4 ms·

Does all AI need a learning component?

4 points by andrew-lucker 10y ago

mindcrime 10y ago
The importance of "GOFAI" techniques was one of the themes in my talk at All Things Open last year. I wanted to cover the "state of the art" in open source machine learning tools, but realized that that just promoted FDD - Fad Driven Development, so I spent a modest amount of time during the talk going over some of the older ideas and pointing out how they still had relevance, and suggesting that the Next Big Thing might involve some hybridization of GOFAI techniques and probabilistic / machine learning ideas (including deep learning). So I guess I'd basically agree with the author here. The only thing I wish he/she would have done, is add some links to papers or whatever, highlighting what they consider to be some of the important new results in AI. There's definitely still plenty of work ongoing that falls under the "AI" rubric, see the cs.AI category on Arxiv.org for evidence of that. Of course some things get cross-posted with the ML category as well, so it's not all "GOFAI". http://arxiv.org/list/cs.AI/recent http://arxiv.org/list/cs.AI/recent
- andrew-lucker 10y ago
  Sorry for the lack of substance. I don't like to get too specific on Medium, but rather just use it to dump general feelings and emotion. That seems to be the community standard too. Reposting to HN is different though. What gave me the original feeling of FDD rather than full spectrum research is all of the things that don't get reposted or show up on news. I am sure that people are still doing that research, it just doesn't "have a hat on", as one of my professors would say.
  - mindcrime 10y ago
    Sorry for the lack of substance. I don't like to get too specific on Medium, but rather just use it to dump general feelings and emotion. That seems to be the community standard too. No worries, I didn't intend that to be a heavy criticism. Just a passing observation, that it would be nice to see some pointers to the things that you are excited about.
    - andrew-lucker 10y ago
      My personal passion is linguistics, and I find this whole LSTM or probablistic grammar stuff ridiculous. It would be nice to see some research on connecting knowledge models rather than just trying to mimick speech like a parrot.

4 points by andrew-lucker 10y ago

The importance of "GOFAI" techniques was one of the themes in my talk at All Things Open last year. I wanted to cover the "state of the art" in open source machine learning tools, but realized that that just promoted FDD - Fad Driven Development, so I spent a modest amount of time during the talk going over some of the older ideas and pointing out how they still had relevance, and suggesting that the Next Big Thing might involve some hybridization of GOFAI techniques and probabilistic / machine learning ideas (including deep learning). So I guess I'd basically agree with the author here. The only thing I wish he/she would have done, is add some links to papers or whatever, highlighting what they consider to be some of the important new results in AI. There's definitely still plenty of work ongoing that falls under the "AI" rubric, see the cs.AI category on Arxiv.org for evidence of that. Of course some things get cross-posted with the ML category as well, so it's not all "GOFAI". http://arxiv.org/list/cs.AI/recent http://arxiv.org/list/cs.AI/recent

andrew-lucker 10y ago
Sorry for the lack of substance. I don't like to get too specific on Medium, but rather just use it to dump general feelings and emotion. That seems to be the community standard too. Reposting to HN is different though. What gave me the original feeling of FDD rather than full spectrum research is all of the things that don't get reposted or show up on news. I am sure that people are still doing that research, it just doesn't "have a hat on", as one of my professors would say.
- mindcrime 10y ago
  Sorry for the lack of substance. I don't like to get too specific on Medium, but rather just use it to dump general feelings and emotion. That seems to be the community standard too. No worries, I didn't intend that to be a heavy criticism. Just a passing observation, that it would be nice to see some pointers to the things that you are excited about.
  - andrew-lucker 10y ago
    My personal passion is linguistics, and I find this whole LSTM or probablistic grammar stuff ridiculous. It would be nice to see some research on connecting knowledge models rather than just trying to mimick speech like a parrot.