This page summarizes some of the themes of our discussion in class Nov 12.
Here are some highlights that people resonated with.
From Dennett, Kinds of Minds, discussed Sep 24.
The story: Our hero keeps track of one particular object, despite the fact that there are many others like it, exploiting conceptual, perceptual and behavioral abilities. Then something goes wrong, our hero loses track: the object is right in front of him but he can no longer identify it.
Variations: a lucky penny, thrown into a fountain. Pumpkins in a pumpkin patch. Peeps.
From Agre, Computation and Human Understanding, Discussed Sep 24.
The story: Agents can act effectively in the world without keeping track of objects' identities. Imagine what it would mean to recognize each fork you use as the same or different from all the other forks you've used in your life. It would be like looking up each fork's identity in a big database, a "cosmic registry". You can imagine somebody at a big, dreary, imposing desk, Brazil-style, flipping pages in the registry, trying to find your fork.
Variations: similar to the lucky penny.
From Putnam, The Meaning of 'Meaning', Discussed Sep 24.
The story: Two agents have identical experiences on different planets. The only difference is that there is water on Earth, and XYZ on Twin Earth. We think about water, they think about XYZ.
Variations: any natural kind can take the place of water. Stock examples: aluminum, molybdenum, elm trees, beeches.
From Dennett, Evolution, error and intentionality, Discussed Oct 1.
The story: People are placed in suspended animation in caretaker robots; the robots have general capabilities to preserve their human occupants. Twilight zone twist: we are these robots, carrying our genes. The thinking poiint is that maybe even people have only derived intentionality.
We have a bunch of experience that we can draw on ourselves. This includes both concepts and explanations, and perhaps even visual material and demonstrations prepared with our own robots.
After Oved, Mental Coining of Terms, Discussed Oct 15.
The story: We build a robot to get us oranges (from the fridge, say). What we want is oranges, the real thing, not just things that look, feel, or even taste like oranges. There's lots of room in this story for funny mistakes when the robot brings you back things that are not oranges, to illustrate the role of concepts in allowing robots to learn, to categorize, to interact with people, etc.
Variation: Again, any natural kind would work here.
From the Rutgers Lab for Real-Life Reinforcement Learning, in relation to Cohen et al, Contentful mental states for robot baby, Discussed Oct 22.
The story: we watch a robot behave, and it seems to exhibit sophisticated behavior. But we open the hood, and we see that its connection with the world is mediated by fairly specific and 'unnatural' data structures. It is difficult to find any commonsensical representations there.
Here are some ideas that we latched onto for presenting the issues in an engaging way.
Having an "inner monologue" as a voiceover that describes ongoing events in a way that more accurately captures the mechanisms behind the robot's behavior.
Showing the world in a way that indicates the way that the robot perceives, organizes or categorizes the world. An animation could present the world as a buzzing, blooming confusion; as (untrackable) indistinguishable objects; with labels for the identities of objects; or for perceptual features; or for underlying categories…
As a final note, there are some features of cognition that connect with arguments about representations and that seemed particularly challenging but important to get across:
And there are some mechanisms that are implicated in realizing these features: