3 ms·

> Algorithms are less nuanced, hard to write, and will often fail, people are expensive and biased. Algorithms have a fantastic property of repeatability which

> Algorithms are less nuanced, hard to write, and will often fail, people are expensive and biased. Algorithms have a fantastic property of repeatability which means that it is (a) possible to adjust them over time (b) possible to identify what exactly happened (c) blind to the context and inputs outside of the algorithm That's exactly what is needed.

majewsky 9y ago
> Algorithms [allow to] identify what exactly happened "This video was demonetized because at the 1:15 mark, there is a patch of light dirt that our neural network classified as naked skin with a 70% confidence." Oh, you meant you were going to do image recognition without any sort of machine learning?
- notyourday 9y ago
  Your alternative is the idiot on his last day at Twitter who in his post Trump account blocking interview basically admitted that it was the totality of things that happened that day to him and it being his last day and that Trumps account popped up in front of him which caused him to make the arbitrary decision to block it. Please remove humans from making decisions that need to be uniform. So yes, algorithm it is.
  - cnewey 9y ago
    You're implicitly making the assumption that an algorithm is unbiased and less likely to be fallible - and unlikely to be susceptible to bias or manipulation (at least, compared to a human). There is a considerable amount of academic and popular literature showing that these assumptions are false, at least at a technical level. The problem of vetting YouTube videos isn't necessarily something that can be solved by an algorithm, at the end of the day. Lots of the simple cases, sure - but such an algorithm won't work on even the most moderately challenging content. After all, moderating complex content such as video doesn't just require a well-trained neural network to recognise objects portrayed in images and sounds in audio... deciding whether a video is appropriate or not for monetisation is something that requires an explicit understanding of the social and cultural context which the video is going to be viewed in - which is both highly subjective and difficult to express in an algorithm. NB; - Neural networks (and other AI techniques) can be tricked and manipulated (see [0,1]; examples of adversarial images) - Machine learning algorithms can induce or confirm bias through inadequate training data and poor assumptions. (see [2]; a book about bias in machine learning algorithms) [0]: https://arxiv.org/abs/1510.05328 https://arxiv.org/abs/1510.05328 [1]: http://www.evolvingai.org/files/DNNsEasilyFooled_cvpr15.pdf http://www.evolvingai.org/files/DNNsEasilyFooled_cvpr15.pdf [2]: https://weaponsofmathdestructionbook.com https://weaponsofmathdestructionbook.com
    - notyourday 9y ago
      > You're implicitly making the assumption that an algorithm is unbiased and less likely to be fallible - and unlikely to be susceptible to bias or manipulation (at least, compared to a human). There is a considerable amount of academic and popular literature showing that these assumptions are false, at least at a technical level. No, I'm arguing that algorithms are transparent, which means that they can be adjusted. > The problem of vetting YouTube videos isn't necessarily something that can be solved by an algorithm, at the end of the day. Lots of the simple cases, sure - but such an algorithm won't work on even the most moderately challenging content. of course they would here: while (!decision_pass_the_spot_check()) { tweak_algorithm() } > deciding whether a video is appropriate or not for monetisation is something that requires an explicit understanding of the social and cultural context which the video is going to be viewed in - which is both highly subjective and difficult to express in an algorithm. Certain cultures think it is acceptable to stone women. In their societal context it is OK. We do not, however, care that it is OK in their societal context. So we make sure that in our algorithm stoning women ==> demonetized. Now we move onto the next problem. Did someone flag a video that we processed that had women stoned that was not demonetized? We open a ticket with the people responsible for that model, with the video that was not demonetized and have them figure out why it did not happen. That's how you solve a complex problem. You break them into simpler ones and when you see an issue you address it. Claiming that the problem is just too complex for algorithms is avoidance.
      - cnewey 9y ago
        Your argument that algorithms are transparent is equally flawed. There is an entire field of research dedicated to introspection of layers in convolutional neural networks! A system that is complex and nuanced enough to be able to deal with a problem such as auto-moderating YouTube videos will be phenomenally difficult to inspect (and nearly impossible to "adjust" in the way that you seem to be suggesting). To clarify; I am not saying that it will never be possible to solve this sort of problem with an algorithm - just that doing so would require solving the small inconvenience of general artificial intelligence first... :-). It certainly isn't beyond the realm of possibility in the (very distant) future, but to suggest that it's possible right now (and that it would be better than a human!) is to over-egg the pudding somewhat.
strangecasts 9y ago
(b) possible to identify what exactly happened (c) blind to the context and inputs outside of the algorithm Neither of these are really true for neural networks: it's hard to pinpoint the exact feature or hidden layer which results in a particular classification. Training also requires datasets which are often proprietary, and stochastic approaches mean you can get different networks from the same dataset.
- notyourday 9y ago
  You may get a different network from a same dataset. You will get a different result if you let two humans classify the data.
  - strangecasts 9y ago
    Not "may" but will - it's very easy to render your network essentially unauditable (you don't have thousands of petabytes' worth of storage available? you can't fab your own ML coprocessor? too bad!) You bring up Trump's Twitter suspension in the other subthread, but in that case the failure was transient, and the person responsible and their reasoning were quickly identified; YouTube has demonetized videos with little to no explanation or recourse. Uniformity is absolutely a goal where people's livelihoods are affected, but ML algorithms can't guarantee it by themselves: they make it even more important to also maintain accountability and transparency.
    - notyourday 9y ago
      > Not "may" but will - it's very easy to render your network essentially unauditable (you don't have thousands of petabytes' worth of storage available? you can't fab your own ML coprocessor? too bad!) Google can. So that does not apply. > You bring up Trump's Twitter suspension in the other subthread, but in that case the failure was transient, and the person responsible and their reasoning were quickly identified; YouTube has demonetized videos with little to no explanation or recourse. There's no recourse or explanation because it is not youtube's business model regardless of what/who makes a decision to demonetize them. Adding arbitrary human behavior to those decisions only makes it worse.

