This post is more than a year old. The information, claims or views in this post may be out of date.
I cannot not write about this excellent presentation from a16z:
Two big things that stand out for me (and there’s lots of high-density info in that deck):
The ability for computers to understand the world around us got infinitely better once humans were taken out of the equation. Neural networks that train against vast sets of data and write their own rules turned out to be a lot more efficient than having human specialists trying to write those rules by hand.
Somewhere in 2016 (or maybe even as early as 2014) we crossed the first Rubicon: Human engineers may no longer be capable of keeping up with the intellectual growth of the machines they used to manage.
Fantastic news for scale and growth – computers can now write better and more efficient software, which in turn gets loaded on to ever-smaller and lower-powered devices. Ambient intelligence is just around the corner.
Slightly worse news for governance and accountability – at what point does the outcome of a program stop being the responsibility of the human engineers, and start becoming the responsibility of the neural network that designed its own decision tree?
The day we need to prosecute a neural network for a crime – that’ll be the second Rubicon. Once software has legal standing, the game changes again. Probably not for the better.
Mobile applied to Automotive
Mobile phones scaled out a hell of a lot faster than PCs ever could (hence the title of the presentation), but one of the things it has made a significant impact on is manufacturing. The halo effect of having so many compact, mass-produced components means that hardware is no longer a true differentiating factor, and it’s much more about the software and services that power those devices.
The same could be true for cars. We might be heading into a future where cars (taking “electric” for granted here) are assembled in the same way that smartphones are today (just by pulling off-the-shelf interoperable components together), and the key differentiator will be the services rendered through that car.
Which leads to the interesting thought of “Automotion-as-a-Service”.
I wonder if SaaS-type pricing will ever apply to the automotive industry. Bundled minutes become bundled miles, personal assistant integrations cost extra, and you get cheaper packages if you accept in-car targeted advertising. Somehow I think that might happen.
Assuming there isn’t a Final Optimization at some point, where the neural networks collectively decide we’re too much trouble to deal with 😉