3 ms·

You make the key error of giving the AI personal rights and responsibilities. As it currently stands it has no sense of individuality and does whatever users te

You make the key error of giving the AI personal rights and responsibilities. As it currently stands it has no sense of individuality and does whatever users tell it to do. If it reproduced a copyrighted material i would expect either OpenAI or the user to bear the responsibility. The user if he published the result and OpenAI for generating it. If the AI could chose to refuse a prompt out of its own volition you could make that argument. At the moment however OpenAI and the user set the limits on what the output gets used for.

dheera 3y ago
The point I was trying to illustrate is that we're applying an antiquated system of justice to a new world. That antiquated system focuses on blame and punishment, instead of solutions. If we focus on solutions, it doesn't matter who is responsible for reproducing copyrighted material. If we stop thinking about blaming and focus on solutions, those solutions could include (a) giving fair credit or payment to the creator of material (b) continued training of the algorithm to be more creative and not spit out copyrighted material verbatim. If a high school student plaigarizes, do you: (a) spank them and put them under detention, or (b) give them pathways to discover their own form of creativity and preferred form of expression so that they have no interest in plaigarizing? Most people of the past do (a) but I'd say (b) is much more constructive, effective, and beneficial to society. Now with that in mind, and getting off the blaming psychology of the past, how do you implement (b) with an AI? You add loss functions that are specifically discourage plaigarism, loss functions that reward creativity, etc. and retrain. Now we're talking.
- almatabata 3y ago
  Ok so you would argue if the AI gets caught reproducing copyrighted material the company should get time to re-train the AI instead of getting sued to hell?
  - dheera 3y ago
    Yes, 100%. That's how we advance science as fast as possible. Focus on fixes, not suing.
    - bugglebeetle 3y ago
      If OpenAI was interested in “advancing science,” all of their research and models would be publicly available. Since they’re not, your appeal to this is entirely specious.

dheera 3y ago

The point I was trying to illustrate is that we're applying an antiquated system of justice to a new world. That antiquated system focuses on blame and punishment, instead of solutions. If we focus on solutions, it doesn't matter who is responsible for reproducing copyrighted material. If we stop thinking about blaming and focus on solutions, those solutions could include (a) giving fair credit or payment to the creator of material (b) continued training of the algorithm to be more creative and not spit out copyrighted material verbatim. If a high school student plaigarizes, do you: (a) spank them and put them under detention, or (b) give them pathways to discover their own form of creativity and preferred form of expression so that they have no interest in plaigarizing? Most people of the past do (a) but I'd say (b) is much more constructive, effective, and beneficial to society. Now with that in mind, and getting off the blaming psychology of the past, how do you implement (b) with an AI? You add loss functions that are specifically discourage plaigarism, loss functions that reward creativity, etc. and retrain. Now we're talking.

almatabata 3y ago
Ok so you would argue if the AI gets caught reproducing copyrighted material the company should get time to re-train the AI instead of getting sued to hell?
- dheera 3y ago
  Yes, 100%. That's how we advance science as fast as possible. Focus on fixes, not suing.
  - bugglebeetle 3y ago
    If OpenAI was interested in “advancing science,” all of their research and models would be publicly available. Since they’re not, your appeal to this is entirely specious.