majewsky 9y ago

> Algorithms [allow to] identify what exactly happened "This video was demonetized because at the 1:15 mark, there is a patch of light dirt that our neural network classified as naked skin with a 70% confidence." Oh, you meant you were going to do image recognition without any sort of machine learning?

notyourday 9y ago
Your alternative is the idiot on his last day at Twitter who in his post Trump account blocking interview basically admitted that it was the totality of things that happened that day to him and it being his last day and that Trumps account popped up in front of him which caused him to make the arbitrary decision to block it. Please remove humans from making decisions that need to be uniform. So yes, algorithm it is.
- cnewey 9y ago
  You're implicitly making the assumption that an algorithm is unbiased and less likely to be fallible - and unlikely to be susceptible to bias or manipulation (at least, compared to a human). There is a considerable amount of academic and popular literature showing that these assumptions are false, at least at a technical level. The problem of vetting YouTube videos isn't necessarily something that can be solved by an algorithm, at the end of the day. Lots of the simple cases, sure - but such an algorithm won't work on even the most moderately challenging content. After all, moderating complex content such as video doesn't just require a well-trained neural network to recognise objects portrayed in images and sounds in audio... deciding whether a video is appropriate or not for monetisation is something that requires an explicit understanding of the social and cultural context which the video is going to be viewed in - which is both highly subjective and difficult to express in an algorithm. NB; - Neural networks (and other AI techniques) can be tricked and manipulated (see [0,1]; examples of adversarial images) - Machine learning algorithms can induce or confirm bias through inadequate training data and poor assumptions. (see [2]; a book about bias in machine learning algorithms) [0]: https://arxiv.org/abs/1510.05328 https://arxiv.org/abs/1510.05328 [1]: http://www.evolvingai.org/files/DNNsEasilyFooled_cvpr15.pdf http://www.evolvingai.org/files/DNNsEasilyFooled_cvpr15.pdf [2]: https://weaponsofmathdestructionbook.com https://weaponsofmathdestructionbook.com
  - notyourday 9y ago
    > You're implicitly making the assumption that an algorithm is unbiased and less likely to be fallible - and unlikely to be susceptible to bias or manipulation (at least, compared to a human). There is a considerable amount of academic and popular literature showing that these assumptions are false, at least at a technical level. No, I'm arguing that algorithms are transparent, which means that they can be adjusted. > The problem of vetting YouTube videos isn't necessarily something that can be solved by an algorithm, at the end of the day. Lots of the simple cases, sure - but such an algorithm won't work on even the most moderately challenging content. of course they would here: while (!decision_pass_the_spot_check()) { tweak_algorithm() } > deciding whether a video is appropriate or not for monetisation is something that requires an explicit understanding of the social and cultural context which the video is going to be viewed in - which is both highly subjective and difficult to express in an algorithm. Certain cultures think it is acceptable to stone women. In their societal context it is OK. We do not, however, care that it is OK in their societal context. So we make sure that in our algorithm stoning women ==> demonetized. Now we move onto the next problem. Did someone flag a video that we processed that had women stoned that was not demonetized? We open a ticket with the people responsible for that model, with the video that was not demonetized and have them figure out why it did not happen. That's how you solve a complex problem. You break them into simpler ones and when you see an issue you address it. Claiming that the problem is just too complex for algorithms is avoidance.
    - cnewey 9y ago
      Your argument that algorithms are transparent is equally flawed. There is an entire field of research dedicated to introspection of layers in convolutional neural networks! A system that is complex and nuanced enough to be able to deal with a problem such as auto-moderating YouTube videos will be phenomenally difficult to inspect (and nearly impossible to "adjust" in the way that you seem to be suggesting). To clarify; I am not saying that it will never be possible to solve this sort of problem with an algorithm - just that doing so would require solving the small inconvenience of general artificial intelligence first... :-). It certainly isn't beyond the realm of possibility in the (very distant) future, but to suggest that it's possible right now (and that it would be better than a human!) is to over-egg the pudding somewhat.

strangecasts 9y ago

(b) possible to identify what exactly happened (c) blind to the context and inputs outside of the algorithm Neither of these are really true for neural networks: it's hard to pinpoint the exact feature or hidden layer which results in a particular classification. Training also requires datasets which are often proprietary, and stochastic approaches mean you can get different networks from the same dataset.

notyourday 9y ago
You may get a different network from a same dataset. You will get a different result if you let two humans classify the data.
- strangecasts 9y ago
  Not "may" but will - it's very easy to render your network essentially unauditable (you don't have thousands of petabytes' worth of storage available? you can't fab your own ML coprocessor? too bad!) You bring up Trump's Twitter suspension in the other subthread, but in that case the failure was transient, and the person responsible and their reasoning were quickly identified; YouTube has demonetized videos with little to no explanation or recourse. Uniformity is absolutely a goal where people's livelihoods are affected, but ML algorithms can't guarantee it by themselves: they make it even more important to also maintain accountability and transparency.
  - notyourday 9y ago
    > Not "may" but will - it's very easy to render your network essentially unauditable (you don't have thousands of petabytes' worth of storage available? you can't fab your own ML coprocessor? too bad!) Google can. So that does not apply. > You bring up Trump's Twitter suspension in the other subthread, but in that case the failure was transient, and the person responsible and their reasoning were quickly identified; YouTube has demonetized videos with little to no explanation or recourse. There's no recourse or explanation because it is not youtube's business model regardless of what/who makes a decision to demonetize them. Adding arbitrary human behavior to those decisions only makes it worse.