AI Surveillance Startup Caught Using Sweatshop Workers to Monitor US Residents

I want to preface this by saying that I don't like Flock or what they do. I am not intending to call them "ethical". Below, I will say that this article gives no evidence that Flock is being unethical in THIS INSTANCE. I do think they do unethical things in other instances.

With that said, this is a non-story for anyone who understands how machine learning works. An ML or AI model is nothing more than a bunch of well-labeled data describing the phenomena you'd like to predict and a bunch of math that connects that data in interesting ways. If you want to "train" an ML or AI model, you must have a good, well-labeled training dataset. Something that the model can "learn" from.

The ONLY way to get a large, well-labeled dataset is to have a bunch of humans look at a lot of data and annotate it to say what is in it. It is possible to do this in an ethical way by anonymizing data and feeding it to the annotators in unconnected ways. They may or may not be doing this ethically but the article gives no evidence either way.

Unlike Amazon, with their cashier-less stores, this doesn't accuse Flock of using human labor to look at real, live data. They aren't being accused of being an "AI" model composed of inexpensive labor. They are being accused of doing what every ML or AI model builder does; using human labor to label training data for their model. This is the same for corporate models (which tend to use employees or gig workers) as it is for academic models (which tend to use graduate students).

Flock is unethical in other ways but this is just how the ML and AI sausages are made.

3 comments

I want to preface this by saying that I don't like Flock or what they do. I am not intending to call them "ethical". Below, I will say that this article gives no evidence that Flock is being unethical in THIS INSTANCE. I do think they do unethical things in other instances.
With that said, this is a non-story for anyone who understands how machine learning works. An ML or AI model is nothing more than a bunch of well-labeled data describing the phenomena you'd like to predict and a bunch of math that connects that data in interesting ways. If you want to "train" an ML or AI model, you must have a good, well-labeled training dataset. Something that the model can "learn" from.
The ONLY way to get a large, well-labeled dataset is to have a bunch of humans look at a lot of data and annotate it to say what is in it. It is possible to do this in an ethical way by anonymizing data and feeding it to the annotators in unconnected ways. They may or may not be doing this ethically but the article gives no evidence either way.
Unlike Amazon, with their cashier-less stores, this doesn't accuse Flock of using human labor to look at real, live data. They aren't being accused of being an "AI" model composed of inexpensive labor. They are being accused of doing what every ML or AI model builder does; using human labor to label training data for their model. This is the same for corporate models (which tend to use employees or gig workers) as it is for academic models (which tend to use graduate students).
Flock is unethical in other ways but this is just how the ML and AI sausages are made.
- Thats a lot of bullshit words for something that has no right to exist in American society. The company flock is unethical in its existence. We should tar and feather the ceos and investors and punish and shame people who don't value freedom and privacy. You should be ashamed.
  
  For saying that they are unethical but that this article is not evidence of that I should be ashamed? This conclusions in this article would label everyone who develops ML models as unethical. How about we focus on the ways in which Flock is unethical rather than making non-issues into issues?
  Flock is absolutely unethical but this is not evidence of that.