Philosophical fundamentals of AI - Lecture 2
What are the two main parts of the lecture "Philosophical fundamentals of AI"?
Part 1 is "Practitioner foundations" and Part 2 is "Philosophical foundations for intelligent machines" (Page 2).
How does the lecture define "AI" (Artificial Intelligence)?
(designed) learning + reasoning machines (Page 3).
How does the lecture define "Learning"?
Getting better at a task through experience (Page 3).
What two things does "Learning" require?
A metric that measures improvement (better/worse) and being relative to a specific task (Page 3).
How does the lecture define "Reasoning"?
Logical deduction or probabilistic inference (Page 3).
What is the goal of European AI policy as identified in the "White paper on AI"?
Trustworthy AI (Page 7).
The "White paper on AI" identifies trustworthy AI as necessarily connected with what three things?
Robustness, legality, and ethical standards (Page 7).
What is the "Plus One" in the "Four Principles Plus One" of ethical standards?
Explainability (Page 7).
What is "Robust AI"?
The property of an AI model that its sensitivity/specificity will continue to obtain for new data (Page 8).
What is the difference between "weak robustness" and "strong robustness"?
Weak robustness applies to new data that is similar, while strong robustness applies to new data that is different (Page 8).
What is an "argument"?
A set of propositions containing at least one premise and one conclusion, where premises are intended to support the conclusion (Page 9).
What is "Validity" in a deductive argument?
The property that if the premises are true, then the conclusion is guaranteed to be true (Page 10).
What is a "sound" argument?
An argument that is valid and has all true premises (Page 11).
What is the "Is-Ought Gap" (or "Hume's Law")?
"No ought-judgment may be correctly inferred from a set of premises expressed only in terms of 'is'." (Page 12).
What is the key feature of "Inductive Arguments"?
They are "risky arguments," meaning the premises do not guarantee the conclusion (Page 14).
What is the main aim of "Classical AI"?
To build artificial machines that have intelligence, holding that any feature of human intelligence can be built into a machine (Page 18).
What is the claim of "Strong AI"?
That the AI *actually* has a mind and cognition (Page 18).
What is "Artificial General Intelligence" (AGI)?
The ambition to recreate human or superhuman intelligence using the methodology of Technical AI (Page 19).
What are two main arguments that intelligence is "computation-plus" (i.e., computation alone is not enough)?
The 4E argument (intelligence must be embodied and interact with the environment) and the Chinese room argument (intelligence requires intentionality) (Page 21).
What is the "Chinese room argument" (by John Searle)?
It's an argument that computation alone is not enough for intelligence; it must also have "intentionality" (or understanding) (Page 21).
What is the logical structure of an argument from "necessary conditions"?
1. E is strongly intelligent only if E can learn and reason. 2. AI does not have that capacity. 3. Therefore, AI is not strongly intelligent. (This is *modus tollens*) (Page 23).
What is the logical structure of an argument from "sufficient conditions"?
1. If E has such-and-such a capacity, E is a person. 2. AI has that capacity. 3. Therefore, AI is a person. (This is *modus ponens*) (Page 24).
What are some candidate necessary conditions for "personhood"?
Consciousness, Selfhood, Responsibility, and Goal-setting (Page 25).
What is the moral significance of "personhood"?
Being a person is considered a basic kind of moral status, requiring respect or possessing rights (Page 25).