5 ms·

Is this really AI? This seems like simple classification and ranking. Honestly I didn't see anything new in there that hasn't been around for the past 10 years.

Is this really AI? This seems like simple classification and ranking. Honestly I didn't see anything new in there that hasn't been around for the past 10 years. KNN? NDCG? That's entry level ML. TFA does throw around neural networks a bit, but doesn't go into any detail. EDIT: Maybe I'm just thrown off by the "Powered by AI" part of the article title. I was expecting more I suppose.

brenden2 7y ago
Anything moderately related to software and data is called AI now, because that's what gets the upvotes.
- thosakwe 7y ago
  AFAIK classification, deep learning, ML, etc. all are valid subsets of AI
  - brenden2 7y ago
    It's more like complex linear regressions. The term "AI" evokes the idea of computers possessing something similar to human intelligence which it very much is not. We don't even properly understand how the human brain works.
    - brootstrap 7y ago
      Yeah i am totally with you. AI can mean linear regressions and often does. "look Ma i'm writing my own AI algorithm!!!" ... "writes linear regression by hand in python"
    - baddox 7y ago
      There’s no need for AI to actually work the same way that the human brain works, and that’s never been part of the definition of “AI.”
- bigbadgoose 7y ago
  During a recent discussion on this topic, this emerged with regard to "what is AI": A. System should, without prompting, identify areas of improvement and innovation B. Automated collection of data and the processing thereof, combined with application towards a concrete goal — does not qualify under A. C. Part of A. is willing and unwilling discovery and exposure to both benevolent and adversarial environments and operating conditions
  - brenden2 7y ago
    This is just moving the goalposts around so that you can take advantage of the hype boost you get from calling everything "AI".
confeit 7y ago
Just read it as "Powered by big data" or "Powered by software" or "Powered by mathematics". It sounds fake and dumb. If it sounds fake and dumb to you, a ML practicioner, then maybe this article was not written for you. If you read the article itself, you get the feeling that AI was tacked on at the last minute, likely with an intent that has nothing to do with knowledge sharing.
CabSauce 7y ago
Embeddings have only really been popularized in the last few years. The 'original' word2vec paper was 2013. It's really only been maybe two years that using embeddings on other things has been popularized.
- tanilama 7y ago
  http://www.jmlr.org/papers/volume3/bengio03a/bengio03a.pdf http://www.jmlr.org/papers/volume3/bengio03a/bengio03a.pdf Bengio has a paper in 2003 that describes almost the same idea as word2vec (CBOW model to be exact).
baddox 7y ago
Why would something need to use techniques less than 10 years old to qualify as being AI? The field of artificial intelligence has been around for well over 50 years. Long before the most recent boom in AI research and marketing, AI has been a serious academic field and has been taught as part of a standard computer science curriculum. Very basic techniques like A* search are widely recognized as being AI techniques. There’s really no point in having a semantic argument, but just know that if you wish to do so you are arguing against many decades of wide usage of the term.
huac 7y ago
the concepts are simple but doing it at scale is hard. there are some good insights on scaling here: * scaling to more engineers/products: IGQL is an interesting way to compose ML pipelines with straightforward syntax * scaling to more ranking candidates: an active user with a large follow graph who loads the explore tab likely has millions of eligible candidates - how do you load those fast? the idea of using a "distilled" model as a first, light ranking before using a full model as the final predictor is a good intuitive idea that I haven't seen described before. * scaling KNN is hard: FB has done interesting work to make approximate nearest neighbor search fast, and opensourced it (the FAISS library which is referred to in the post). the improvements here are certainly non-trivial. * scaling to more users: creating useful general purpose user embeddings is hard! * scaling to more objectives: instagram has many business objectives, e.g. likes, follows, minimizing hides, so there is a need to have multiple models making many predictions. There is also a need to weight them intelligently, which is where the Bayesian optimization libraries come in. in some sense, nothing is truly AI, but this is useful work which you can learn a lot from.
tanilama 7y ago
True. It sets rightfully under the engineering category, because all this does is just engineering. At scale is challenging, but with Facebook's resources, it is more like problem waiting to be tackled, but the stakes are not high, because it CAN be tackled, just how much resources you want to pour into. What makes this mildly interesting is the IGQL, but again, without knowing the full syntax, it feels pretty restrictive. Youtube is doing much more advanced stuff, like Reinforcement learning@Scale, as comparing to Instagram in this regards.

