GoogleX projects can 'graduate' and become their own companies that issue their own stock.

Employees of Google that work on GoogleX projects that graduate no longer technically work for Google, but for the new company and they no longer receive Google shares, but new company shares (in this case Waymo, Loon is another example [0]).

This forces the company and employee incentives to be aligned since the value of the employee equity is tied to the success of the specific GoogleX project and not Google itself.

When a company graduates if a lot of employees don't want to stay on the project and choose to stay at Google, that itself might be a bad sign (though it could also just be employee risk preference).

I think companies graduate when they're less pure R&D and closer to marketability. If they can't survive on their own at that point it may be because there isn't a market, the timing isn't right, or any other reason that can cause a startup to fail.

This system prevents Google from pouring money into something that's never tested in an environment where it can actually fail. It puts moonshot projects in an environment where they're truly tested and don't just become places where money is spent without any ability to measure success.

It's my understanding that graduated companies have their own hiring methods and can make their own independent decisions in general. They would also need to raise their own money.

[0]: https://loon.com/

It seems like a slightly bizarre system: have any of the "graduated" companies actually been successful by conventional (i.e. profitable) metrics?

I don’t know, but even if that’s the case it would seem like it’s working as intended?

It’s better to know if something will not succeed and you can’t easily tell without testing in the market.

The incentive alignment with equity seems like the most critical piece and it has the bonus of rewarding employees for taking on the extra risk.

This lets you spin out big ideas without spending all of your money - a little like running internal VC (one big success could make up for all of the other failures).

I think at this level capital is not really the main constraint. It's attracting talent and creating a culture/environment where success is possible, even then that's just a necessary but insufficient prerequisite for success among all of the other non-capital reasons a startup might not succeed.

I don't think it was a GoogleX project, but Chronicle was spun out from Google as an independent company. It grew well, and seemed ready to succeed on its own.

Google then re-acquired it.

Speculation, but perhaps Google was concerned about it getting acquired by a competitor as it used a lot of core Google search technology. Or, perhaps Google knew that the acquisition cost would spiral upwards if they waited much longer. Or, perhaps it just proved out the hypothesis that the product was viable, so better to bring it back in-house.

But I would call it successful by conventional metrics: it found product-market fit, had good customer traction, and was acquired at a decent (though not crazy) valuation.

This is the scenario I expect, Waymo gets "spun out" and does its R&D on someone else's nickel and Google doesn't have to report the losses, then if it takes off, Google re-acquires it in a sweetheart deal and reaps all the goodness.

From my experience working there, this is the sort of concept they would be attracted to. Google gets a benefit from either scenario.

From the counter party part of this deal, the "investors" either lose their money, or they get a fixed amount of 'upside' when Google re-acquires.

Or maybe they know Waymo is there to burn through cash without doing anything useful (like any other “research” project at any place that has infinite stream of money from oil or, let’s say, ads) and expect it to be overtaken by competition and want to get out.

All research projects from Google (and similar companies with "infinite stream of money") aren't doing anything useful? Seems overly cynical.

They're at least responsible for a large chunk of ML research, which is directly to directly improve their products (Maps, Photos, YouTube, etc.).

A common practice is to perform market research by expressing interest in acquiring competitors and then having them provide technical information to show acquisition worthiness. It’s all supposedly under NDA but the reality is that the big tech companies can easily sue you into oblivion and your lawsuit isn’t going anywhere. If you even try, welcome to the blacklist. Also the lawyers fees on your side are enough to put you out of business.

They are fishing for understanding of the market and sometimes general approaches or parameters of possibility. You are providing free insight. It’s the same when you pitch a VC. You just give them more and more free insight. Hopefully they provide value in return.

If this is merely a scheme for Google to sweep the losses under the carpet, what's in it for the investors who'll be left holding the bag? And mind you, in order for the losses to go away, the value of the company at the time it's spun out needs to exceed the billions of dollars already invested in it over the span of 12 years. (Including $120m paid out in bonuses to Anthony Levandowski alone.)

