Let's talk self-driving cars for a bit. I don't know why, though it seems my feed is talking about them lately.

First, some context.

In computer science, NP-complete (or NP-hard) is what you use to discuss unsolved algorithms. Shorthand is NPC.

https://en.wikipedia.org/wiki/NP-completeness

AI has same notion in AI-complete (or AI-hard).

https://en.wikipedia.org/wiki/AI-complete

A very very basic distillation of NPC problems is that if *any* in the set of problems can be solved quickly, all can be solved quickly.

It's one of those fundamental things in computer science that we haven't solved yet.

AI-complete (I don't know if it gets shortened to AIC, though I'm going to do that for twitter reasons) is similar.

AIC problems are thought to include computer vision, natural language processing, and unexpected real world situations.

Currently, AIC requires human intervention to be solved, meaning, computer technology (either compute or algorithms) is not enough to solve these problems.

What this also likely means is that full self-driving cars are not currently achievable.

Note: I say "likely" in the sense that the unknowable is by definition uncertain and we have to talk likelihood etc. Twitter isn't good for nuance.

Putting it another way, self driving cars, autonomous coding, replace all humans for every task robots are not happening any time soon.

That said....

While self-driving cars are unlikely, self-driving features are for sure achievable. In fact, that's the point of AIC saying 'human intervention'. We can build systems that do some, but not all, and definitely not fully autonomous right now.

But we shouldn't stop pushing them because useful features and things will emerge out of this exploration.

But don't buy a Tesla thinking it will be a robotaxi anytime soon in our current imaging of it.

Same goes for construction/factory/surgery robots. These are going to revolutionize the various industries *with* humans helping, not by replacing humans.

What you'll see is robots that do things more precisely or quickly or easily, though do require people for the tricky bits still.

Good example is surgery robots.

Overall, my quick views on this as it currently stands.

1. I'm a fan of appropriate marketing hype. And self-driving as the prime example of marketing having gotten out ahead of the tech.

Do I think we'll achieve full self-driving cars?

In my lifetime, no.

Not unless we do things differently. Right now we are pushing *the car* to achieve this because we are trying to replicate and replace the *driver*.

If we want to achieve full self-driving, my view is we would have to look at the whole driving system, meaning, roads, cars, cameras, etc etc etc.

This scale of change and coordination is for sure achievable, we would simply have to fund it and work at it more than we are now.

There is some awesome work in this area. In fact Ohio has one of the more forward looking approaches to this with @DriveOhio ( https://drive.ohio.gov/wps/portal/gov/driveohio) and a stretch of technology proving grounds highway route 33 https://www.33smartcorridor.com/ 

2. We should absolutely continue to fund efforts to get to self-driving, autonomous coding etc, though I would love if people understood better what they were funding.

I love funding moonshots because there is great advances (and money) to be made in the problems solved along the way.

NASA is my all-time favorite example here. The $ generated by funding the race to the moon is almost incalculable. Likely one of the largest returns ever.

This is how I look at funding these efforts today. You are unlikely to get self-driving, but you'll get immense value just going in that direction. Capture that

3. I appreciate car manufacturers wanting to push the marketing envelope, though consumers must understand what they buy today can have self-driving features, though a car completed today cannot with any degree of certainty guarantee it can achieve self-driving at any time.

Do not buy a car thinking you'll get this anytime soon.

Last few thoughts.

While self-driving cars and robotaxis are not likely to happen anytime soon and probably not in my lifetime, I love seeing the advances that come from pushing in that direction.

I would prefer companies not marketing hype this because then the only choice is for people to think something was an eventual failure because it didn't achieve X even if it invented XYZ things along the way.

I encourage people not to get discouraged by not achieving full-self driving anytime soon because the things we build along the way will be amazing.