• 21 Posts
  • 463 Comments
Joined 2 years ago
cake
Cake day: June 27th, 2023


  • That paragraph begins,

    Like his predecessor critics of artificial intelligence, Taube, Dreyfus and Lighthill, Weizenbaum is impatient, implying that if the problem hasn’t been solved in twenty years, it is lime to give up.

    Weizenbaum replies,

    I do not say and I do not believe that “if the problem hasn’t been solved in twenty years, we should give up”. I say (p. 198) " . . . it would be wrong . . . to make impossibility arguments about what computers can do entirely on the grounds of our present ignorance". That is quite the opposite of what McCarthy charges me with saying.

    It’s a snidely jokey response to an argument that Weizenbaum didn’t make!





  • Scientists and philosophers have spilled a tanker truck of ink about the question of how to demarcate science from non-science or define pseudoscience rigorously. But we can bypass all that, because the basic issue is in fact very simple. One of the most fundamental parts of living a scientific life is admitting that you don’t know what you don’t know. Without that, it’s well-nigh impossible to do the work. Meanwhile, the generative AI industry is built on doing exactly the opposite. By its very nature, it generates slop that sounds confident. It is, intrinsically and fundamentally, anti-science.

    Now, on top of that, while being anti-science the AI industry also mimics the form of science. Look at all the shiny PDFs! They’ve got numbers in them and everything. Tables and plots and benchmarks! I think that any anti-science activity that steals the outward habits of science for its own purposes will qualify as pseudoscience, by any sensible definition of pseudoscience. In other words, wherever we draw the line or paint the gray area, modern “AI” will be on the bad side of it.