I wasn't trying to say it is "merely" a scheme. Its sort of like shorting some stock and at the same time buying a call option for twice that many shares above the price that you shorted it for.

The short part gives you some cash now and if the stock drops even further, you can cover the short and keep the profit. But if the stock goes up, just when there is a margin call you've got an "in the money" call option to cover it. So if it is shooting up you exercise the call, cover the short with half the shares, and profit when the other half keep shooting up.

Anyone can nominally do this on any stock, but there are tax advantages to the company that does it with one of their own subsidiaries.

What you're desrcbing is fiduciaries violating their responsbilities to shareholders.

I am sure that at least 50.1% of voting shares explicitly approve any crazy scheme Alphabet comes up with.

That has been their get-out-of-jail-free card since, well forever.

50.1% doesn't magically absolve you of the responsibility to maximize shareholder value. They could vote to sell to Alphabet for $1 - it would result in a massive lawsuit.

Yes, people can and will sue for any reason. But in this case this is how the suit would likely go ...

plaintiff: "Your honor these share holders are suing the company for violating its fiduciary duty."

company; "Your honor, we know that some share holders may not always agree with the majority, but we made sure that over 50% of the share holders were on board with every decision we've made. We move to dismiss."

Judge: "You have documented that the majority agreed?"

company: "Yes your honor"

Judge: "And every shareholder has access to the bylaws of the company which state in clear and unequivocal terms that all decisions will be decided by a simple majority vote?"

company: "Yes your honor."

"case dismissed."

Chronicle had two businesses under one roof.

One is VirusTotal, which is super important to the security community but I don't think generates enough revenue to live on its own.

The other is a strange "Splunk Lite" offering that as far as I could tell the main selling point was it was way cheaper because Google gave them free/discounted storage. The search was terrible. It didn't highlight important things or hide the mundane. When I saw a demo it didn't even support IPv6 yet.

But I guess every moonshot factory has to have its Challenger disaster to learn a few lessons.

Chronicle actually busted pretty bad! Have a friend who worked there and he described it as a nightmare, and not the productive kind. https://www.vice.com/en_us/article/9kej3e/chronicle-is-dead-...

sounds like a spin-out/spin-in model. you do these primarily to create special compensation packages.

the inclusion of a non-revocable license to core technology is a common part of such structures because it gives the people who take the risk some proof against management changes that otherwise sink such deals.

These guys did pretty well for themselves:

https://en.m.wikipedia.org/wiki/Pokémon_Go

Niantic was spun off from Google.

It was also acquired by Google, so it might not be the best example.

No, Niantic began as a "startup within Google". It wasn't an acquisition.

You're right, I was confusing the fact that the founder came in with an acquisition.

It's my understanding is that Google X is a marketing / PR platform for Google. Google X doesn't ever need to actually produce anything or create any real-world value because their value is in the PR they generate for Google - keeping them positioned through press as a technology company vs. the largest advertising company in the world. Many of the Google X projects are acquisitions/acquihires purchased to continue the narrative. It's why they never ship products and can't get anything out the door. If this works - it'll be one of the first after billions invested. They've done an excellent job I would say.

It was more like a playground for Sergey while he was still involved. I’m not sure if he voluntary chose to do it or not.

This is the main topic in The Innovators Dilemma, it's a pretty interesting book if you like this sort of graduation of companies

It makes sense from a focus point of view. Rather than Google/Alphabet pissing away their profits on ventures outside their core competence, spin them off and leave them for investors and managers that want to focus on self driving cars or whatever. It's hard enough to do one thing well.

Alphabet holding structure first and foremost is there to increase stock price. It’s specifically structured to separate and downsize risks. One of the investor’s concern is uncontrollable pouring of money into moonshots. Making them separate companies creates some transparency and illusion of accountability.

>> Making them separate companies creates some transparency and illusion of accountability.

