Lecture 12: Philosophy and Ethics of AI Flashcards
What is the difference between weak and strong AI?
Weak AI: can do individual tasks well such as playing chess, can learn how to improve on that, but it is bounded by the human ability to train it and adjust parameters
Strong AI: can do multiple unrelated tasks well, can teach itself how to solve new tasks, eventually surpasses human intelligence
Does strong AI exist?
No. All the available models are examples of weak AI. Increasing computational resources with current models would not help make current weak AI stronger, but only faster
True or False. AI is good in pattern recognition and surpasses humans in the ability to find relationships in large and complex data
True. But, humans are good at spotting causality, making inferences, and behaving in a new situation based on prior knowledge
What is algorithmic bias?
Algorithmic bias means systematic errors in a program that results in unfair outcomes, e.g., giving privilege to one arbitrary group of users over others
In the context of ML, models inherit social patterns reflected in their training data without any direct effort by programmers to include such biases
Give some examples of bias in data (that is a form of algorithmic bias).
Bias in data highlights the importance of understanding the underlying data before using it, even if a model works well
- old data represents old societal values and outcomes
- unrepresentative samples
- bad labeling of data
- features that are proxy (postal code =! ethnicity)
Explain how there could be bias in the models (also a form of algorithmic bias)
Black box solutions that make decisions without giving humans the ability to understand the decision process are problematic
Algorithmic optimizations without considerations of consequences
e.g., social media giants try to maximize consumption of content to get more revenue (and promote posts that are based on the number of clicks), but this content might be affecting the society (like promoting extremist values and so on)
Give some examples of real-life cases when algorithmic bias actually happened
- Amazon’s recruiting tool
- Healthcare risk algorithm
- Profiling criminals with COMPAS
True or false. You always have to think whether the research that you are conducting is ethical, and even if the algorithms you build or the ideas you have could be used in the future by someone else to use it and cause harm to the society and others.
What do you think? Obviously true :)