7 ms·

I’m a PhD student focusing on large scale video analysis. It’s an unfortunate fact that a lot of the applied research in this space is motivated by surveillance

I’m a PhD student focusing on large scale video analysis. It’s an unfortunate fact that a lot of the applied research in this space is motivated by surveillance, particularly in systems, eg what’s come out of Microsoft’s efforts here [0]. A colleague who attended CVPR said most of the industry booths were either surveillance or self driving cars. I wish researchers would start to think more creatively about what we can do with large video datasets and tools like face recognition. The least we can do is try to use this technology for social good. For example, a USC/Google team used computer vision tools to identify gender bias in modern cinema [1] (for my own research, we’re doing a similar kind of analysis on TV news at 100x the scale). [0] https://www.usenix.org/conference/nsdi17/technical-sessions/presentation/zhang https://www.usenix.org/conference/nsdi17/technical-sessions/... [1] https://www.google.com/about/main/gender-equality-films/ https://www.google.com/about/main/gender-equality-films/

walrus01 8y ago
> I wish researchers would start to think more creatively about what we can do with large video datasets and tools like face recognition. Some very smart people in China did exactly that, and arrived at a conclusion on what the local market for surveillance tech was lacking. Based on what the government will fund, they came up with an answer we seem to dislike.
- make3 8y ago
  I think he means, aside from dystopian population surveillance
  - jerkstate 8y ago
    yeah, like a thing that would remind me what peoples names are at parties. probably not as much money in that though.
    - make3 8y ago
      you can probably scrape your own Facebook friends' pictures & use the GCS or Azure free services
bachbach 8y ago
> The least we can do is try to use this technology for social good. For example, a USC/Google team used computer vision tools to identify gender bias in modern cinema That sounds entirely worthless. A social good involves helping somebody who needs aid. This proposal merely fills a perceived need by political dogmatics. This is exactly a sociopolitical memory leak.
- throwawayjava 8y ago
  > A social good involves helping somebody who needs aid. The citation provided by the parent post presents concrete claims about who is being helped and how. Here's my summary of the argument in the paper: 1. Gender bias exists in films (intro). 2. This bias has a tangible negative impact on women's lives (section titled "Role models on-screen make a difference"). 3. Quantifying this bias helps filmmakers create films that are less biased (section titled "How can this technology change things?") Summarily, women need aid because of how they are portrayed in films and quantifying/identifying bias helps portray women in a less damaging way. To be clear, I think it's entirely reasonable to disagree with any one of these assertions (esp the third)! But you don't present any opposing arguments, you just claim that the research doesn't meet a prime facie burden of "helping somebody who needs aid" even though the linked website goes to great lengths to make the case that it does meet this exact burden. > This proposal merely fills a perceived need by political dogmatics The same is true for surveillance technology, and military technology, and renewable energy technology, and even economic development. There are very few non-political notions of "social good". Defining and achieving "social good" is fundamentally a political act. If you want research to have an impact on the world, then you've got to be comfortable wading into the political. Because any time you change the world in non-trivial ways, politics is eventually at hand.
- dang 8y ago
  Please don't take HN threads into political flamewar. https://news.ycombinator.com/newsguidelines.html https://news.ycombinator.com/newsguidelines.html
  - mark_edward 8y ago
ineednoprocrast 8y ago
I rarely post, but I feel it necessary to chime in here. I studied computational linguistics/natural language processing in grad school, out of an interest in accessibility issues (I had RSI) and related applications, and had naively neglected to look very hard about how I'd be likely use it in a career. Other people in my cohort went to work for the CIA. I couldn't move to the bay area for... reasons, (this was pre-Google) and it seemed everyplace I could find anything wanted me to have a security clearance, with all the implications that has. One could say I'm not mercenary enough, or that I took too many philosophy classes, or any number of other rationalizations, but one way or another, none of the jobs I've taken have had much need for what I studied. I feel morally obligated to use what I know to improve the state of the world. This has, unfortunately, led to a lot of unemployment and serious depression. Pseudonym for obvious reasons.
- chronic821 8y ago
  > I feel morally obligated to use what I know to improve the state of the world. This has, unfortunately, led to a lot of unemployment and serious depression. And as you would suspect, I'm a PhD grad working in the Bay area on mas surveillance AI. Half a million dollars a year. I honestly don't know how to invest the money anymore. Index funds are boring at this point.
  - prolikewh0a 8y ago
    What point was there in posting this?
    - arcticbull 8y ago
      It seems like what he's suggesting is that selling out can be quite lucrative, and if you can deal with it, there are rewards to be had.
  - ineednoprocrast 8y ago
    I'm a little confused by the beginning of your response, because you don't appear to be the person I replied to. When you wrote this, what did you anticipate my reaction would be? Why? I must admit, I really don't know what it would be like to be working in the bay area on mass surveillance AI for half a million a year.
est 8y ago
> identify gender bias in modern cinema Looks like a way to sell more ads.
dbcooper 8y ago
Can't wait for algorithm homogenised movies!
- arcticbull 8y ago
  I think this is the internal slogan at Netflix :)
- severine 8y ago
  I'd say you've seen a bunch of them, like everyone (sigh)
thrantaway 8y ago
It's disheartening that so many bright people work on research that will undoubtedly be abused by authoritarian governments and make the world a worse place for everyone. Ask a thousand people what video analysis can be used for and I bet the vast majority will say surveillance, and most would probably struggle to think about something else. I wish researchers would start to ask themselves if what they're building can be abused by authoritarian governments, and if so, switch focus to a field of research that will make the world a better, rather than a worse, place to live in. History will not look kindly on researchers who built the tools that enabled Xi the Pooh to achieve a dystopian surveillance state.
- extralego 8y ago
  History? You mean Google? Google will glorify it’s history in it’s middle schools just like America glorifies it’s history in it’s middle schools. Considering the researchers are hired according to their sheltered backgrounds, holding them as indivuals accountable is futile. If you care, vote accordingly, for politicians willing to refuse corporate donations. If you care a lot, study revolutionary history, learn from it and organize. I’m just paraphrasing the only thing that seems to ever actually work.
  - thrantaway 8y ago
    Google certainly played a major role, and it wasn't my intention to shelter companies/governments from blame. However, I certainly don't believe holding individuals accountable is futile. They need to consider if people can be suppressed, stripped of their rights, imprisoned, and murdered thanks to their research/code. We should encourage developers to fight for privacy, and discourage them from engaging in research/writing code that can be used to commit evil acts. And yes, I wholeheartedly agree that everyone should vote.
  - v1vek 8y ago
    Google wasn't responsible for any of the major breakthroughs in AI. The current AI boom is driven by deep learning[1], due to Alex Krizhevsky [2][3], who was then a Ph.D. student @ University of Toronto. Subsequently, several significant breakthroughs have been in Universities & outside Google or FAANGs (Siri, DeepMind were all questions) [1] https://en.wikipedia.org/wiki/Deep_learning https://en.wikipedia.org/wiki/Deep_learning [2] https://beamandrew.github.io/deeplearning/2017/02/23/deep_learning_101_part1.html https://beamandrew.github.io/deeplearning/2017/02/23/deep_le... [3] https://arxiv.org/abs/1803.01164 https://arxiv.org/abs/1803.01164
extralego 8y ago
To be fair, the absolute last thing “modern cinema” needs is another formulaic measurement of demographically correlated screentime. I am not personally against girls seeing depictions of other girls in movies (have you consulted with the anti-bodyshaming sect?) however this claim to “social good” is telling of a desperate effort to ignore actual social problems, none of which are so conveniently morally distinct from the economic disparities Google instigates and defends on a daily basis. Studies show little girls who are homeless have bigger problems.
Noos 8y ago
If you asked them, they would effusively tell you all this surveillance is being used for the social good, and probably even believe it. Not many people are brutally honest about how their work destroys the social fabric, and will always point to the benefit it provides. Pornographers think its a social good if we would just be a little less uptight about sexuality, and more porn access is a solution. And the same people who try to work for good can find hey've just armed others for evil by giving them tools.

