimbottles]

I solved numerous problems in my working life while riding a bicycle.

Some of those problems getting solved were critical to the success of our company.

[Jamie]

I don’t see the difference with “organic “ learning except that computers don’t forget and can have access to larger data sets.

“Free” market is a myth. Even chaotic systems develop “rules” over time.

Jamie, I agree that computers have massive data sets to work with and they appear to process it faster than we can. But.... I think that we are attending to the conscious mind, the one that thinks rationally and/or logically. We are ignoring, or at the least disregarding, the unconscious mind. I believe this is a flaw in our understanding of "thought" and "understanding".

From what I've read, Mozart simply wrote down the music as he knew it should be. He wasn't "composing" he was "documenting" what was flowing through his mind. I've considered this for a long time and feel that there is some other form of thought which isn't available to the linguistically and logically oriented bits of our brains. It is the "other brain" which takes over when one is skiing a really difficulty line, when driving insanely fast and the world slows down, when sitting at the piano and just letting whatever is in there come out. It's then that I feel that other metal capacity straining to get out.

It's fun to consider what capacity is held by this other part of the brain.

When faced with an insanely difficult problem, I go for a walk. I try to get lost. I try to "not think". Then, all at once, the answer appears. All at once I know what the solution is. I have no idea how that knowledge entered my mind, but it's obviously correct. It's blindingly correct.

I've spent most of my adult life trying to be able to call up this feeling on demand; mostly failing.

Perhaps this is what the zen master really knows: how to enter this mental state. I have no idea how to do it.

The mind is an incredible tool that we mostly don’t understand. I often do the same. My most creative ideas come at the oddest times. Ideas sometimes are like the water wiggle toys; you have to let go to hold on to them.

But, what exactly is consciousness? I think the conscious mind is more common than we think, comes in gradations, and in forms we do not recognize. That consciousness is unique is a bit of a conceit of humans. ( Define rational or logical? What is experience? You usually end up boiling it down to data and observed relationships. The Newtonian world we live in just an approximation of the statIstical probabilities of the quantum. We rely on gestalt when we we do not or cannot understand all the parts and must make inferences. My guess is that teach algorithms to make inferences about a car park that is as good as or better than we can is closer than we think. It might take lidar (sorry Elon) and it might be a low level form of machine consciousness.

[Olaf Hart]

That intuitive mind works on pattern recognition Beau.

My mind saves long term memory in files based on patterns, not time sequences or topics.

When I am in the zone, or I am sleeping on a problem, I am working by pattern recognition, it’s very powerful but also potentially flawed.

The trick is to have some other thought processes to go to if I am uncomfortable, that is I don’t recognise a pattern.

And also to have systems to check if I am missing something.

I am sure Eric has a lot more to tell us about this, it’s a very interesting topic.

[Benno von Humpback]

That intuitive mind works on pattern recognition Beau.

I am sure Eric has a lot more to tell us about this, it’s a very interesting topic.

Not that ignorance ever stopped me from holding forth, but I really don't know much about how the machines do it.

[kdh]

On the brain and how we think, I only know what I read in "Thinking, Fast and Slow" by Daniel Kahneman. Great book.

In school I lived down the road from where Emily Dickinson grew up and I used to walk to campus. All creativity was accomplished on those walks.

[BeauV]

On the brain and how we think, I only know what I read in "Thinking, Fast and Slow" by Daniel Kahneman. Great book.

In school I lived down the road from where Emily Dickinson grew up and I used to walk to campus. All creativity was accomplished on those walks.

I used to listen to books on the iPhone when making long drives and listening to the news in long-form (EG: ProPublica). Now, I try to spend an hour a day without any external input at all - complete silence. I walk, drive, bike, sail, something that takes a little low-level thinking but no external input of complex concepts. I've found it quite useful.

[Rob McAlpine]

A lot of days I walk to and from work in the morning - 2 miles each way, let's me sort out my thoughts for the morning's work.

My son doesn't use his car at college except for long trips. He walks everywhere. He's always been walk and think person.

[Jamie]

Everyday these days I get to walk to work along the intracoastal. It’s a good view now that it’s no longer hotter than the center of the sun.

[kdh]

I watched this NOVA on self driving cars last night. There was a lot of skepticism from those working on the problem about level 5 fully automated self driving in the wild happening any time soon. Way too many "edge cases" or unusual examples for neural nets to accommodate.

For controlled environments like highways we're already there.

[TheOffice]

I’ve never considered 128 a controlled environment! :D

[kdh]

I’ve never considered 128 a controlled environment!

Joel, you have an ex-wife in Concord or is that someone else? Did you grow up around here?

All who have know not to refer to 128 as "Rt 95" nor to refer to our main toll highway as "Rt 90" as those originally from New Jersey tend to do. It's always "the Pike." Going into Boston is always "going into town." The "city" is New York City.

128 is not that bad, just be prepared to stop!

[TheOffice]

Keith,

Grew up in Cranston. Close enough to fear 128!

[Benno von Humpback]

I’ve never considered 128 a controlled environment!

Joel, you have an ex-wife in Concord or is that someone else?

I'll take a look if no one claims her.

[TheOffice]

my ex is in Bethesda and is still bat-shit crazy. Look elsewhere!

[BeauV]

Interesting article in the WSJ on how the automakers who don't build their own batteries are having a tough time. It's behind a paywall, so here's the stuff I thought was interesting. Pretty clear that Tesla has a serious strategic advantage in building its own batteries.

SHANGHAI—A little-known Chinese company has become the world’s biggest maker of electric vehicle batteries.

Beijing engineered a scenario that didn’t give the world much choice.

China is by far the biggest EV market, and to boost its standing in the fast-growing industry, China began pressuring foreign auto makers to use locally-made batteries in the country several years ago. One company—Contemporary Amperex Technology Ltd., known as CATL—was the only shop capable of producing them at scale.

Auto makers weren’t pleased, but they fell in line. During a visit to CATL headquarters in 2017, three Daimler AG executives displayed their irritation shortly after the meeting started, recalled Jiang Lingfeng, then a CATL project manager who prepared a technical briefing for the visitors.

One Daimler executive cut off his briefing, said Mr. Jiang. “We’re not interested,” the executive said, according to Mr. Jiang. “The only reason we’re here is that we have no choice, so let’s just talk about the price.”

Article here

[Ajax]

Is Musk ever going to show us the new pickup truck? Depending on the final specs, that is a real contender for my next vehicle in a few years.

[TheOffice]

I thought the reveal is this month. Production date TBA (and missed)