Connectionism and Embodiment

Question 1

What are two philosophical challenges to Symbolic AI?

Answer 1

The Chinese Room argument: Rule-based symbol-manipulation is not sufficient for understanding (which could be an important aspect of intelligence)

Lady Lovelace’s Objection: It is not the computer’s intelligence, but the programmer’s.

(Or the consciousness argument, but I believe consciousness is a bit out of scope of this course)

Question 2

What are two practical challenges to Symbolic AI?

Answer 2

The Frame problem: It is difficult to distinguish relevant from irrelevant problems.

For many problems, it is difficult to articulate rules that are both effective and tractable. Internal representation do not always seem intuitive for everything, because we cannot define strict rules about wat an object is or is not.

Question 3

Why can machine learning pose a solution for the practical challenges of Symbolic AI?

Answer 3

Human engineers don’t need to identify and articulate rules with which to solve a particular problem. Instead, they just need to specify a basic architecture, learning algorithm, and data environment in which the computer can learn the relevant rules on its own.

Machine learning is useful for solving problems whose solutions require rules which are implicit, subtle, and complex (and for this reason difficult for human programmers to understand).

Question 4

What are some limitations of neural networks according to the slides?

Answer 4

The more complex the problem and the longer-term its horizon, the more specific things need to be remembered for longer time. But since networks memorize facts by changing weights throughout the network, every newly-learned fact may cause“ catastrophic forgetting” of previously-learned facts.

Question 5

For solving problems that involve sensing and acting, it is often less important for the AI system to “think” abstractly and correctly than for it to exploit the way it is embodied and embedded.

What is meant with Embodiment and Embedding?

Answer 5

Embodiment: The way in which an AI system is physically
realized

Embedding: The physical location and surrounding environment in which an AI system operates.

Question 6

What is meant with embodied cognition according to the paper by L. Shapiro?

Answer 6

Embodied cognition is an approach to cognition that puts more emphasis on the significance of an organism’s body in how and what the organism thinks, in comparison to traditional ‘computational’ cognitive science.

In the paper, Shapiro also makes a case that organisms always needing to consult a representation about something seems redundant if the ‘representation’ is right in front of our eyes.

Question 7

Which three aspects of embodied cognition research were mentioned in the paper by Shapiro?

Answer 7

1. The body itself shapes how the mind processes information. An example is the distance between the ears are allowing us to hear where a sound is coming from.
2. The thoughts we have are shaped in relation to our body. The cognitive metaphor theory is an example of that. For example when we say we feel low, we feel sad. A being that is spherical and lives without gravity would not create the same metaphor.
3. Items in our environment are part of our cognition. Out phone is an extension of our cognition, and even pen and paper as well. We offload cognition on external things.