walrus01 8y ago

> I wish researchers would start to think more creatively about what we can do with large video datasets and tools like face recognition. Some very smart people in China did exactly that, and arrived at a conclusion on what the local market for surveillance tech was lacking. Based on what the government will fund, they came up with an answer we seem to dislike.

make3 8y ago
I think he means, aside from dystopian population surveillance
- jerkstate 8y ago
  yeah, like a thing that would remind me what peoples names are at parties. probably not as much money in that though.
  - make3 8y ago
    you can probably scrape your own Facebook friends' pictures & use the GCS or Azure free services

bachbach 8y ago

> The least we can do is try to use this technology for social good. For example, a USC/Google team used computer vision tools to identify gender bias in modern cinema That sounds entirely worthless. A social good involves helping somebody who needs aid. This proposal merely fills a perceived need by political dogmatics. This is exactly a sociopolitical memory leak.

throwawayjava 8y ago
> A social good involves helping somebody who needs aid. The citation provided by the parent post presents concrete claims about who is being helped and how. Here's my summary of the argument in the paper: 1. Gender bias exists in films (intro). 2. This bias has a tangible negative impact on women's lives (section titled "Role models on-screen make a difference"). 3. Quantifying this bias helps filmmakers create films that are less biased (section titled "How can this technology change things?") Summarily, women need aid because of how they are portrayed in films and quantifying/identifying bias helps portray women in a less damaging way. To be clear, I think it's entirely reasonable to disagree with any one of these assertions (esp the third)! But you don't present any opposing arguments, you just claim that the research doesn't meet a prime facie burden of "helping somebody who needs aid" even though the linked website goes to great lengths to make the case that it does meet this exact burden. > This proposal merely fills a perceived need by political dogmatics The same is true for surveillance technology, and military technology, and renewable energy technology, and even economic development. There are very few non-political notions of "social good". Defining and achieving "social good" is fundamentally a political act. If you want research to have an impact on the world, then you've got to be comfortable wading into the political. Because any time you change the world in non-trivial ways, politics is eventually at hand.
dang 8y ago
Please don't take HN threads into political flamewar. https://news.ycombinator.com/newsguidelines.html https://news.ycombinator.com/newsguidelines.html
- mark_edward 8y ago

