5 ms·

I'm sorry folks have to confront this graphic imagery so often. If you could use ML to classify killing in a scene better, you could do the public and the rese

I'm sorry folks have to confront this graphic imagery so often. If you could use ML to classify killing in a scene better, you could do the public and the researchers a favor. Use the classifier to limit the impact of a terrorist video before it is spread to social media and give researchers a chance to organize and consent to watching certain sections of the video when and if needed.

journalctl 7y ago
When will this ML-fueled fantasy end? You can’t use machine learning for most of these tasks. Not now, maybe not ever. And the fact is: real people are being paid shit wages to watch ISIS beheading videos right now. Today. As we sit and sip our coffee. So what can we do /right now/ to help solve this problem?
- reportgunner 7y ago
  So much this. I often wonder if the people clever enough to propose a ML solution in a random comment on the web have an idea of how ML/AI/nested IFs work. (or rather how anything works)
- state_less 7y ago
  I disagree that there isn't information in those images that could be used to filter out some of these scenes. You have blood, impact wounds, bodies going limp, and other factors. It's something you can work on right now, in addition to other remedies, to reduce the quantity and impact of the trauma.
SolaceQuantum 7y ago
That still would require massive human resources to categorize the training data, and then more human resources to continually quirk the algorithm on identifying traumatic media. I don't personally see a way to avoid using humans here, so it's important that we identify and deal with the damage done to people in an empathetic and humane manner.
- state_less 7y ago
  Humans (researchers) are already doing the work. What I'm suggesting is that you add a classifier so that in the future, some of the work is done for you by the machine.
  - psychometry 7y ago
    You can rest assured that many people in the ML world are way ahead of you. If ML could be effectively used, it already would be.
    - state_less 7y ago
      There are a couple papers from 2017 where they talk about getting some effective results, but there is a lack of datasets around this topic and that the area is less studied. Can you point to the efforts that have been made here? https://www.cs.cmu.edu/~rahuls/pub/caip2011-rahuls.pdf https://www.cs.cmu.edu/~rahuls/pub/caip2011-rahuls.pdf https://arxiv.org/pdf/1711.02578.pdf https://arxiv.org/pdf/1711.02578.pdf
  - SolaceQuantum 7y ago
    Yes, and I'm responding that that still requires human labor and trauma, therefore it is still important to classify these responses as harm and treat these jobs as dangerous to the psychological wellbeing of the people participating in them now- give them access to healthcare related to their workplace hazard.

journalctl 7y ago

When will this ML-fueled fantasy end? You can’t use machine learning for most of these tasks. Not now, maybe not ever. And the fact is: real people are being paid shit wages to watch ISIS beheading videos right now. Today. As we sit and sip our coffee. So what can we do /right now/ to help solve this problem?

reportgunner 7y ago
So much this. I often wonder if the people clever enough to propose a ML solution in a random comment on the web have an idea of how ML/AI/nested IFs work. (or rather how anything works)
state_less 7y ago
I disagree that there isn't information in those images that could be used to filter out some of these scenes. You have blood, impact wounds, bodies going limp, and other factors. It's something you can work on right now, in addition to other remedies, to reduce the quantity and impact of the trauma.

SolaceQuantum 7y ago

That still would require massive human resources to categorize the training data, and then more human resources to continually quirk the algorithm on identifying traumatic media. I don't personally see a way to avoid using humans here, so it's important that we identify and deal with the damage done to people in an empathetic and humane manner.

state_less 7y ago
Humans (researchers) are already doing the work. What I'm suggesting is that you add a classifier so that in the future, some of the work is done for you by the machine.
- psychometry 7y ago
  You can rest assured that many people in the ML world are way ahead of you. If ML could be effectively used, it already would be.
  - state_less 7y ago
    There are a couple papers from 2017 where they talk about getting some effective results, but there is a lack of datasets around this topic and that the area is less studied. Can you point to the efforts that have been made here? https://www.cs.cmu.edu/~rahuls/pub/caip2011-rahuls.pdf https://www.cs.cmu.edu/~rahuls/pub/caip2011-rahuls.pdf https://arxiv.org/pdf/1711.02578.pdf https://arxiv.org/pdf/1711.02578.pdf
- SolaceQuantum 7y ago
  Yes, and I'm responding that that still requires human labor and trauma, therefore it is still important to classify these responses as harm and treat these jobs as dangerous to the psychological wellbeing of the people participating in them now- give them access to healthcare related to their workplace hazard.