Besides self-driving autopilot, why would Tesla need AI scientists ?

[Taipan]

I'm studying, as a part-time student via MOOC such as Coursera/Udacity and Fast.ai, the current SOTA (State Of The Art) techniques in Machine/Deep learning.

Many students/instructors refer to Tesla (and other Elon's projects such as SpaceX and OpenAI) as a cutting-edge "real-life" applications of AI.

I can deffo understand the use of AI for the integrated module of Autopilot in a Tesla, same for the Autopilot in a SpaceX rocket returning to land.

But when I read Tesla's mission statement "to accelerate the advent of sustainable transport by bringing compelling mass market electric cars to market as soon as possible", I wonder what AI can really do there.

It's not like purchasing a Tesla car is similar to buying an item on Amazon or selecting a movie on Netflix (ie. recommendations generated by AI)

How can data science help in accelerating electric vehicle adoption ?

 

[Tam]

...what AI can really do there...

1) Less cost and Less Time Delay:

My understanding is: Tesla doesn't need to write all scenarios for Artificial Intelligence as it learns from its failures and will eventually corrects itself.

Code writing is very intensive so Tesla might need to hire lots of programmers to write for all scenarios.

On the other hand, Tesla can hire less code writers for AI, thus, saving cost and time.

Less cost and time mean more affordability.

More affordable means consumers can adopt the technology faster and easier.

2) Adding technology to persuade upgrading customers:

Like cellphone, people keep upgrading because they like new features, it's the same with Tesla.

I had my 2012 Model S and I was not about to upgrade at all (no, not even with weaker MobilEye Autopilot).

Not until Tesla got a more powerful nVidia-powered AI-platform Autopilot and that did me in.

The more people like me who spend money to upgrade, the more cash for Tesla which in turn, the faster for Tesla to afford cheaper Tesla car versions including Model 3 which in turn, will promote EV adoption faster.

 

[calisnow]

You're big on mission statements eh OP? I see a future committee volunteer in you son!

 

[mblakele]

There is some scope for an Amazon-like recommendation engine, I suppose. Tesla could conceivably recommend vehicles and accessories based on what they know about you. They may already do this, but I haven't seen much evidence that it's done very well.

Leaving that aside, AI and data science have the potential to help in pretty much any area where people have to make decisions. Consider manufacturing applications. Tesla makes various stuff: vehicles and (in partnership with Panasonic) batteries. Apply big data, ML, etc. to improving manufacturing efficiency. Delivery and service too.

Then there are the everyday functions of the vehicles, PowerWalls, etc. Self-driving is just one of these. Consider battery management: could the vehicle do a better job of learning individual driving patters, so it can use shore power to pre-condition the battery? PowerWalls should learn how individual households use power, to optimize charging and discharging.

Finally Tesla has financials and assets to manage. Here too, improving efficiencies would further Tesla's mission.

 

[S4WRXTTCS]

My take on it is that self-driving cars, and electric cars go hand in hand.

The two things combined is really the only sustainable way forwards.

AI is critical not just in the task of self-driving, but in getting from point A to point B. I have to laugh a little bit at the concept of self-driving Tesla when a regular Tesla can't even navigate its owner to a destination without screwing up.

I can't even say I'm happy with any of the navigation apps from anyone. They're all missing things AI would give them. I'm still regularly treating navigation prompts as suggestions versus actually following it. Navigation apps don't take into account that traffic makes it impossible to do certain maneuvers at various times in the day.

Obviously getting from point A to point B is the entire point of a car. So improving the ability for the driver to do so is monumental regardless of whether the driver is human or a robot.

I can see data science helping if you think of how many times people have had to override navigation prompts to place X. A navigation app used be a sufficiently large amount of people should be extremely fast at self correcting.

 

[Taipan]

Greetings @Tam, @mblakele & @S4WRXTTCS !

Many thanks for your detailed responses, these make a lot of sense and confirm what I was taught (in theory, not in practice/IRL) regarding both the AutoPilot/Self-Driving function, which I worked on with CNN's applied to Computer Vision, and battery/manufacturing/servicing management-projections.

Aren't these implementations of AI (in the wide sense of it), and their benefits, only available/detectable to Tesla cars' owners/users ?
As in "Once converted, these former ICE owners will be convinced of the benefit of an EV built and serviced by Tesla, and become Evangelists".

What I wonder is: "Are there AI experts at Tesla focusing on converting ICE owners into EV drivers/owners? And if so, how/what are they proceeding ?".

T.

 

[mblakele]

Are there AI experts at Tesla focusing on converting ICE owners into EV drivers/owners? And if so, how/what are they proceeding ?

That sounds like marketing. Tesla doesn't spend much money on that. What little they do is mostly concentrated on making the brand and products look as sexy as possible in social media. Maybe there's scope for AI-ish help with that: deciding when and what and where new messaging should appear. I haven't seen any evidence that Tesla currently tries to automate those decisions.

However I think it's an idea worth exploring. Maybe you could design an A-B test around it?

 

[Tam]

...AI experts at Tesla focusing on converting ICE owners into EV drivers/owners...

There is no reasons that AI cannot be applied in non-Autopilot areas such as websites, social medias, and as mblakele mentioned marketing... to accelerate faster EV-adopters.

However, Tesla's supply has never been able to keep up with demands, so diverting AI for marketing is a misuse of resource.

That's why Tesla has open-sourced patents so the EV supply as a whole, not just Tesla's, will be able to keep up with the demand.

Tesla-inspired Chinese EV startup launches all-electric SUV using open-source patents

 

[RDoc]

Could AI be used to infer that if a USB stick had both music and playlist files on it, that if the user selected the playlist file, they wanted to have the music files played in the order they appear in the playlist file? That's probably asking too much.