Earlier this week, Tesla updated the software in its popular Model S to introduce a vast swathe of new artificial intelligence. The auto-pilot mode brings autonomous driving to a mainstream road vehicle for the first time. But at its heart is the arrival of a stunningly effective business model for artificial intelligence.
There are some jaw-dropping elements at play:
* It was delivered as software over-the-air without a visit to a garage. Ponder on that, within the frame of software eating the world. We turned a car into an AI self-driving robot through the equivalent of a glorified text message.
There are two implications of this networked intelligence. One a business implication, the second about the nature of non-human intelligence. I’ll explore the first in more detail.
The business impact is crucial. Musk has created a method where every single human driver is generating a training set of data for the ecosystem of algorithms that allow a car to self-drive. Forget Google’s 1m miles of autonomous driving, Tesla can get a million miles of driving every week from 25,000 average drivers. The training set - which includes statistically rare weird edge cases - will become enormous. Tesla’s AI will literally be better able to navigate the world than competitor AIs.
And from great data, come effective AI systems. Google, of all, know this, perhaps best espoused in their classic “The Unreasonable Effectiveness of Data
” (five years old, but worth a read.)
So Tesla’s AI systems will get better and better than the competition. And they will have a substantial lead time on anyone else putting autonomous vehicles on the road.
That lead time is, in Warren Buffet’s term, the ‘economic moat’. At PeerIndex, we spent a lot of time thinking about how to build and generate proprietary data around topic-based expertise rankings, and by-and-large succeeded through a range of simple and complex methods to generate this data.
And to the second point: no exploration or hypothesis, just things to ponder. What questions does this network of AIs raise for our understanding of artificial or other non-human AIs? The car is not like a humanoid robot. It is embodied in an entirely different way. Where will the car’s intelligence located? In your own vehicle; in a ‘cloud’ or shared across all vehicles in the network? As these systems develop, does one ever ‘own’ one’s car if core parts of its navigation systems reside in a network or a higher-order emergent attribute of that network?