6 ms·

I wonder if at some point the amount of extra data necessary to achieve an n-fold improvement will outstrip what we can provide. I think the time for AI legisl

I wonder if at some point the amount of extra data necessary to achieve an n-fold improvement will outstrip what we can provide. I think the time for AI legislation is now - before FAAMG deploys something like the next-gen of GPT-3. Of course with the legislative lag that exists even for decade-old tech I don't have the highest confidence in this being achieved by a federal government in the state it is in now.

rewq4321 6y ago
On that data point: I wonder if anyone can comment on how much useful training data we could get out of generating text based on knowledge graphs/databases that we have. You can construct an awful lot of sentences out of just a few facts (e.g. weights of various classes to generate sentences like: "x's are heavier than y's, but not as heavy as z's"). All the variations would contain the same information (of subsets of it), but the same could be said of lots of text online. Obviously this is an inefficient way to incorporate the databases into a GPT-like model, but it might make sense economically given the race that is now playing out - just shoehorn it in or you'll be left behind (at least in the short term) by those who do. "We can work out how to make it efficient after we're rolling around in cash." The knowledge databases could be used to generate what would essentially be "word problems" (in math classes), starting with simple things like "If I put three marbles in a cup, and then I take one out, and each marble weighs 20g, then the remaining marbles weigh 40g in total" and moving on to progressively more complex ones. If that were to happen, then you'd see companies employing people to create templates which essentially convert databases into sentences/paragraphs, which can then be consumed by the GPT-like model. It seems like this data would need to be used in a sort of pre-training step though, because you want the model to encode all the relationships, but you don't want it to learn to generate these types of concrete sentences, specifically.
inetsee 6y ago
As blueeyes has already pointed out "Legislation in one nation will simply handicap that nation." I don't have a lot of faith in our legislators ability to legislate safety without relegating us to an AI backwater.
- datameta 6y ago
  You're right. I think ideally the legislation should be international. Maybe something like the Washington Naval Treaty that set an upper limit on the tonnage and armament of new battleships. Or perhaps more aptly something akin to SALT I & II where older models are taken offline to avoid derelict AI systems from falling into malicious hands and to keep the number from growing out of control. Although this parallel is somewhat weak considering the capabilities of one advanced model are more valuable than 10x models of the last generation. Theoretical wishful thinking, I suppose, but I strongly believe that corp/govt scale ML research should be treated like advanced weaponry because it isn't a matter of if but when AI will be weaponized (whether the flavor of warfare is physical or informational). Although of course as with weapons treaties - the major powers would likely tend to be selective in what they commit to limiting themselves in.
  - Jach 6y ago
    The world couldn't even come together on controlling 3D printed weaponry, there's no hope for an arms treaty for AI right now. The "it's not feasible to regulate even if you tried" stance applies too -- you can restrict central actors without much difficulty, and that would work for AI just as well as it works for battleships, but there's a lot of distributed compute whereas there's not a lot of distributed shipyards. Like, you just have to follow what's been done with anime image nets to see that something like GPT-3 is possible for a distributed worldwide group to achieve and is not limited to firms or governments. Maybe when we have a disaster directly attributable to AI, nations can get on-board with something like the BWC and CWC. Until then, be even more pessimistic. (If you want a fun if rather dry book to read on material technology developments that were in the pipeline a couple decades ago, some of which have come to fruition, as well as some policy recommendations for the technologies that aren't generally good, check out Jürgen Altmann's Military Nanotechnology.)
    - sbierwagen 6y ago
      >The world couldn't even come together on controlling 3D printed weaponry Beg pardon? Plastic guns have been banned in the US since 1988 https://en.wikipedia.org/wiki/Undetectable_Firearms_Act https://en.wikipedia.org/wiki/Undetectable_Firearms_Act I assume other countries have similar bans.

rewq4321 6y ago

On that data point: I wonder if anyone can comment on how much useful training data we could get out of generating text based on knowledge graphs/databases that we have. You can construct an awful lot of sentences out of just a few facts (e.g. weights of various classes to generate sentences like: "x's are heavier than y's, but not as heavy as z's"). All the variations would contain the same information (of subsets of it), but the same could be said of lots of text online. Obviously this is an inefficient way to incorporate the databases into a GPT-like model, but it might make sense economically given the race that is now playing out - just shoehorn it in or you'll be left behind (at least in the short term) by those who do. "We can work out how to make it efficient after we're rolling around in cash." The knowledge databases could be used to generate what would essentially be "word problems" (in math classes), starting with simple things like "If I put three marbles in a cup, and then I take one out, and each marble weighs 20g, then the remaining marbles weigh 40g in total" and moving on to progressively more complex ones. If that were to happen, then you'd see companies employing people to create templates which essentially convert databases into sentences/paragraphs, which can then be consumed by the GPT-like model. It seems like this data would need to be used in a sort of pre-training step though, because you want the model to encode all the relationships, but you don't want it to learn to generate these types of concrete sentences, specifically.

inetsee 6y ago

As blueeyes has already pointed out "Legislation in one nation will simply handicap that nation." I don't have a lot of faith in our legislators ability to legislate safety without relegating us to an AI backwater.

datameta 6y ago
You're right. I think ideally the legislation should be international. Maybe something like the Washington Naval Treaty that set an upper limit on the tonnage and armament of new battleships. Or perhaps more aptly something akin to SALT I & II where older models are taken offline to avoid derelict AI systems from falling into malicious hands and to keep the number from growing out of control. Although this parallel is somewhat weak considering the capabilities of one advanced model are more valuable than 10x models of the last generation. Theoretical wishful thinking, I suppose, but I strongly believe that corp/govt scale ML research should be treated like advanced weaponry because it isn't a matter of if but when AI will be weaponized (whether the flavor of warfare is physical or informational). Although of course as with weapons treaties - the major powers would likely tend to be selective in what they commit to limiting themselves in.
- Jach 6y ago
  The world couldn't even come together on controlling 3D printed weaponry, there's no hope for an arms treaty for AI right now. The "it's not feasible to regulate even if you tried" stance applies too -- you can restrict central actors without much difficulty, and that would work for AI just as well as it works for battleships, but there's a lot of distributed compute whereas there's not a lot of distributed shipyards. Like, you just have to follow what's been done with anime image nets to see that something like GPT-3 is possible for a distributed worldwide group to achieve and is not limited to firms or governments. Maybe when we have a disaster directly attributable to AI, nations can get on-board with something like the BWC and CWC. Until then, be even more pessimistic. (If you want a fun if rather dry book to read on material technology developments that were in the pipeline a couple decades ago, some of which have come to fruition, as well as some policy recommendations for the technologies that aren't generally good, check out Jürgen Altmann's Military Nanotechnology.)
  - sbierwagen 6y ago
    >The world couldn't even come together on controlling 3D printed weaponry Beg pardon? Plastic guns have been banned in the US since 1988 https://en.wikipedia.org/wiki/Undetectable_Firearms_Act https://en.wikipedia.org/wiki/Undetectable_Firearms_Act I assume other countries have similar bans.