Factors driving self-driving car adoption

I have an informal wager running with a handful of people on the following question:

What year will cars that are autonomous-capable constitute over 50% of the cars sold in that model year?

I began this wager in 2015, and I said "within 7 years". Giving myself credit for the fact that 2016 model year cars start being sold in late 2015, I'm going to clarify my prediction to say that the 2023 model year is my target prediction.

I acknowledge this is optimistic. Here's a list of the beliefs that drive my prediction (which I may update from time to time on this blog post):

1. Technology will advance faster than mainstream pundits give credit.
2. Insurance factors will drive rapid adoption. Accident losses will be lower on self-driving cars, and to maximize profit, auto insurors will offer insurance at lower cost to cover self-driving cars, quickly growing demand.
3. Fleet-economics driven by Uber, Lyft & others will incentivize & accelerate needed manufacturer engineering.
4. Multiple teams will be attacking the machine intelligence software needed, so manufacturers will have the choice of using licensed tech, or inventing in-house (if able). The rate of change in machine intelligence has increased dramatically in the last several years compared to the decades prior. The sensor technology is already available off-the-shelf, cheaply, and will be rolled into cars WELL before the software is fully able to use them.

There's one factor relating to this second assumption that I failed to recognize: Self-driving cars might dramatically shrink the auto insurance industry. If loss-rates are in fact much lower for self-driving cars, and if fleet accident rates are low-enough to encourage self-insuring (for collision loss), this could actually dramatically drive down overall auto insurance revenue. I had not considered this until reading this article at the Boston Globe on the topic. This factor might slow insurer embrace, and eliminate the speedy lowering of rates assumed in belief #2 above.

Nonetheless, I remain more optimistic than pundits (even "expert" ones) that predict a longer time frame for this.

What do you think?

PS - I've always kept one ace-in-the-whole: How do you define "autonomous-capable"? To be self-serving, I could cheat with my definition. E.g. See item #4 above regarding sensors. ;-)