I would not be so dismissive of it -- there are good reasons to have separate companies. You can offload some risk by getting co-venture (as in Waymo's case.) More importantly, you can have other critical partners be co-invested with skin in the game. In the case of Waymo, if I were trying to raise funding, I might get money from P&C Insurance companies -- it could help align great future partnerships that make 1+1=3.

Interesting! I had no idea.

You can get more info about this by reading up on GoogleX, but more importantly the articles that came out when Alphabet was formed. Those graduated companies would need a parent company, and that company was Alphabet. It was done to separate these graduated companies from Google's name (and accounting depts).

Alphabet wants Waymo to become a company that can make it on its own steam, rather than be dependent on the Google revenue stream. For the first several years of Waymo's life they did exactly what your describing, but now they're at a point where they want to "push it out of the nest" a little bit. Sharing the profits is probably not much of a concern, if Waymo doesn't manage to become self sustaining there won't be any profits to split, so it makes sense to optimize for that.

Indeed, and when it kills someone, nothing to do with Google.

But it would have everything to do with Waymo. Clearly this is something they’re thinking about, and one hopes they learned a lot from Uber’s accident in AZ.

I think if anything, it causes the bar to be super duper high to fully launch, which is why I’m very pessimistic on Level 5 driving happening anytime in the near future. AGI is not solved, not even remotely, and if death is the consequence, Waymo will probably dissolve before launch.

Are humans AGI? They kill 35k on US roads alone each year. If Level 5 means 0 deaths per year, then it will never happen until the singularity. It is quite possible America's safety first attidue will let 35k people die a year because automatic cars would kill hundreds per year, but I hope not.

I wonder if all this AI to build self driving cars should have gone into AI that simply saves lives a lot more easily: - detect if the driver is insane, drunk, swerving (tired?), stolen. - detect baby in the backseat parked, and temp rising?

If all the AI did was decide you're unfit to drive, that it take over for a few minutes to get to the next stop and pull over and call the cops on you, we might save more lives by that alone.

Automatic crash avoidance is already a thing that's steadily rolling out to more cars, and improving over time.

For stopping people from driving, you'd have to solve the social issues there first somehow. "Our car stops you from driving if it thinks you're incompetent" isn't something that sells more cars. Maybe the government could mandate it, and that's the government's responsibility.

> "Our car stops you from driving if it thinks you're incompetent" isn't something that sells more cars.

I could see something like that as part of a “supervised driver” safety feature. Like maybe you enable it when your kids use the car or if you rent the car out to strangers.

The AI that you talk about should be used in traffic lights. They should affix cameras to every single traffic light and have AI figure out the traffic flows. Then they can make traffic much more efficient throughout an entire city. It should be able to track a car through its entire journey through license plate recognition, and then adjust traffic throughout the entire city so that it's the most efficient for the amount of traffic.

So basically a super high tech car breathalyzer that uses some black box algorithm to decide if I'm drunk and call the police on me? The self driving car seems a little bit cooler.

Why not just install the existing breathalyzers in every car right now?

Yes but who am I going to go after if another car hits me if its autonomous? The manufacturer which is a 10 year plus lawsuit taking up a large portion of my life. I definitely think keeping the responsibility to the person driving the vehicle is the only reason the death toll is somewhat acceptable to society.

This seems backwards to me. If you get hit by Google, you’ll wake up in a hospital bed full of flowers from ambulance-chasing lawyers who are salivating over the chance to represent you so they can get in on that sweet settlement money.

Hit by a human, decent chance they flee and if you track them down, it’ll probably be some uninsured drunk with a negative net worth.

I was assuming after widespread adoption, some friendly obscurely worded legislation and the current supreme court.

You'd probably go after the cars insurer just the same whether it's driven by a human or a computer.

For reference, 35k deaths in the USA is about 1 death every 85 million miles driven. In comparison, Waymo has supposedly driven about 20 million miles. It's good that Waymo has had no deaths so far, because otherwise it would likely be far less safe than human drivers if there were any.

Not sure why this is downvoted.

