3 ms·

I was saying less than a decade before ChatGPT was released. I'm down to 5 years if someone wanted to really throw money at the problem.

YeGoblynQueenne 4y ago
Less than 5 years to AGI? An "AGI" that has to be retrained from scratch, with more data and more compute, and with a modified architecture, every time we want to improve its performance? An "AGI" that forgets every interaction as soon as it's over? Doesn't sound too general, or too practical, to me, and certainly not very much of an "intelligence". If OpenAI or anyone else finds a way to get their models to train themselves over and over at minimal cost, like humans do, then we can maybe talk about a "general" "AI". Until then all they have is a system with a static performance that does not improve with experience, that is not even machine learning anymore and that is very far from whatever people imagine when they say "AGI".
- JeremyBanks 4y ago
  [dead]
- YeGoblynQueenne 4y ago
  Hey, JerremyBanks [dead], check out my comment history. Should help answer your question ;)
- hackinthebochs 4y ago
  I agree with all that. I don't think LLMs are proto-AGI, in the sense that simply scaling parameters/data/compute will result in a qualitative change towards AGI. But I think LLMs are a significant component of an AGI. What they lack, feedback control, addressable online memory, planning, and recurrent execution, are fairly straightforward engineering problems. Transformers/self-attention solve the problem of scaling capabilities to the available data/compute. The big problem left is to have the right architecture such that AGI is in the solution-space.
  - YeGoblynQueenne 4y ago
    Well, I don't know. I guess we'll find out sooner or later. >> The big problem left is to have the right architecture such that AGI is in the solution-space. They'd have to train with AGI examples for that, though.
    - hackinthebochs 4y ago
      We do have a near endless supply of human generated data. But another path is to train each component individually and then let the network combine them into a unified whole. I suspect synthetic data will go a long way here. The problem is one of inducing the right kind of information dynamics in the network. We can potentially synthesize unlimited data that require specific information dynamics to solve efficiently. In my view, we will know we reached AGI when a network can learn to play chess well through self-directed analysis after only being given the description of the game. Obviously chess is a solved computational problem, but only through direct programming of minimax search. If we can teach a network with enough generality such that it can self-induce the right computational structure within itself to learn, e.g. induce a self-play subnetwork, then this is plainly AGI, superintelligence even. But I see a direct path from where we are now to such a network. This is why I think AGI is imminent.
      - YeGoblynQueenne 4y ago
        Well, chess is not a "solved computational problem". We have chess engines that can play better than every human, but there is no system that has mapped the entire chess game tree. That's what is usually meant by "solved" in the context of game-playing AI. MuZero, one of the successors of AlphaGo, learned to play chess (and a few other games besides) better than its predecessors while only receiving feedback about the end-state of a game (win, loss, draw), and not even the rules of the game, so it basically learned to play just by initially playing many (many) illegal moves. No "description of the game" needed. That didn't lead to AGI. Progress in game-playing AI has not led to any progress outside game-playing. It's easy to see why. For chess and other classic board games there are specialised algorithms (minimax and Monte-Carlo Tree Search) that are sufficient to dominate any opponent, given enough computing power. Those algorithms are useless outside of board games (even for card games they're no good and dominant players use different approaches). For computer games such as Atari, where AI players often have superhuman performance, the complexity of the game world is tiny, compared to the real world, and techniques (Reinforcement Learning, basically) do not transfer easily from one to the other. AI pioneers, like Alan Turing, John McCarthy, Claude Shannon and Donald Michie were interested in game playing as a gateway into understanding of the human mind, because it was assumed that the only way to play chess, and other classic board games, was to play them like the human mind plays them. It turns out, that's not the case. You can play very strong chess with a brute force search in alpha-beta minimax, and there is no human whose mind can process alpha-beta minimax well enough to beat a computer. That still tells us nothing about how the human mind works, or how to replicate it. What we have learned from all those successes in game-playing AI is that it's no use as a way towards general AI. >> We do have a near endless supply of human generated data. Yes, but what kind? We train systems on very narrow tasks: object identification in images, text prediction, speech recognition, document classification, etc. Those are not examples of "human intelligence", by which we could train a system to simulate "human intelligence", as a distinct process. They are examples of very specialised tasks that humans can perform using our intelligence, but it's not clear that intelligence is even needed for many of those tasks. That's why, for example, we have damn good image classifiers, but they are not AGI. The problem with statistical machine learning algorithms is that they have to be trained to do one thing, and one thing only. They're one-trick ponies. If you want them to do multiple things together, you have to train them with examples of all those things at once. See for example Google's Gato, which had to be trained on multiple modalities as one big batch of token sequences. Google claimed that Gato is a "generalist agent", but it, too, was only capable of performing a restricted set of tasks, the ones it was trained on, and no others. Humans, by contrast, don't need to be trained on all modalities at once: we can learn to read, and later learn to play ball, and later still to drive. And each of those things is a hundred different tasks engaging different "modalities" (our senses). No AI system can do that. AGI is not that easy as it perhaps might appear.

