I watched a panel discussion on self driving cars in 2015 with several legit experts in the technology (I believe Sebastian Thrun was one of them). There were also some CEOs. The experts all said there’s no way the tech would deploy before 2025, potentially later. The CEOs were saying 2018 or 2019.
The experts had a clear view way back then, the CEOs just don’t want to listen.
This is always a game of boundary-setting. In one way that's true, in another way Waymo is 21+ years behind[1]. People have been setting up 'particular vehicles' to navigate 'particular areas' for decades. If the Waymo "self driving car" is an expansion of older site-specific tech, then there's nothing new under the sun. If their car is can be put anywhere then it just being able to drive in Tempe isn't proof of that.
IMO the truth is more with the experts. Each new location seems to require a lot of tuning to get right and function in a way the company is happy with. If a "self driving car" is one we can drop anywhere and have it drive we are still waiting on that.
By the way, Particular Vehicles have been safely serving Particular Areas for decades: airport driverless shuttles. Slowly expanding from there is the right way to go, not "New York to Palo Alto by 2017", as Musk promised.
Yah, without any evidence to back it up my personal suspicion is that we'll arrive at a local maxima of "full self driving on highways and instrumented roads" in ~20 years. I think the area has a lot of potential, but so much of the hype (and stock price) is tied up with "a car that can drive better than a human everywhere" which seems impossible for anyone to produce.
Just like with IoT we'll eventually arrive at a boring, useful state...it will just take a while.
Oh, come on, these kinds of false analogies don't help to elucidate, only to confuse.
There is a universe of difference between something like ParkShuttle (its own right of way, using magnets in the road to detect position) and modern autonomous vehicles. Saying modern autonomous vehicles have environmental constraints is valid, saying that makes then no different than a "people mover on wheels" is not.
The point is that "deployed to a limited area" is not what people are thinking when they talk about a "self driving car." Of course ParkShuttle isn't the same as Waymo (or any other modern self-driving car) - but the question is how far we've come!
Are you claiming Waymo is ready to drive anywhere in the world? I do not think that's Waymo's position. So where are we? Cut through the rhetoric and tell it like it is - or join the rest of us who are speculating from the sidelines with incomplete information.
I'd argue even waymo is still far away from FSD. Driverless cars on a restricted set of roads with a remote operator monitoring things (and the ability to quickly resolve issues), is nowhere close to what I (and the general public I would argue) understands as fully self driving.
IDK, the Waymo cars that drive around SF meet my definition of full self driving. They handle a ton of weird traffic patterns, pedestrians, pickup zones, and very dense urban environments much better than I think you realize. Unless you live here, you probably aren't aware of their capabilities.
The goal is to replace a human driver in every situation where a human driver could or would drive, with an equivalent or (ideally) better safety record.
Are you suggesting that Waymo is there? I don't think the evidence would support that. Even if we relax that goal a bit so the self-driving car can disengage and refuse to drive in, say, the most difficult 20% of situations, I don't think we can say Waymo is there, either.
Sure but the panel discussion was talking about consumer release of self driving cars, which really requires level 5 capabilities in a wide variety of conditions, not just ideal weather.
Not many people, even experts, predicted the success of ChatGPT even five years ago - and most have moved their timeline for AGI up significantly because of its release.
Right but "hard to predict" does not mean "it will happen whenever I want it to happen". You can't look at a very hard problem, where a good sample of the world's experts say it will take more than 10 years to solve, and just say "I bet we can do it in 4 years" just because you want that to be true.
Do you have sources on those predictions moving up “significantly” because of ChatGPT? Predicting words that sound good together, as a result of a prompt, based on a pre trained set of data is vastly different from AGI.
I can't find the exact article I read it in at the moment, but ChatGPT was definitely a turning point where a bunch of people reeled in their predictions by a decade.
(The median expert prediction was between 2040-2060 previously, now quite a few people have moved that timeline to before 2035).
Experts are wrong all the time. The experts said that solar prices would come down but were off by nearly an order of magnitude in how long it took.
With something this hard it's extremely tough to tell for sure when it will happen and often progress goes in chunks where it seems like it's going to happen but then doesnt.
Many experts predicted price declines with pretty fair accuracy. The large groups of experts that were making these forecasts for purposes of global planning for mitigation of climate change were often intentionally conservative because the danger of planning with that assumption outweighs anything you lose by making more conservative predictions.
Experts have a narrow, deep view while CEOs have a wide, shallow view.
If the expert works in a lab developing new experimental solar panels they probably don’t see a clear path to mass production.
The CEO might know another manufacturing expert that does see a path to production and have enough high level understanding to know the methods are compatible.
In this case yes, but in this case we have to assume the research side has achieved it's goal (its proven a material exists with the desired properties), now it is a manufacturing problem. If the lab can't produce the material, the manufacturing lines have nothing to manufacture with. And that seems to be the case with FSD. CEOs make wild claims, but the tech isn't there. The material has not been proven to exist in a lab with the desired properties.
Like a CEO saying, I have a material that can protect wearers from nuclear fusion blasts and it will be on the market in 6 months, but the experts in the field have yet to actually prove that material exists and create it in a lab.
But the point is that we are all equally capable of being wrong. Especially when we step outside our area of expertise. CEOs are just another type of expert but their domain is organization. We have to consider the source’s experience relative to the domain in question before we can decide if their prediction is trustworthy.
Both might be equally capable but the incentives are entirely different. An Engineer is motivated to deliver accurate predictions because that's their job. A CEO is not motivated to deliver accurate predictions because their main job is to hype their product and services.
In this case the actual CEO we all know we're talking about is a serial confabulist who publicly agrees with anti semitic conspiracy theories and tells his advertisers to go fuck themselves, so your abstract hypothetical arguments generously giving some unspecified billionaire CEO the benefit of the doubt don't hold any water.
He also does dubiously claim to be an expert in many fields, although despite the success of the engineers working for him, his greatest personal expertise appears to be sycophant creation — competence can only propel one so far.
Although one should never believe the ravings, useful work does surprisingly often nucleate around them, just not to the degree or with the speed promised.
The experts had a clear view way back then, the CEOs just don’t want to listen.