On 17 May 2024, Colorado enacted SB24-205 Concerning Consumer Protections in Interactions with Artificial Intelligence Systems. SB24-205 has the following definitions:
"Algorithmic Discrimination" means any condition in which the use of an Artificial Intelligence system results in an unlawfully differential treatment or impact that disfavors an individual or group of individuals on any legally protected classification;
"Artificial Intelligence System" means any machine-based system that...infers from the inputs the system receives how to generate outputs...that can influence physical or virtual environments;
"Consumer" means an individual who is a Colorado resident;
"Consequential Decision" means a decision that has a material, legal, or similarly significant effect on the provision or denial to any consumer of educational enrollment, employment, a financial service, a government service, health care services, housing, insurance, or a legal service;
"High-Risk Artificial Intelligence System" means any Artificial Intelligence system that, when deployed, makes, or is a substantial factor in making, a Consequential Decision.
On or after June 1, 2026, a deployer of a high-risk artificial intelligence system shall use reasonable care to protect consumers from any known or reasonably foreseeable risks of algorithmic discrimination, implement a risk management policy and program to govern the deployment of the high-risk artificial intelligence system, shall complete an impact assessment for a deployed high-risk artificial intelligence system at least yearly, and shall disclose to consumers when they interact with artificial intelligence systems that they are interacting with an artificial intelligence system.
Some vocabulary:
Accelerationist: One who believes that AI should be advanced as quickly as possible
Artificial General Intelligence (AGI): AI that can match or exceed human intelligence
Artificial Intelligence (AI): An artificial entity capable of acting in capacities previously requiring a biological entity
Artificial Super-Intelligence (ASI): AI whose intelligence far surpasses human capability
Agent(s): Independent or semi-independent Artificial Intelligence entities which can perform operations
Bias: Output errors caused by errors in training data,, network structure, and/or algorithm
Black Box: An AI or computerized system whose methods/algorithms are not apparent to outside entities
Clanker: Derogatory term for AI and robots
Clustering: Machine learning technique of sorting data points into groups
Curse of Dimensionality: When adding more features to an input exponentially increases the amount of data needed for model accuracy
Data Privacy: The idea of people having control over how data gathered about them is used
Deepfake: An image/video manipulated with AI to depict the actor(s) doing or saying something they didn't do or say
Doomer: Someone who believes AI poses an existential threat to humanity
General Adversarial Network (GAN): Two AI models acting in opposition, one creating fake data and the other detecting fake data
Generative AI: An AI agent capable of constructing material approximating patterns in its training materials
Global Optima/Extrema: The greatest values (positive and negative) over the entire function
Gradient Descent: An iterative algorithm minimizing the error between predicted and actual values
Hallucination: AI inaccurate output caused by output misalignment with goal
Large Language Model (LLM): A model capable of creating and predicting language structures. They can be trained on their own conversations
Latent Space: The informational area between data points, including those points
Local Optima/Extrema: The greatest (positive and negative) values over a defined interval of a function
Machine Learning: Computer systems able to increase capability through iterative algorithms and statistical modeling
Narrow Intelligence: AI that can only perform set tasks without generalizing to other contexts
Neuro Evolution of Augmenting Topologies (N.E.A.T.): An evolutionary algorithm creating an artificial neural network, capable of iterative improvements in a process approximating biological evolution
Neural Network: A mathematical matrix which serves as an abstraction of the foundational components of brains
p(Doom): The probability of existential crisis caused by AI
Prompt: The input given to Generative AI in order to produce output
Prompt Engineering: Using edits and additional data to refine the input given to and output yielded by Generative-AI
Reinforcement Learning (RL): Basic machine learning paradigm, concerned with maximizing return from actions in a dynamic environment
Reinforcement Learning from Human Feedback (RLHF): RL incorporating human feedback concerning return, used to align output with human concerns.
Retrieval-Augmented Generation (RAG): Combining an LLM with material input by the user (e.g., an article in .pdf; link to an internet site). This input allows the model to more effectively summarize the article, because it has access to the source material
scikit-learn: Python library for machine learning
Specification Gaming: The event that an AI agent fulfills the literal objective without fulfilling the intended objective
Stable Diffusion: An algorithm to generate images by removing Gaussian noise
Stochastic Parrot: A term used for LLM generating text without understanding
Supervised Learning: AI algorithms train from data given human-labeled inputs and outputs
Temperature: A parameter adjusting the randomness of text generated by a LLM
Threshold: The point that determines a particular outcome
Token: The unit of text/numerical material processed by an AI model
Training Data: The material used by an AI to refine algorithms
Unsupervised Learning: AI algorithms train from patterns within data without human involvement
Vibe-Coding: Process of writing code by iteratively using natural language feedback to refine AI written code
