Hacker News Reader

Comment by rvnx | Hacker News Reader

Wasn't it GitLab actually ?

Otherwise, about the article, it's unfair to say that we need to correct the AI/ML to a more "neutral" perspective.

I'm always worried about artificial "neutrality", as neutrality often ends up just adding more weight to the perspective that you consider more socially and politically acceptable.

In Australia for example, French people have a bad reputation because they tend to steal from shops ("french shopping": https://www.traveller.com.au/french-nickers-cause-a-stink-do... ).

If tomorrow you build software for the shops to identify risks; well, whether you like it or not, it'll target French.

Should we artificially add more examples of non-stealing French in the dataset just because it's socially more acceptable ?

(I took French because it's socially and legally ok to blame French for mistakes of this planet, and I was born there, but replace with whatever suits you)

View on HN

The current problem surrounding using AI/ML in this area is that it can't distinguish between Jean Pierre and John Paul, Luc and Luke and authorities rely on it with no nuance or discernment.

You worry about artificial neutrality creeping in to the algorithms, but the algorithms already reflect human bias. We can't even reliably tell where someone is from yet we expect our AI to do it close to flawlessly.

I'm from East Asia and I think I can distinguish an East Asian's country of origin maybe 70% of the time. But 70% is nowhere near good enough if a person's life literally depends on the decision. And I bet you'd be hard-pressed to find someone even in around 90% accuracy. Migration and shared culture, among other things, make this difficult.

> Should we artificially add more examples of non-stealing French in the dataset just because it's socially more acceptable ?

You are completely missing the point.

This issue is simple. Take two groups A and B. Ceteris paribus if you control twice as much in group A than group B you will see twice as much positive events in group A. That's a sampling bias. Then you conclude group A is worth and should be controlled more leading to even more positive. It's all about sampling bias and feedback loops.

It's GitHub. If you create a new repository it will automatically suggest you change the name to `main` on the 'setup page'.

Actually it's in git now. I just got this warning when doing git init . last night.

Eh I realized the warning isn't in english but it basically says master will be changed to main and how I could set a global setting and how I could rename a newly created branch.

> it basically says master will be changed to main

It doesn’t do this. Here is the message:

    hint: Using 'master' as the name for the initial branch. This default branch name
    hint: is subject to change. To configure the initial branch name to use in all
    hint: of your new repositories, which will suppress this warning, call:
    hint: 
    hint:  git config --global init.defaultBranch <name>
    hint: 
    hint: Names commonly chosen instead of 'master' are 'main', 'trunk' and
    hint: 'development'. The just-created branch can be renamed via this command:
    hint: 
    hint:  git branch -m <name>

You beat me to it posting that output. Still, the 'This default branch name is subject to change' feels a bit awkward.

> Otherwise, about the article, it's unfair to say that we need to correct the AI/ML to a more "neutral" perspective.

I work in ML. its not neutrality, its representation.

if you use your own staff to train a face detector, a detector to detect small faces in CCTV, it'll only work on your staff. It won't pick up children, anybody who is slightly brown, anyone with big hair, or people who wear hats.

thats rubbish if you expect your models to work in the wider world.

don't think of is as political, think of it as trying to release a word processor to the world that only support ANSII, and wondering why its not selling overseas.

Reply on news.ycombinator.com