This definitely would have been a consideration at some point along the journey.

You're naive if you think it wasn't.

I downvoted because it's the sort of flippant comment that comes up in every self-driving car discussion, and there's lots more compelling reasons that google might want to spin off waymo that a fear of bad PR when a car kills somebody. (mostly because that's how the Google X projects have always worked)

Of course a waymo car will kill somebody at some point. Everybody knows it's going to happen, and there's no way google/alphabet hasn't essentially priced in that risk. If they were afraid of killing people, they wouldn't have gotten into cars in the first place.

They are right to be afraid, but can balance risk and reward and not paralyzed by the fact someone will get killed.

Expectations are different on HN than on other sites.

https://news.ycombinator.com/newsguidelines.html

External investments can prove that Waymo was able to convince its own value outside of Alphabet, which might be a significant signal for future investors.

Alphabet can (and will likely) invest $xB more money for initial business ramp-up but if you want to create the next Google probably money from their own pocket might not be enough. This is particularly important as they need to heavily invest into physical assets with very high upfront costs.

FYI, Uber has 10 millions of drivers across the world and if you want to match that number, you're gonna need $100B even with super conservative assumption of $10k per car (and the actual cost will likely be much higher). Doing something like this by Alphabet alone is extremely risky given that the technology is still in a very young stage.

That assumes they want to go 0-100 in an instant, instead of building out a fleet as a region becomes available. Regulation is still a major hurdle.

Cars will quickly start paying off, even if each car only pulls in 10 rides of $10 each every day, it'll probably pay for itself in a year.

Diversify the risk.

Even founders who can self fund, raise money from investors as you don't want to own all the risk in any risky endeavor.

A company has to be able to operate on its own. Look at BRK; they don't fund every last bit of debt for all of their holdings. They will pull profits from company A even if A needs money to e.g. invest in operations if they can get credit at a decent rate. They may put that money in company B which has more growth potential and is not in as good a position to take on debt to do so.

From the last time this story was posted I recall that Alphabet was only offering Waymo investment opportunities to automakers. This would be an important aspect of fostering strategic partnerships, since Waymo is probably not planning to get deep into the car manufacturing business.

That’s a good point. But maybe they don’t have 100% confidence so they want to share the risk. Or maybe alphabet doesn’t see the self driving VEHICLE as their product. Maybe they see the consumer data as the product and want someone else to pay for the effort to collect it.

My take is that there are two interests in self driving cars for Google Proper (i.e. Search, Maps, Youtube, News, Ads).

* Free version of Uber/Lyft based on a Google ad/maps referral as described by this patent [1]

* Free up peoples commute times so there eyeballs are free to use Google more (watch Youtube, use Google Search, read Google News). Perhaps even access free in-Waymo wifi that only allows Google domain access.

They have monopolies in these two areas and therefore neither requires 100% ownership in Waymo.

[1] https://techcrunch.com/2014/01/23/google-awarded-patent-for-...

Bringing a car to market is going to require a lot of partnerships with a lot of companies, and google might prefer that they have skin in the game.

> Can anyone explain why does Waymo need to raise any money? Doesn't Alphabet have practically infinite pockets? If they are confident of Waymo, why would they want to share the profits with anyone else ever?

It's not just about getting money, it's also about getting money from the right people. Investors with the pull/clout/insight/etc they need. Investors who know people or who know people who know people. Perhaps people who have political connections or regulatory access. People with insight in the automotive industry. People with ties to media for favorable coverage. It's about getting money from people in position to help them. It's like a movie production that bring in producers who has connections to local governments that can expedite the process of shutting down a bridge or highway to film on for a day.

Alphabet has "limitless" amount of money but there is also a "limitless" amount of money outside of alphabet wanting to buy into waymo. So for waymo, it's really not about getting money but getting the right type of investors. It's a two way street. Investors want to buy into waymo and waymo wants to buy into investors.

I think also Google share holder probably wont stomach unlimited investment in this. It allows Google stock growth to not be hampered by this investment

How can such a small investment hamper Google stock growth. This might be even less than the average daily volatility in the stock price.

