1: Ethics and Machine Learning

What does it mean to evaluate an action morally (or ethically)? What is an example of a sentence that expresses a moral evaluation of an action?

Assessing whether an action is right, wrong, or a matter of moral duty.

“Telling the harsh truth was the (morally) right thing to do.”

What type of questions does the study of ethics seek to answer, and what kind of statements answer those questions?

Ethics seeks answers to evaluative questions, such as whether a particular action or AI feature is good, wrong, or bad.

Evaluative statements serve as the answers to these questions.

What do all evaluative statements have in common?

They all involve a positive or negative assessment of something.

How is ‘intelligence’ defined in lecture? How is ‘artificial intelligence’ defined?

Intelligence is defined as what people use when acquiring and processing information, recognizing objects and patterns, reasoning, strategizing, and solving problems.

Artificial intelligence is defined as what a machine uses when acquiring and processing information, recognizing patterns, reasoning, strategizing, and solving problems.

What is symbolic AI (sometimes called “Good Old-Fashioned AI” or “GOFAI?”)?

Symbolic AI (or "Good Old-Fashioned AI") is an approach that uses a sequence of explicitly defined statements and instructions to manipulate symbols and derive results that are logically entailed by its pre-programmed rules.

Why are symbolic AI systems described as static ?

Symbolic AI systems are described as static because, given the same inputs, they will always deliver the same outputs based on their fixed rules.

What is an illustration of symbolic AI, and why is it an illustration of symbolic AI?

The home loan decision tree and Deep Blue illustrate symbolic AI because both rely strictly on pre-programmed, step-by-step decision rules and symbol manipulations to evaluate inputs. Neither system updates or discovers its own underlying rules over time.

What is machine learning, and in what sense is a machine learning system dynamic rather than static?

Machine learning refers to an algorithmic system that is dynamic because it is designed to improve "by itself" over time. Unlike static systems, it constructs a model from data, tests it, and iteratively revises its model to account for errors without relying on fixed instructions.

What are the four morally significant features of machine learning?

bias, scale, opacity, and autonomy.

What is redundant encoding? What is an example?

Redundant encoding (or proxies) occurs when a predictive data feature has a strong correlation with membership in a protected group.

An example is an automated recruitment system using zip codes or specific colleges as data points, which act as proxies that replicate demographic bias.

What are the four core ethical questions about bias and machine learning?

1. Whether algorithmic bias makes predictive systems unfair or unusable.

2. Whether large-scale data collection violates citizens' right to privacy.

3. How to justify decisions made by opaque "black box" models.

4. How to assign moral responsibility to autonomous AI systems.