Whenever people ask me which company I think will win the self driving car race, I say Tesla.
And the reason is that they have more data.
And when it comes to training machines to do what humans do, more data is better than more software engineers.
Bloomberg has a good post on that today.
“If we have data, let’s look at data. If all we have are opinions, let’s go with mine.” – Jim Barksdale, former Netscape CEO”Not everything that can be counted counts, and not everything that counts can be counted.”Albert Einstein, Physicist
Jim Barksdale has some of the best business quotes in the universe.
“If you torture the data long enough, it will confess”. – Ronald Coase.
https://uploads.disquscdn.c… https://uploads.disquscdn.c…AI folks are (generally) not very good artists. They think they can measure the limbs of natural language and get to meaning, and that’s what they’ve done since AI as a discipline was founded.This is why the Natural Language Understanding problem is more likely to be solved by someone who’s an artist at heart, yet happens to also have artistry with the tools of science.
An economist (or accountant – take your pick) knows the cost of everything and the value of nothing.
A layman was watching a scientist perform an experiment. He asked the scientist: “What are you trying to prove by this experiment?” The scientist answered: “A scientist does not try to prove things. He tries to find out.” #objectivity #openmindedness- From an old issue of Reader’s Digest.It seems not many people have digested it, though.
It makes you wonder how far behind the other auto manufacturers really are, specifically the US ones. I think MB and BMW have something up their sleeves we can expect soon.
ErikSchwartz: comment aboveAs for data… Toyota (and other majors) sell more cars in a week than Tesla sells in a year. They can close that data collection gap VERY quickly.What independent player can these volume OEMs, with their huge data collection potential, turn to for a credible hardware/software catch-up operation ?
Good point. They need to come out with something soon.
Depends highly on how you sift the data you collect. I see a lot of people make data bend to their will. They don’t do true unbiased estimators or true hypothesis testing. They are too afraid to see their opinion debunked by data. Tesla is pretty cool. That coolness factor will take it a long way as well (branding)
Add cohort analysis to that list of good practices, too!
The issue with pointing to tesla as the data winner is that the type and depth of data they are collecting is very sparse vs. other tech companies, and their gathering conditions aren’t as controlled as it’s up to the user, where Uber can gather data on their own time.Wrote this (https://medium.com/frontier… : “It turns out that the majority of the data that Tesla gathers across those millions of miles isn’t actually as valuable as many believe. They log images every few frames and save video data/multiple frames only during accidents/failures. So maybe that data moat isn’t as big of a head start as many think.”
Well I may be wrong with that response. It is one of the reasons I wrote this post. To get smarter. And the post you linked to is awesome. So thanks!
Since Musk is inspired by Tesla and you made a plea for experimentation rather than building models to “fit the prediction” …This is what’s happening in data, economics & Natural Language AI: https://uploads.disquscdn.c…Scientific experimentation should be about seeking TRUTH and be anchored in practical realities.
I have always thought it ironic that Tesla runs their cars on direct current.
Ha!!!! That is a great irony
Lol, I did say elsewhere to Charlie that edge includes artistry and wit.
Actually, they don’t. The cars store energy as DC voltage, because that’s how batteries work. The motors are AC induction and therefore an inverter is used to convert DC to AC to drive them.
Yep, just like the other dozens of AI and data-related articles written everyday which are nothing more than bullshit.
That is a great well written piece. I love how you give A and B alternatives.What is so stark is seeing the $150B a year of commercial drivers wages that could evaporate. That is the nature of technology. 50 really smart engineers could actually add that much value (or you could say take away that many jobs)The only extra alternative I would give is on car ownership. The real question is does the car wear out from age or mileage.Let me give an analogy. The police around here assign each officer an individual car. Now they could have less cars if they put each car into a pool and had officers share. But they are going to wear out at the same rate, so you really don’t save (except if you count the cost of capital which right now is nil). The two huge added benefits are that the person driving the car really cares about taking care of it because it is theirs, and they can park it in front of their house (we love that). So while I don’t doubt that in an urban environment where parking is an issue people won’t have cars (mmm….parking is another thing that is going to get hurt) For us out in the country having a car in front of the house and not having one that is abused (if I was just renting what do I care about putting those bags of cement in the car) is a big benefit.
Thanks Philip! Agree that in rural areas, owners will still want their own vehicle. I do think that the # of vehicles owned could be reduced to 1 as it can drop you off at work, pick up your kids on its own, etc. -> In addition, if we think about how OEMs design cars today, they must balance price of vehicle vs. quality, because when they sell a car for $30k, that is the majority of $ they will ever see from that car…However if instead they can think of the vehicle with an LTV of $80k or $100k (due to ridesharing), they can build a similar vehicle with significantly higher quality parts, that will allow the car to put on more mileage and take more wear and tear on the internals.