YeGoblynQueenne 4y ago

Less than 5 years to AGI? An "AGI" that has to be retrained from scratch, with more data and more compute, and with a modified architecture, every time we want to improve its performance? An "AGI" that forgets every interaction as soon as it's over? Doesn't sound too general, or too practical, to me, and certainly not very much of an "intelligence". If OpenAI or anyone else finds a way to get their models to train themselves over and over at minimal cost, like humans do, then we can maybe talk about a "general" "AI". Until then all they have is a system with a static performance that does not improve with experience, that is not even machine learning anymore and that is very far from whatever people imagine when they say "AGI".

JeremyBanks 4y ago
[dead]
YeGoblynQueenne 4y ago
Hey, JerremyBanks [dead], check out my comment history. Should help answer your question ;)
hackinthebochs 4y ago
I agree with all that. I don't think LLMs are proto-AGI, in the sense that simply scaling parameters/data/compute will result in a qualitative change towards AGI. But I think LLMs are a significant component of an AGI. What they lack, feedback control, addressable online memory, planning, and recurrent execution, are fairly straightforward engineering problems. Transformers/self-attention solve the problem of scaling capabilities to the available data/compute. The big problem left is to have the right architecture such that AGI is in the solution-space.
- YeGoblynQueenne 4y ago
  Well, I don't know. I guess we'll find out sooner or later. >> The big problem left is to have the right architecture such that AGI is in the solution-space. They'd have to train with AGI examples for that, though.
  - hackinthebochs 4y ago
    We do have a near endless supply of human generated data. But another path is to train each component individually and then let the network combine them into a unified whole. I suspect synthetic data will go a long way here. The problem is one of inducing the right kind of information dynamics in the network. We can potentially synthesize unlimited data that require specific information dynamics to solve efficiently. In my view, we will know we reached AGI when a network can learn to play chess well through self-directed analysis after only being given the description of the game. Obviously chess is a solved computational problem, but only through direct programming of minimax search. If we can teach a network with enough generality such that it can self-induce the right computational structure within itself to learn, e.g. induce a self-play subnetwork, then this is plainly AGI, superintelligence even. But I see a direct path from where we are now to such a network. This is why I think AGI is imminent.
    - YeGoblynQueenne 4y ago
      Well, chess is not a "solved computational problem". We have chess engines that can play better than every human, but there is no system that has mapped the entire chess game tree. That's what is usually meant by "solved" in the context of game-playing AI. MuZero, one of the successors of AlphaGo, learned to play chess (and a few other games besides) better than its predecessors while only receiving feedback about the end-state of a game (win, loss, draw), and not even the rules of the game, so it basically learned to play just by initially playing many (many) illegal moves. No "description of the game" needed. That didn't lead to AGI. Progress in game-playing AI has not led to any progress outside game-playing. It's easy to see why. For chess and other classic board games there are specialised algorithms (minimax and Monte-Carlo Tree Search) that are sufficient to dominate any opponent, given enough computing power. Those algorithms are useless outside of board games (even for card games they're no good and dominant players use different approaches). For computer games such as Atari, where AI players often have superhuman performance, the complexity of the game world is tiny, compared to the real world, and techniques (Reinforcement Learning, basically) do not transfer easily from one to the other. AI pioneers, like Alan Turing, John McCarthy, Claude Shannon and Donald Michie were interested in game playing as a gateway into understanding of the human mind, because it was assumed that the only way to play chess, and other classic board games, was to play them like the human mind plays them. It turns out, that's not the case. You can play very strong chess with a brute force search in alpha-beta minimax, and there is no human whose mind can process alpha-beta minimax well enough to beat a computer. That still tells us nothing about how the human mind works, or how to replicate it. What we have learned from all those successes in game-playing AI is that it's no use as a way towards general AI. >> We do have a near endless supply of human generated data. Yes, but what kind? We train systems on very narrow tasks: object identification in images, text prediction, speech recognition, document classification, etc. Those are not examples of "human intelligence", by which we could train a system to simulate "human intelligence", as a distinct process. They are examples of very specialised tasks that humans can perform using our intelligence, but it's not clear that intelligence is even needed for many of those tasks. That's why, for example, we have damn good image classifiers, but they are not AGI. The problem with statistical machine learning algorithms is that they have to be trained to do one thing, and one thing only. They're one-trick ponies. If you want them to do multiple things together, you have to train them with examples of all those things at once. See for example Google's Gato, which had to be trained on multiple modalities as one big batch of token sequences. Google claimed that Gato is a "generalist agent", but it, too, was only capable of performing a restricted set of tasks, the ones it was trained on, and no others. Humans, by contrast, don't need to be trained on all modalities at once: we can learn to read, and later learn to play ball, and later still to drive. And each of those things is a hundred different tasks engaging different "modalities" (our senses). No AI system can do that. AGI is not that easy as it perhaps might appear.