Depends on the shareholder, doesn't it? Some might buy Alphabet because of Waymo.

It aligns other companies' interests with theirs.

Investors could start getting angry if they feel that Google is blowing large amounts of cash on endeavors that continue to fail. By doing it this way they get a major stake in the potential success of their spin-offs, but they are able to continue delivering profits rather than sinking cash in one black-hole big idea after another. It's probably a matter of walking the line between short-term value of the already successful business versus long-term but riskier value of the potential moonshots.

All of the reasons other commentors gave, and also to show it has value. Think of it like getting a credit card while you can still technically get money from your parents (weak analogy).

If you can get money from someone else it shows that it has value in the marketplace generally.

If I were one of those new investors, I'd ask the same question. Alphabet has been involved in Waymo for 12 years. Why does Alphabet feel that more cash is a better investment than more shares in Waymo? What does Alphabet know that I don't?

Maybe they want to diversify risk.

"Raising money" means people are buying in. So, I'd ask the opposite. Not why they need money, but who feels the desire to invest.

Risk transfer.

Fossuser gave the best answer thus far to your question. [0]

A few confounders I haven't seen though:

1) Waymo is its own entity, but its brand is linked to Google/Alphabet. A reader of this article will be clear on that by the time they have gotten through the sub-headline.

Any success or catastrophic failure of Waymo/self-driving will have Google spoken of in the same breath as Waymo.

2) Capital Allocation & Diversification - Google had $119 Billion of cash on hand at the end of FY2019 earning no return. More than any other company.

- Alphabet is not Berkshire Hathaway waiting in the wings to deploy its capital in a downturn or buy up unrelated businesses.

- Diversification is a fine strategy, but when you have this much capital to allocate—and ostensibly Alphabet has this much because they don't know what to do with it—not fully funding Waymo itself doesn't make sense.

3) The Waymo blog post about this that Rutledge linked below gives a possible alternative [1]

> “Today, we're expanding that team, adding financial investors and important strategic partners who bring decades of experience investing in and supporting successful technology companies building transformative products. With this injection of capital and business acumen, alongside Alphabet, we’ll deepen our investment in our people, our technology, and our operations, all in support of the deployment of the Waymo Driver around the world.”

Building technology is hard, but building out a car company that can produce vehicles on any scale is also hard. Tesla and Elon now know this.

So maybe by opening up to outside investment they allow strategic partners to share in the risk and outsized reward of success.

Curiously though, only one of the six outside investors listed seem to be "strategic." That would be Magna International, "a leading global automotive supplier."

The other investors are:

- AutoNation: a used/new car buying/selling website

- Andreessen Horowitz: VC firm

- Canada Pension Plan Investment Board

- Silver Lake: private equity

- Mubadala Investment Company: United Arab Emirates sovereign wealth fund

From the outside looking in, I have further questions as to how those are "strategic" investments that could help develop and/or bring Waymo cars to market.

tl;dr Alphabet/Google has so much capital it needs to deploy not using it to fully fund Waymo is curious. Waymo is not a distinct brand from Google/Alphabet. It's not readily apparent how six of the seven outside investors chosen for the "strategic" investment add value beyond capital.

[0] https://news.ycombinator.com/item?id=23157082

[1] https://news.ycombinator.com/item?id=23156434

- Canada Pension Plan Investment Board

- Mubadala Investment Company: United Arab Emirates sovereign wealth fund

Couldn't these also be considered strategic in the regional/political/regulatory sense? Access to people with strong connections having skin in the game.

Thanks - I'd argue it's less about brand separation and more about actually testing viability and forcing a culture that considers survival when making decisions.

The ability to raise money is itself a test - if you can't get third parties to invest then that's not a great sign. Google's judgement may be impaired by being so close to it.

They could easily self-fund, but then they wouldn't get that signal. Forcing a company to survive out of the nest will influence the decisions the stakeholders make, they can't relax knowing that Google will just fund things no matter what, they have to be focused on shipping something people want.