ineednoprocrast 8y ago

I rarely post, but I feel it necessary to chime in here. I studied computational linguistics/natural language processing in grad school, out of an interest in accessibility issues (I had RSI) and related applications, and had naively neglected to look very hard about how I'd be likely use it in a career. Other people in my cohort went to work for the CIA. I couldn't move to the bay area for... reasons, (this was pre-Google) and it seemed everyplace I could find anything wanted me to have a security clearance, with all the implications that has. One could say I'm not mercenary enough, or that I took too many philosophy classes, or any number of other rationalizations, but one way or another, none of the jobs I've taken have had much need for what I studied. I feel morally obligated to use what I know to improve the state of the world. This has, unfortunately, led to a lot of unemployment and serious depression. Pseudonym for obvious reasons.

chronic821 8y ago
> I feel morally obligated to use what I know to improve the state of the world. This has, unfortunately, led to a lot of unemployment and serious depression. And as you would suspect, I'm a PhD grad working in the Bay area on mas surveillance AI. Half a million dollars a year. I honestly don't know how to invest the money anymore. Index funds are boring at this point.
- prolikewh0a 8y ago
  What point was there in posting this?
  - arcticbull 8y ago
    It seems like what he's suggesting is that selling out can be quite lucrative, and if you can deal with it, there are rewards to be had.
- ineednoprocrast 8y ago
  I'm a little confused by the beginning of your response, because you don't appear to be the person I replied to. When you wrote this, what did you anticipate my reaction would be? Why? I must admit, I really don't know what it would be like to be working in the bay area on mass surveillance AI for half a million a year.

est 8y ago

> identify gender bias in modern cinema Looks like a way to sell more ads.

dbcooper 8y ago

Can't wait for algorithm homogenised movies!

arcticbull 8y ago
I think this is the internal slogan at Netflix :)
severine 8y ago
I'd say you've seen a bunch of them, like everyone (sigh)

thrantaway 8y ago

It's disheartening that so many bright people work on research that will undoubtedly be abused by authoritarian governments and make the world a worse place for everyone. Ask a thousand people what video analysis can be used for and I bet the vast majority will say surveillance, and most would probably struggle to think about something else. I wish researchers would start to ask themselves if what they're building can be abused by authoritarian governments, and if so, switch focus to a field of research that will make the world a better, rather than a worse, place to live in. History will not look kindly on researchers who built the tools that enabled Xi the Pooh to achieve a dystopian surveillance state.

extralego 8y ago
History? You mean Google? Google will glorify it’s history in it’s middle schools just like America glorifies it’s history in it’s middle schools. Considering the researchers are hired according to their sheltered backgrounds, holding them as indivuals accountable is futile. If you care, vote accordingly, for politicians willing to refuse corporate donations. If you care a lot, study revolutionary history, learn from it and organize. I’m just paraphrasing the only thing that seems to ever actually work.
- thrantaway 8y ago
  Google certainly played a major role, and it wasn't my intention to shelter companies/governments from blame. However, I certainly don't believe holding individuals accountable is futile. They need to consider if people can be suppressed, stripped of their rights, imprisoned, and murdered thanks to their research/code. We should encourage developers to fight for privacy, and discourage them from engaging in research/writing code that can be used to commit evil acts. And yes, I wholeheartedly agree that everyone should vote.
- v1vek 8y ago
  Google wasn't responsible for any of the major breakthroughs in AI. The current AI boom is driven by deep learning[1], due to Alex Krizhevsky [2][3], who was then a Ph.D. student @ University of Toronto. Subsequently, several significant breakthroughs have been in Universities & outside Google or FAANGs (Siri, DeepMind were all questions) [1] https://en.wikipedia.org/wiki/Deep_learning https://en.wikipedia.org/wiki/Deep_learning [2] https://beamandrew.github.io/deeplearning/2017/02/23/deep_learning_101_part1.html https://beamandrew.github.io/deeplearning/2017/02/23/deep_le... [3] https://arxiv.org/abs/1803.01164 https://arxiv.org/abs/1803.01164

extralego 8y ago

To be fair, the absolute last thing “modern cinema” needs is another formulaic measurement of demographically correlated screentime. I am not personally against girls seeing depictions of other girls in movies (have you consulted with the anti-bodyshaming sect?) however this claim to “social good” is telling of a desperate effort to ignore actual social problems, none of which are so conveniently morally distinct from the economic disparities Google instigates and defends on a daily basis. Studies show little girls who are homeless have bigger problems.

Noos 8y ago

If you asked them, they would effusively tell you all this surveillance is being used for the social good, and probably even believe it. Not many people are brutally honest about how their work destroys the social fabric, and will always point to the benefit it provides. Pornographers think its a social good if we would just be a little less uptight about sexuality, and more porn access is a solution. And the same people who try to work for good can find hey've just armed others for evil by giving them tools.