4 ms·

its probably finding bounding boxes based on highest probability %. So, it's a shame the refs head looks more like a soccer ball than the soccer ball!

by grp000 6y ago

hda2 6y ago
.. to the AI. The fundamental problem is that no training set can ever be really complete. Humans have general intelligence to deal with this. ML does not.
- novia 6y ago
  the biggest problem is that ML/AI has no conception of time, context, continuity, or cause and effect.
  - NateEag 6y ago
    Which is why some of us avoid calling these systems "AI". Yes, the term has become accepted, but it's a horribly irresponsible label because it massively misleads laypeople about how these systems work and how trustworthy they are.
    - gentleman11 6y ago
      What’s a better catch all term? Data modelling system? Prediction system?
      - Reelin 6y ago
        ... machine learning? Technically AI is correct, since AI doesn't necessarily refer to human level AGI. Unfortunately it's highly misleading because that distinction seems to be lost on the average person. Machine learning doesn't seem to carry the same science fiction implications for whatever reason.
    - joveian 6y ago
      I wouldn't say "has become" since the term was already in widspread use in games by the early 90s at least if not earlier. I would guess many people are first exposed to the term in games and I would be surprised if the expectation you describe is actually widespread at all. Games or not, it seems like the first exposure anyone has to anything called AI would cure them of any such misconceptions.
      - NateEag 6y ago
        I meant it had become widespread in describing ML / DL systems, specifically.
  - Reelin 6y ago
    > ML/AI has no conception of Time: False. Time can be included as an input. Previous system state can also be communicated forward in various ways (ex the LSTM architecture). Conceptual context: I'll agree, since that seems like it would require general abstract reasoning ability (ie strong AI). Similarly, robust treatment in the general case of physical continuity and cause and effect both seem like special cases of conceptual context to me. That being said, bear in mind that we already have examples of not so robust treatment for special cases of physical continuity as well as cause and effect. (Example: https://news.developer.nvidia.com/transforming-standard-video-into-slow-motion-with-ai/ https://news.developer.nvidia.com/transforming-standard-vide...) Who knows how close (or far) we might be from a general solution?
    - novia 6y ago
      > Time: False. Time can be included as an input. I mean that they don't understand time as something that's unidirectional. (To the best of my knowledge! Would be happy to be proven wrong.)
      - Reelin 6y ago
        Well I guess without abstract reasoning ability (and thus no conceptual context in the human sense) "understanding" in the way the word is typically used is presumably impossible. But I don't see any reason you couldn't train a model to incorporate the underlying assumption that the flow of time is unidirectional. I expect you would just need inputs (ie training data) reflecting that fact coupled with an appropriate loss function. (Aren't networks that predict future physical states, such as motion, more or less doing this?)
  - ad404b8a372f2b9 6y ago
    Maybe not the model they used but AI in general does. A simple model like P(ball is in position y at time t1 | ball is in position x at time t0) accounts for all of that.
- grp000 6y ago
  I think in this case, the fundamental problem that other replies also stated is the lack of a generality of context. The system was trained to recognize balls, but it had no concept of a person. Train it on balls and people, and I'm sure the performance would improve because you can tell the system to not track things classified as a person. But then your system is also trying to draw boxes around people. So, now maybe your BB (bounding box) for people and ball keep overlapping. Now, you train it on image segmentation (draw an outline around objects) as opposed to bounding box selection and the performance also improves. The downside is now you require significantly more compute power on hand/lower sampling rate, etc. I think it is a testament to not just human complexity, but human efficiency.

its probably finding bounding boxes based on highest probability %. So, it's a shame the refs head looks more like a soccer ball than the soccer ball!

by grp000 6y ago

its probably finding bounding boxes based on highest probability %. So, it's a shame the refs head looks more like a soccer ball than the soccer ball!

hda2 6y ago

.. to the AI. The fundamental problem is that no training set can ever be really complete. Humans have general intelligence to deal with this. ML does not.

novia 6y ago
the biggest problem is that ML/AI has no conception of time, context, continuity, or cause and effect.
- NateEag 6y ago
  Which is why some of us avoid calling these systems "AI". Yes, the term has become accepted, but it's a horribly irresponsible label because it massively misleads laypeople about how these systems work and how trustworthy they are.
  - gentleman11 6y ago
    What’s a better catch all term? Data modelling system? Prediction system?
    - Reelin 6y ago
      ... machine learning? Technically AI is correct, since AI doesn't necessarily refer to human level AGI. Unfortunately it's highly misleading because that distinction seems to be lost on the average person. Machine learning doesn't seem to carry the same science fiction implications for whatever reason.
  - joveian 6y ago
    I wouldn't say "has become" since the term was already in widspread use in games by the early 90s at least if not earlier. I would guess many people are first exposed to the term in games and I would be surprised if the expectation you describe is actually widespread at all. Games or not, it seems like the first exposure anyone has to anything called AI would cure them of any such misconceptions.
    - NateEag 6y ago
      I meant it had become widespread in describing ML / DL systems, specifically.
- Reelin 6y ago
  > ML/AI has no conception of Time: False. Time can be included as an input. Previous system state can also be communicated forward in various ways (ex the LSTM architecture). Conceptual context: I'll agree, since that seems like it would require general abstract reasoning ability (ie strong AI). Similarly, robust treatment in the general case of physical continuity and cause and effect both seem like special cases of conceptual context to me. That being said, bear in mind that we already have examples of not so robust treatment for special cases of physical continuity as well as cause and effect. (Example: https://news.developer.nvidia.com/transforming-standard-video-into-slow-motion-with-ai/ https://news.developer.nvidia.com/transforming-standard-vide...) Who knows how close (or far) we might be from a general solution?
  - novia 6y ago
    > Time: False. Time can be included as an input. I mean that they don't understand time as something that's unidirectional. (To the best of my knowledge! Would be happy to be proven wrong.)
    - Reelin 6y ago
      Well I guess without abstract reasoning ability (and thus no conceptual context in the human sense) "understanding" in the way the word is typically used is presumably impossible. But I don't see any reason you couldn't train a model to incorporate the underlying assumption that the flow of time is unidirectional. I expect you would just need inputs (ie training data) reflecting that fact coupled with an appropriate loss function. (Aren't networks that predict future physical states, such as motion, more or less doing this?)
- ad404b8a372f2b9 6y ago
  Maybe not the model they used but AI in general does. A simple model like P(ball is in position y at time t1 | ball is in position x at time t0) accounts for all of that.
grp000 6y ago
I think in this case, the fundamental problem that other replies also stated is the lack of a generality of context. The system was trained to recognize balls, but it had no concept of a person. Train it on balls and people, and I'm sure the performance would improve because you can tell the system to not track things classified as a person. But then your system is also trying to draw boxes around people. So, now maybe your BB (bounding box) for people and ball keep overlapping. Now, you train it on image segmentation (draw an outline around objects) as opposed to bounding box selection and the performance also improves. The downside is now you require significantly more compute power on hand/lower sampling rate, etc. I think it is a testament to not just human complexity, but human efficiency.