Hacker News Reader

Comment by gibsonf1 | Hacker News Reader

You cannot predict the future with patterns from the past, regardless of how many patterns you crunch - the world is always changing. ML can just never solve FSD - conceptual understanding and causality is needed to achieve human level driving capability. We instantly classify things conceptually from our perception, instantly understand how those things may behave, and then learn the specifics of what that thing is doing to instantly predict what might happen and how it is interacting with other causally connected things in the environment.

If ML could predict the future, there would be some very rich stock brokers right now.

Autonomous driving efforts need a major pivot to start working with concepts, causality, and the integrated space time model of the world we humans use.

View on HN

I think you uave some major misconceptions about how this stuff works. Most vision models output the current state of the world, to be integrated with other sensor data like lidar that is then used for planning. There are models that try to guess things like the crossing intent of pedestrians, but even those models just output a confidence of them crossing. I work for a different self driving company, but this high level stuff is pretty much the same everywhere.

Exactly, that is how that stuff works. Past patterns in the Ml neural net generated from massive data and training try to figure out the label of things from various sensors, and then given that label, try to figure out other things that might happen based on the label and event stream. These patterns can recognize with a certain probability similar patterns, but they have no way to cope with brand new patterns. Those brand new patterns, although they may only occur 1 out of 100 experiences - or lets say 3 days a year when your're driving, something fully novel occurs, this will kill you if you are relying on FSD. Until the systems try to think like humans, they simply wont.

> Autonomous driving efforts need a major pivot to start working with concepts, causality, and the integrated space time model of the world we humans use.

Dear god, no. I don't know what you think AV companies are doing past the perception layer, but they would be pivoting away from controls and planning -- you know, stuff that actually works outside of neurips/iclr/&c fantasy land.

Reply on news.ycombinator.com