Maybe ahead in data but mitigated by lack of single focus? given their other critical challenges for the company itself including getting to scale, skating the cash financing challenges and integrating Solar City?
Andreessen: “One brilliant person. Not 1,500 people.”https://uploads.disquscdn.c…* https://techcrunch.com/2016…Google, Facebook, Amazon, Apple, MS, IBMWatson have more data and more engineers and more $million AI researchers than anyone.Yet they haven’t been able to solve the Natural Language Understanding and “fake news” problem.It isn’t a matter of “They don’t have the will to do it because it affects $ ads.” It is that Computer Science hasn’t been able to think outside the box and invent the tools needed.
I wouldn’t be too quick to bet against George Hotz.”It ain’t bragging if you can back it up”- Muhammad AliGeorge Hotz may not be the eventual winner, but I’m willing to wager he will be there at the finish line.
It doesn’t make sense to bet against anyone who can do what would take 1,500 engineers at Google / Uber / Tesla to do.There’s that adage: “Two heads are better than one.”Well, is it better to have 1 Einstein or 1,000 ordinary physicists? It depends if the measure of success is the creation of new fields of physics or managing the power supply and engineering needs of a country.More is not necessarily better. We needed more people during the building of the Great Wall and the Pyramids.WhatsApp needed all of 55 people (32 engineers) to serve 900 million users and be acquired for $19 billion.So for every quote and maxim, there’s an alternative paradox. Lol.
Its the software engineers that figure out the algorithms that actually do the mining of the data to make it useful. Minor detail.
All the data in the world is only useful if you know what to do with it. And you can likely train an AI just as well with less raw, but better curated and annotated, data – so I wouldn’t rule out the competition. That said, I’m a tesla fan as well, looking forward to getting my model 3 in a few years 🙂
Lately I’ve been thinking that the winner in this space could be UBER do its enormous scale and amount of data. It’s true that UBER cars don’t have the technology that Tesla cars do and that their data might no be as solid, but that scale… god. Also, UBER has acquired Otto and started rolling out its self driving cars lately. Besides who the winner is, I think one of the key insights in this industry is about car ownership. Will this even matter in the future?
“Tesla has acquired Otto”I know you meant “Uber has acquired Otto”That portends well for them, assuming the other parts come together as they have for Tesla.
Damn yeah, you’re right, thanks for pointing it out! (I already edited it) I agree that it ”could” be good for them.
tesla magic. video doing the rounds todayhttps://twitter.com/l1ad/st…
United health care / Optum now processes about 1.5 billion health care claims a year… talk about a great place to innovating in right now!
CONTRIBUTORS:True wisdom is knowing what you don’t knowConfucius, Sayings of Confucius
Didn’t the Encyclopaedia Britannica have more data than Wikipedia ?
CONTRIBUTORS:Amazed that all the technology concentrated contributors forget history so quickly.There was a short time ago that Sony was creating a Blue Ray disk to replace DVD format and Toshiba was creating the HD DVD. Toshiba conceded in 2008 to the Blue Ray technology. But still produced a Blue Ray player. The one who wins no matter the leader is the consumer.
Data helps, vision is a must.Elon has vision and communications skills and a belief that technology will simply get invented as needed.I’m a huge believer in him.
Is Tesla a platform on which other businesses can be built?There are several companies building Tesla specific transportation companies.
There won’t be one winner. For there to be any winners there need to be standards. To get to standards car makers will have to collaborate (which kills the whole “winner” idea).As for data… Toyota (and other majors) sell more cars in a week than Tesla sells in a year. They can close that data collection gap VERY quickly.
also, while other OEMs may not have the autonomous vehicle data that tesla has, they’ve got a ton of other useful data in the form of telematics and maps. the scale of this data might make it more useful or sufficiently useful so tesla’s edge is not meaningful. and of course, TSLA still bleeds cash…..
It’s a worthy question though: IT has huge network effects, and smart cars are an IT-ization of cars.Network effects come from scale economies, skills, infrastructure, data and apps. I think much of that does apply to cars ?Maybe we’ll see something akin to what happened in PC and then Mobile: a vertically-integrated enclave (Apple) in a mostly OEM/layered market (Wintel, Android) ? There’s a key difference though in that Service and device sharing are probably key elements for self-driving cars, but barely register in PC and Mobile.
One may “win” the race but a bunch of the others will also cross the finish line.the established automobile industrial conglomerates are not about to hand over their power to Tesla. their political power is enormous. to have gotten away with distributing pollution machines on mass for decades is proof of that.
