4 ms·

> OpenAI is clear about where it stands. “Our aim is to create general intelligence,” says Dhariwal. “Building models like DALL-E 2 that connect vision and lang

by dvt 5y ago

> OpenAI is clear about where it stands. “Our aim is to create general intelligence,” says Dhariwal. “Building models like DALL-E 2 that connect vision and language is a crucial step in our larger goal of teaching machines to perceive the world the way humans do, and eventually developing AGI.” (Emphasis mine.) Philosophical nit: this by definition would not be AGI. AGI would be the ability of the entity to figure out its own subjective qualia-inducing way of interpreting the world. Thomas Nagel already explained this in the 70s (in his famous essay What Is It Like to Be a Bat?). By enforcing our own worldview (technically speaking, model), we're just making some artificial (and in all likelihood inferior) version of a human.

chaps 5y ago
Reread it -- that's not what they're saying. They're suggesting that human-like perception is a stepping stone (ie, not necessarily a linear path towards). It reads to me as if they're acknowledging a significant gap between the two.
- dvt 5y ago
  > They're suggesting that human-like perception is a stepping stone I fully understand the discourse; my point is that it's not (Nagel, Chalmers, Searle†, and other substance dualists would agree). We need a breakthrough akin to what analysis was to algebra in neuroscience before we even understand what our brain is doing. Whatever it's doing, it's definitely not what OpenAI implemented. † Yes, I know he says he isn't, but he kind of is[1]. [1] http://www.edwardfeser.com/unpublishedpapers/searle.html http://www.edwardfeser.com/unpublishedpapers/searle.html
  - hackinthebochs 5y ago
    >Whatever it's doing, it's definitely not what OpenAI implemented. Defining intelligence as "the thing brains do" is a mistake. Planes fly, but not like birds. Biology is a great tool to understand some relationships between form and function, but mimicking biology isn't the end-goal.
    - dvt 5y ago
      > Defining intelligence as "the thing brains do" is a mistake This is a pretty bold claim. Brains are our only vantage points, and we also define life as "what's going on in a petri dish." But in any case, your point is self-defeating: what GPT-3 and DALL-E are doing is literally creating models about how we (humans) perceive the world.
      - hackinthebochs 5y ago
        >what GPT-3 and DALL-E are doing is literally creating models about how we (humans) perceive the world. Not really. They're creating models of how pixel-space covaries with word sequences over human-meaningful samples. These models are modelling the space of human meaning, not human perception.
  - anikan_vader 5y ago
    I understand the gist of your comment, but perhaps you could explain more regarding the relationship between analysis and algebra? Cauchy's work certainly fundamentally changed our understanding of calculus (in that Newton and Liebnitz's earlier hand-wavy proofs were validated with a rigorous definition of a limit), but I'm not aware of any large impact it had on Galois, who was a contemporary of Cauchy.
visarga 5y ago
Humans are also receptacles of culture creating new cultural artefacts, DALL-E is like us in that regard. But it's a tool AI, it's not a full agent. It's not embodied, embedded and enacted in the environment, its goal is to minimise loss, not to survive. I think the "what it's like to be an X?" comes from being in the environment with needs and goals to accomplish. It's not a magical neural architecture, it's the framing of the whole process that makes the difference.

by dvt 5y ago

chaps 5y ago

Reread it -- that's not what they're saying. They're suggesting that human-like perception is a stepping stone (ie, not necessarily a linear path towards). It reads to me as if they're acknowledging a significant gap between the two.

dvt 5y ago
> They're suggesting that human-like perception is a stepping stone I fully understand the discourse; my point is that it's not (Nagel, Chalmers, Searle†, and other substance dualists would agree). We need a breakthrough akin to what analysis was to algebra in neuroscience before we even understand what our brain is doing. Whatever it's doing, it's definitely not what OpenAI implemented. † Yes, I know he says he isn't, but he kind of is[1]. [1] http://www.edwardfeser.com/unpublishedpapers/searle.html http://www.edwardfeser.com/unpublishedpapers/searle.html
- hackinthebochs 5y ago
  >Whatever it's doing, it's definitely not what OpenAI implemented. Defining intelligence as "the thing brains do" is a mistake. Planes fly, but not like birds. Biology is a great tool to understand some relationships between form and function, but mimicking biology isn't the end-goal.
  - dvt 5y ago
    > Defining intelligence as "the thing brains do" is a mistake This is a pretty bold claim. Brains are our only vantage points, and we also define life as "what's going on in a petri dish." But in any case, your point is self-defeating: what GPT-3 and DALL-E are doing is literally creating models about how we (humans) perceive the world.
    - hackinthebochs 5y ago
      >what GPT-3 and DALL-E are doing is literally creating models about how we (humans) perceive the world. Not really. They're creating models of how pixel-space covaries with word sequences over human-meaningful samples. These models are modelling the space of human meaning, not human perception.
- anikan_vader 5y ago
  I understand the gist of your comment, but perhaps you could explain more regarding the relationship between analysis and algebra? Cauchy's work certainly fundamentally changed our understanding of calculus (in that Newton and Liebnitz's earlier hand-wavy proofs were validated with a rigorous definition of a limit), but I'm not aware of any large impact it had on Galois, who was a contemporary of Cauchy.

visarga 5y ago

Humans are also receptacles of culture creating new cultural artefacts, DALL-E is like us in that regard. But it's a tool AI, it's not a full agent. It's not embodied, embedded and enacted in the environment, its goal is to minimise loss, not to survive. I think the "what it's like to be an X?" comes from being in the environment with needs and goals to accomplish. It's not a magical neural architecture, it's the framing of the whole process that makes the difference.