The Alignment Problem: Machine Learning and Human Values

Reinforcement Learning is a “type of machine learning technique that enables an agent to learn in an interactive environment by trial and error using feedback from its own actions and experiences” (Bhatt, Shweta. “Reinforcement Learning 101.” Medium, 19 Mar. 2018). Unlike supervised learning where the model is trained with the correct answer, in reinforcement learning, the agent learns from the consequences of its actions through rewards or penalties. The concept of reinforcement learning was developed by B.F. Skinner, whose wartime experiments on pigeons used external rewards to “sculpt” the pigeons’ behavior.

Word embedding is a technique in language processing where words are mapped to vectors of numbers. This process captures the semantic relationships between words. Common models used to generate word embeddings include Google’s word2vec and Stanford’s GloVe, which use large datasets of text to learn these representations.

Neural network systems are computational models inspired by the human brain’s architecture. They consist of interconnected “neurons” that process information in a dynamic way. They also recognize patterns, respond to external stimuli, and solve problems in a similar way to humans. They are used in applications ranging from voice recognition and image processing to complex decision-making tasks.