Michael Shulman's Shared Notes

Powered by 🌱Roam Garden

A second type, or dimension, of technology has captured our imaginations from the earliest days of science fiction: machines that can reason independently of humans, and make novel contributions. We are in the early days of Artificial Intelligence (AI), and machines cannot quite think on their own, but they can and do make physical and intellectual contributions that are not the direct products of queries from the humans they are working with. The intelligence of machines can range on how independent they are from their human counterparts.

  • In considering how machines can range in their intellectual ability, it is useful to consider varying levels of human intelligence. While all humans possess the ability to think and act independently, their ability to do so in a useful way varies considerably. At low levels of intelligence, humans can carry out repetitive tasks that they are taught how to do, and with rising intelligence comes the ability to learn more quickly and to reason independently. While machines and humans differ qualitatively in the way they find solutions to problems, the varying intelligences of humans does provide a useful analogy for understanding the variance in artificial intelligence.
  • Just like humans of simple intelligence, machines with low levels of AI can act in specified ways, carrying out repetitive tasks, usually following a set of if-then rules. Machines of this intelligence have been part of our world for decades: they may call the fire department if there is smoke, triage our emails, or keep a car at a set speed and slow it down when it approaches a car in front.
  • (Need to find that paper on the continuum of AI I came across sometime over the past week or two).
  • At low levels, machines act in specified ways, usually following a set of if-then rules. Machines of this intelligence have been part of our world for decades: they call the fire department if there's smoke, triage our emails, keep the car at a set speed and slow it down if it comes close to a car in front. *
  • At higher levels, machines can learn, taking information from the environment, making predictions about the future, and adjusting future predictions based on the accuracy of past predictions. At higher levels yet, machines can create new ideas. Emerging technologies allow machines to write essays, to to create art and design and unique illustrations. *
  • A current limitation of AI is that it needs to be fed a lot of information to learn. Machines need to see millions of pictures, paintings, and illustrations of an elephant to "understand" what it is. Humans, by contrast, can see just a single elephant, and then recognize a drawing or abstraction of it as an elephant (provided they already have developed nuanced enough schema to recognize it as a species unique from other, similar animals). Should AI develop the ability to extrapolate from limited sets of information, that would be a higher level of AI. And should AI ever reach the ability to evolve on its own, to adjust its own programming (software) and physical makeup (hardware), that might be the highest level of AI. *
Scribbles on dimensions of technology