At least that would be my guess.

It's a clever way to try and prevent the lethargic energy that comes from being a big company along with the inability for large companies to innovate outside their narrow domain (often even within it).

That all makes a lot of sense especially in light of the lack of urgency to bring things to market that has been reported about Google X throughout its history.

Because google finally realized it's impossible and so they're selling this turd to the greater fool.

Very simple: Google shareholders don't want to be tied up to that extent in self-driving cars.

By keeping Waymo separate, shareholders can make their own choice as to how much they want to invest in search/advertising vs. self-driving cars. They're drastically different risk/reward profiles.

Obviously this isn't something a company usually does, but that's because either a division is small, or it has significant links to the rest of the company, or there's a clear joint strategy.

But Waymo is expected to be huge, and there's really essentially zero link to the ads/search/cloud/media that is Google's bread and butter.

I'm a Google shareholder. If Waymo is a good business I have no problem with Google holding all of the equity.

I think you answered your own question: if they are confident of Waymo

As I said a couple months ago, the "Waymo" name is a sign. If they really believed in the product, it would be called "Google Self-Driving Cars".

https://news.ycombinator.com/item?id=22632588

Different brands can make sense though. Youtube isn't called Google Video and Blogger isn't Google Blogs.

Google Video was an actual product, competing with Youtube. When they decided to acquire the much more successful Youtube, its brand already had a lot of value and recognition, so it survived.

https://en.wikipedia.org/wiki/Google_Video

Similar, on a smaller scale, for Blogger. Google acquired it.

Yet they applied the google name to nest? Android remains without a google moniker (other issues there).

Point being that google has quite a few inconsistencies in their naming (which is a thing in itself) and their not applying Google brand to Waymo at this point is any indication.

Android was also acquired.

Google Video existed and was shut down after Google acquired YouTube:

https://en.wikipedia.org/wiki/Google_Video

Blogger was also an acquisition (from the future founder of Twitter), and it already had a strong brand.

Whereas in 2011 there was a lot of marketing tying self-driving cars <=> Google.

There could be other reasons for the separate brand, like managing bad PR from a crash, but I would also classify that under "don't have confidence".

In contrast Tesla seems to be staking their brand on self-driving. I feel like this could backfire in the next 5-10 years, but Elon will probably make some other advance/launch to distract from that. Self-driving has already fallen behind Tesla's claims, but he can keep the media busy enough with other stories.

It last several years beyond the acquisition of YouTube. In college, I used to be something of a history buff and remember watching and enjoying many old documentaries on WW2 and Cold War stuff on Google Videos. According to the Wiki article, they only turned off uploading to Google Videos in 2011, almost 6 years after acquiring YT.

Everyone is pointing out that those two are acquisitions, but they're missing Verily, Calico, Jigsaw, Wing, Chronicle, etc.

It's par the course for X graduates to get their own brand.

> Youtube isn't called Google Video

Google Video is a thing, it was Google's competitor to YouTube before Google bought YouTube. (And YouTube was already called YouTube before Google bought it.)

Youtube was an acquisition.

Lots of conglomerates have sub brands to avoid antitrust attention. Also maybe they don’t want consumers to think the cars are watching them.

“We will drive you to your requested shopping destination after this brief targeted advertisement for laundry detergent.”

Or the car will take a detour past certain billboards

Too crass. The billboards will display video ads optimized for the occupants passing by.

Hyperbole clearly. Feel free to not use the taxi if you care. But of course you'd want to rant on HN

Even if you thought Waymo technology was amazing and would be a huge success, there is a non-trivial risk that a self-driving car could kill people and turn into a PR nightmare. Thus it's totally reasonable that Google would want to insulate themselves against that risk considering that they own one of the world's most valuable brands in their search business. Also, it might make sense to have a real privacy wall between Google and a potential fleet of millions of cars outfitted with multiple cameras and sensors. The only way to do that effectively is to have a separate company do it.