brenden2 7y ago

Anything moderately related to software and data is called AI now, because that's what gets the upvotes.

thosakwe 7y ago
AFAIK classification, deep learning, ML, etc. all are valid subsets of AI
- brenden2 7y ago
  It's more like complex linear regressions. The term "AI" evokes the idea of computers possessing something similar to human intelligence which it very much is not. We don't even properly understand how the human brain works.
  - brootstrap 7y ago
    Yeah i am totally with you. AI can mean linear regressions and often does. "look Ma i'm writing my own AI algorithm!!!" ... "writes linear regression by hand in python"
  - baddox 7y ago
    There’s no need for AI to actually work the same way that the human brain works, and that’s never been part of the definition of “AI.”
bigbadgoose 7y ago
During a recent discussion on this topic, this emerged with regard to "what is AI": A. System should, without prompting, identify areas of improvement and innovation B. Automated collection of data and the processing thereof, combined with application towards a concrete goal — does not qualify under A. C. Part of A. is willing and unwilling discovery and exposure to both benevolent and adversarial environments and operating conditions
- brenden2 7y ago
  This is just moving the goalposts around so that you can take advantage of the hype boost you get from calling everything "AI".

confeit 7y ago

Just read it as "Powered by big data" or "Powered by software" or "Powered by mathematics". It sounds fake and dumb. If it sounds fake and dumb to you, a ML practicioner, then maybe this article was not written for you. If you read the article itself, you get the feeling that AI was tacked on at the last minute, likely with an intent that has nothing to do with knowledge sharing.

CabSauce 7y ago

Embeddings have only really been popularized in the last few years. The 'original' word2vec paper was 2013. It's really only been maybe two years that using embeddings on other things has been popularized.

tanilama 7y ago
http://www.jmlr.org/papers/volume3/bengio03a/bengio03a.pdf http://www.jmlr.org/papers/volume3/bengio03a/bengio03a.pdf Bengio has a paper in 2003 that describes almost the same idea as word2vec (CBOW model to be exact).

baddox 7y ago

Why would something need to use techniques less than 10 years old to qualify as being AI? The field of artificial intelligence has been around for well over 50 years. Long before the most recent boom in AI research and marketing, AI has been a serious academic field and has been taught as part of a standard computer science curriculum. Very basic techniques like A* search are widely recognized as being AI techniques. There’s really no point in having a semantic argument, but just know that if you wish to do so you are arguing against many decades of wide usage of the term.

huac 7y ago

the concepts are simple but doing it at scale is hard. there are some good insights on scaling here: * scaling to more engineers/products: IGQL is an interesting way to compose ML pipelines with straightforward syntax * scaling to more ranking candidates: an active user with a large follow graph who loads the explore tab likely has millions of eligible candidates - how do you load those fast? the idea of using a "distilled" model as a first, light ranking before using a full model as the final predictor is a good intuitive idea that I haven't seen described before. * scaling KNN is hard: FB has done interesting work to make approximate nearest neighbor search fast, and opensourced it (the FAISS library which is referred to in the post). the improvements here are certainly non-trivial. * scaling to more users: creating useful general purpose user embeddings is hard! * scaling to more objectives: instagram has many business objectives, e.g. likes, follows, minimizing hides, so there is a need to have multiple models making many predictions. There is also a need to weight them intelligently, which is where the Bayesian optimization libraries come in. in some sense, nothing is truly AI, but this is useful work which you can learn a lot from.

tanilama 7y ago

True. It sets rightfully under the engineering category, because all this does is just engineering. At scale is challenging, but with Facebook's resources, it is more like problem waiting to be tackled, but the stakes are not high, because it CAN be tackled, just how much resources you want to pour into. What makes this mildly interesting is the IGQL, but again, without knowing the full syntax, it feels pretty restrictive. Youtube is doing much more advanced stuff, like Reinforcement learning@Scale, as comparing to Instagram in this regards.