Self driving cars have massive network effects, just like any product dependent on machine learning. More users -> more data -> better product -> more users.
It's the same reason Google is invulnerable in web search.
Google isn't invulnerable in search because of a network effect, which is about your product becoming more valuable because you connect your users to your users (e.g., the Bell System or Facebook). Google is invulnerable in search because search has a very long tail, requiring massive investment to reach adequacy for most users. Google's advantage is economy of scale.
Self-driving cars definitely don't have a network effect. They also don't have the came kind of economy of scale as Google. A person who learns to drive in one, specific part of the US is able to drive, with minor additional learning, in any part of the US. Yes, there's some minimum size necessary to get the data for self-driving cars. But unlike with search, where the minimum size is generally "the whole web", the minimum size for self-driving cars is much smaller. That's why we're seeing many different companies work on it.
Network effects are not limited to products that connect users. A network effect is when more usage increases the value of the product [1]. More Google users means more click data, which means better search ranking.
I don't think that qualifies, as network effects keep increasing with the size of the network. Click data's value tails off pretty rapidly once you nail the top few things that people click on, which is the great bulk of the usage.
It's actually exactly the opposite. Google's most valuable data is the long tail, which is substantial in search, since they have some data for rare searches, whereas Bing as none.
The network effect for google comes from it's relationships with users and advertizers. This is why Microsoft couldn't overtake them despite spending so much on bing.
More users -> more data -> better product -> more users
s/b
More users -> more labeled data -> better product -> more users
ML systems need lots of labeled data, not just lots of data. This is one of the primary why game playing AI's have had such great successes, tons of labeled data are relatively cheap. Great discourse on this and other related issues here:
Supervised learning requires labels. Unsupervised doesn't. Self driving cars use elements of both. This is why companies like Cruise and Tesla are rushing to get basic self driving cars out there. They need driving data.
You can get creative about labels. For instance, with known velocity of the car, predict where an object in view of the camera will be 5 frames in the future... This makes all data recorded by the camera effectively labeled, and is actually a useful thing to have.
It's the same reason Google is invulnerable in web search.