People are going to die in Waymo cars, and sometimes we'll wonder if the accident was avoidable and if the car made the wrong choice leading to someone's death.

It'll look bad on the brand when the news is reporting how, "Google self-driving car kills family en route to Disneyland".

I'm not saying self-driving cars are dangerous, but it's just the numbers. People will die and self-driving car deaths will be shocking news and make headlines for a long time. I wouldn't want my brand name anywhere near them.

Sure, just look at Rolls Royce. A related company with that brand name makes jet engines, which of course are thus involved in tragic and headline-making air accidents.

As a result Rolls Royce cars are... huh, no, people actually nod right along, that makes sense, Rolls Royce jet engines, I would want a reliable, high quality jet engine and that's what I associate with this car brand. Most people have no idea it isn't even the same company.

News outlets will connect Google to such deaths regardless. "Google car kills two" the headline will say, and Google's PR person will know better than to insist "Actually it should say Waymo not Google".

It looks bad if it becomes a trend. "Man dies in car crash" isn't going to destroy your brand. Ask every single car company in the world. Even Volvo, whose brand is specifically all about safety. It requires a trend, which is what hurt Boeing recently, bad news after bad news without respite, that'll do it. But there's no reason to think Waymo will ship something dangerous enough to cause such a trend.

I don't think it's a fair comparison.

There's a difference between someone crashing their own Volvo, and someone sitting in the backseat of a Waymo that misinterprets the lane markers and crashes into a barrier. One is not news, one is front page news.

I guarantee the first Waymo death is going to be publicized everywhere. It doesn't need to be a trend, and as I said, it doesn't mean the cars are more dangerous than human drivers. However, it's going to be news. There are going to be all sorts of moral debates when an algorithm decides to drive over a child instead of turning into an oncoming car. People will want answers. How does Waymo rank the value of different lives? I imagine it'll be news for the next decade until there are thousands of deaths and it becomes normal.

It's an industry that's going to be full of "firsts" and that's going to be the news. Waymo drives over dog while auto-piloting to a parking lot. Waymo mistakes grocery cart for stroller and swerves into elderly man. Waymo kills cyclist when poor weather disrupts censors. It doesn't matter if they ship something safe. Get enough cars on the road and these things will happen and people will be talking about it.

> There are going to be all sorts of moral debates when an algorithm decides to drive over a child instead of turning into an oncoming car.

No. That's a trolley problem. It's an interesting intellectual exercise, you can maybe win a debate team trophy for your rousing defence of one choice or the other - but these moral decisions aren't actually what drivers do whether they're humans or a powerful AI.

People keep acting as though this is an unprecedented situation and invoking weird moral beliefs about thinking machines, when it's actually utterly routine. Let's try another exercise:

How many headlines have you read about a specific brand of elevator decapitating a child? Is it none? Do you see anybody pushing for the big elevator manufacturers to have to reveal how they "rank the value of different lives"? No?

That's not because nobody dies this way, it's because we say oh that's just a machine obviously if things go wrong you can get seriously injured and the machine doesn't know if you're a nun or a basketball champion it isn't trying to kill/ not kill anybody in particular it's just a machine.

Humans are often tempted to try "escape manoeuvres" and these almost invariably go wrong, we don't teach machines to try such manoeuvres because the machines are trained based on real performance data not someone's model of themselves as an immortal superhero.

One of the first Waymo crashes was somebody trying an escape manoeuvre. They found themselves in a potential collision so rather than the correct thing (brake to reduce speed, hit the thing you're colliding with because it's too close) they tried an abrupt swerve, lost control of course, crossed a median and smashed into the unrelated oncoming Waymo car at high speed writing off both vehicles. Humans do stuff like that, you can try training them not to but they won't listen. But the machines do not have that problem, so less "Should I kill the nun, the pregnant woman or the Olympic champion?" and more "Despite maximum braking effort a collision has become inevitable. Preparing safety systems for impact".