Money line of the postmore data is better than more software engineersBTW – Have my Tesla model 3 on order
I want to suggest something I think several people have hinted at already… I believe one of the core problems of web + mobile design currently, is that they do a poor job of representing the actual data around us.Many VCs, engineers, and data scientists believe otherwise. They see a world where we simply need to continue to increase the amount of data captured and analyzed, and our abilities to process it. But I think that’s a mistake, one that is borne out of dominant attitudes towards design. I’m not capable enough to break that down into technical interface terms, but Olia Lialina does a strong job of it here —> http://contemporary-home-co…*her work is fantastic in general, for anyone that hasn’t run across… http://art.teleportacia.org…In the simplest terms, to me that means: we have this entire web and data architecture that’s built on the idea that design means reducing barriers between digital and physical (instead of giving the transition space and time to exist). And then we end up saying things like, hey cool if we can just get those trillions of data points from all around us into a useable database or template, we could do so much more. What I’m trying to say is, that alone won’t help us with training machines to do what humans do. The distinction is important and it’s often ignored.
This past year we (Nexar) collected tens of millions of Miles of road data and my conclusion is that you should not be so fast at giving Tesla the reign. Here’s why:1. The key in training AVs is not any old data. It is data of collisions, near collisions and other dangerous situations. Millions of Miles just going up and down the 280 do not advance you that much. The key is to figuring out a way to efficiently collect these problematic events, which ironically, Tesla owners get into at a much lower rate than the overall global population.2. Different societies/geographies lend to different issues. You can’t take a car trained in Mountain View and throw it in Mumbai, or even Berlin, and expect it to smoothly sail. Variation and richness is key. To give one vivid example, we have in our database incidents with cows on the road in New Delhi. Hard to train for those in Mountain View or Pittsburgh.3. If all you’re collecting at the end of the day is IMU+imagery data, then using a $80K device to do that is really not economical. Better deploy a $100-$200 device on any car out there, or even better, just package it as an app and get to a $0 marginal cost. Extra benefit of an app is that worldwide distribution is frictionless and does not depend on importing expensive cars to your country. You will have to provide a lot of end-user value though..4. Finally, this is not about technology. It’s about economy. The reason other OEMs do not upload data en masse is the cost of data upload through cellular. Once the economic value of said data will become clearer and OTA updates (through Wifi) will be adopted, you’ll see massive fleets of vehicles start uploading data. And then it won’t be about 50-100K vehicles. It will be Millions.In general, though, I don’t think there will be one clear winner in the autonomous vehicle space. Google had a shot but they missed it. This problem will be solved simultaneously by several teams, and while Tesla may be the equivalent of the Mac of the 80’s, the OEM market will continue to be highly fragmented with other OEMs playing the role of the PC clones of that era. As an example, recommend looking at the great work done at Berkeley Deep Drive (http://bdd.berkeley.edu/), which brings together several leading OEMs and partners like Nexar to solve this problem together.
If I was a car manufacturer I would want car-driver apps to be a non-differentiated commodity component (pretty much like generic car tyres are now)So I would seek to agree a x-industry protocol such that my hardware can be both/either act as a data source and/or respond to driving instructionsI would seed development of a self-driving car-IP eco-system, lets face it the solutions needed are pretty generic controls (target a response from analog and digital i/o) plus smart pattern recognition algorithms to substitute awareness – which can be broken down into deep vertical domains and A/B tested or similar for an optimisation heuristicThis way i can continue to do what I am good at (building cars) and share the on-cost of closing out pretenders with proprietary solutions. I also have a good network of people used to dealing with regulatory frameworks and diverse global specifics.
Tesla’s data is certainly a valuable asset. And its head start. But Tesla only claims to have data on 1.3 billion driving miles, while estimates are that in the USA alone roughly 3 trillion miles are driven every year. Seems like its still at “jump ball”?
>And when it comes to training machines to do what humans do, more data is better than more software engineers.I would say both are needed. Erring in either direction – is erring.
Data is very important to Tesla’s success but I think their overall business ideology (reconstructing every new touch point experience in the personal auto industry) will eventually help drive out most direct competition and redistribute market wealth in their favor.
Data you can buy, borrow, steal.what to do with it, that’s a problem on its own
It’s hard to derive insight from data you don’t have.
Great comment and we’ll add edge also includes an amount of artistry and wit.https://www.youtube.com/wat…
I’m not utterly gung-ho on M. Musk. He seems to have lots of salesmanship above all. The “autopilot” name is borderline murderous.
+1 – also, I think the risk taking and creativity of a Musk is something we want to optimize for (data that supports not supplants)…rather than assuming massive amounts of linear, processed data will lead to wins.
