5 ms·
I think it's underappreciated just how much human labour goes into annotating data fro autonomous vehicles. Scale is a fast growing start-up, with 35 employees
by Fricken 8y ago
I think it's underappreciated just how much human labour goes into annotating data fro autonomous vehicles. Scale is a fast growing start-up, with 35 employees and around 10,000 contractors, many based out of India and Africa, who do data annotation for Cruise, Zoox, Lyft, and Nutonomy among others.
https://techcrunch.com/2018/08/07/scale-whose-army-of-humans-annotate-raw-data-to-train-self-driving-and-other-ai-systems-nabs-18m/ https://techcrunch.com/2018/08/07/scale-whose-army-of-humans...
- threeseed 8y agoAlso the recent Maps story about Apple and how they might be using human labour to trace satellite imagery to identify the buildings. I also wonder how many companies are using Amazon Mechanical Turk to automate their processes.
- api 8y agoGiven that this training data basically gets loaded into neural nets, etc., can it be said that in some indirect sense these data labelers are actually driving these cars? Are we still just building the mechanical turk? To be it shows how far we still are from biological intelligent systems. I don't recall having to point out what an apple is to my daughter more than a few times. Watching children learn is breathtaking. Even much simpler brains like those of rodents or cats are far beyond anything built so far in silico.
- david-gpu 8y agoGenerally, neural networks do not memorize their training data. If they worked that way, they would not do well with samples they've never seen before. So no, I would not say that the labellers are driving the cars. They do play an important role in the process of building these systems, though.