Exactly.Even in art, there’s a difference between linear thinking …https://uploads.disquscdn.c…And the creativity of the Masters.https://uploads.disquscdn.c…The “data” (the facial features of a woman) is the same. Yet the interpretation and creativity are very very different.
The world needs more brilliant salespeople as much as it needs engineers if not more so.Great products don’t win on their own
Yes and no. Engineers deal with realities, whereas salesmen (I’m one) deal with perceptions, which opens a much wider door to manipulation and deceit… Trump is an example of a good salesman. Sales are amoral in a worse way that engineering is I think; and can be applied to a lot more subjects ?
#SteveJobs He did a lot of salesmanship along the way.
Yes and no back at ya.Agree that the possibility for ill is greaterBut so be itThe greatest lack in startups is building an market and acquiring customersSure we need brilliant engineersYou almost always find themBrilliant storytellers are so scarceBut there is nuance to this I agree
Sorry disagreeBoth are scarce.When one group takes the other for granted. That is what kills companies
Happy to be nudged to be more careful.Though honestly I’ve never met a smart saleperson who ever downtalked the people who make the product or support it.
The mark of a good salesperson is they buy engineers gifts for the holidaysIf you are running strong the salespeople love and laugh at the engineers jabsManagement is where both groups can heckle. See my comment the other day. The bet for our holiday party was that if people put $2k for our school in Cambodia in the hat I would ride a mechanical bull. I didn’t sign up for that and they said well once or twice a year you “write a check you can’t cash”. My term. Here is your turn.I was successful and much fun was had
Yup and in early enterprise software solutions sales is marketing and the most important focul point for building brand.Spent a half day with one of the best a few weeks ago and felt damn inspired by it.
One of the best.Also the person who said–somewhere–that the best CEOs are marketers who can manage to a P & L.
Guessing we’ve traded this link before, but re-reading just now and I feel like you’d enjoy another pass, too! — > http://contemporary-home-co…
There are many things you can learn from earlier entrants, but I’m not sure how you learn from the ML results of another company’s privately held data.The other factors you list are indeed important. I think it is unsurprising in this case that people focus so much attention on the quality of the software enabling hands off driving.
Exactly this.HP machines supposedly had all the data until Apple popped along with its portable devices.Different utility, design, business model, fabrication and mindsets.
Sales is like being a head coach. If you aren’t winning you are out.But I disagree with your premiseIf you don’t have good players you lose and you are out.If you lose the players in your locker room and they stop working…you are out.Why do you think Zeke Elliott bought every Offensive Lineman a John Deere Gator for Christmas?? They don’t block and he is out.A salesperson only cares about totals? She will be out. Yes Bobby Abreu only cared about stats. Out.
Theranos ?We retroactively endow successful people/companies with lots of qualities, forgetting in the process that a lot of failed people also qualified for the same qualities, at one time.Having a vision is great. Being able to articulate it and convince buyers/VCs even better. But what counts in the end is delivering on it, and the jury’s still out about that.Tesla does have a credible chance. But so did lots of eventually failed companies. And the game is about to change a lot. Also, there’s a huge disconnect between the Tesla hype and the reality, they’re a really minor car manufacturer, and not even leading in electric cars.
I don’t know how hard the problem is, so I can’t assess whether companies will converge that fast. I do know that ML benefits from truly huge amounts of data so that access to such data lakes is a competitive advantage. But perhaps this is a case where the problem is easier than it might appear. I doubt it, but I freely confess that I don’t know.Not killing people would indeed be helpful.
Thanks for sharing again.The pulley the Egyptians used to build their temples was as much a technology as today’s computers.Most problems in the world boil down to differences and dissonance gaps in definitions, unfortunately.
I have a system that’s so stealth it’s in the Mariana Trench, lol.
That’s were the salesmanship part bothers me, because these are utterly untrue statements, and this leads me to wondering what else Tesla+Musk are lying and cheating about:- “They’re leading in luxury cars in the US, including the gas burners. “. NO. https://www.yahoo.com/news/…- “And they’re leading in electric cars–sorry, who’s ahead in units?”. NO. http://www.forbes.com/sites…
The industry would be moving with or without Tesla.Tesla isn’t leading in electric car production (only in luxury car production, which might be why it gets so much top-down good press, and much of the luxury is perception and looks, the actual amenities aren’t luxurious).Tesla seems to have a lead in deployed drive assist, though insisting on calling it autopilot and letting it kill people is repugnant.They do have a lead in charging infrastructure in the US.And finally, their business is dependent on bailouts, subsidies and capital burn.Maybe one day it will become an admirable business. Right now it’s mostly a subsidies- and VC-fueled hype machine.
We agree. Don’t make numbers out