Artificial intelligence (AI): Computer programs that can complete cognitive tasks typically associated with human intelligence
AI augmentation: The process of using AI to improve a work product, whether by making it easier to do or higher in quality, but the human still maintains control and makes the final decisions.
AI automation: The process of using AI to accomplish tasks, without any action on the user’s part
AI model: A computer program trained on a set of data to recognize patterns and perform specific tasks
AI tool: AI-powered software that can automate or assist users with a variety of tasks
AI user: Someone who leverages AI to complete a personal or professional task
Allocative harm: Wrongdoing that occurs when an AI system’s use or behavior withholds opportunities, resources, or information in domains that affect a person’s well-being
Biased data: Data that is incomplete, does not accurately represent populations, or includes preferential treatment for certain individuals or groups
Chain-of-thought prompting: A prompting technique that involves requesting a large language model to explain its reasoning processes
Cognitive task: Any mental activity, such as thinking, understanding, learning, and remembering
Conversational AI tool: A generative AI tool that processes text requests and generates text responses
Data bias: A circumstance in which systemic errors or prejudices lead to unfair or inaccurate information, resulting in biased outputs
Deepfakes: AI-generated fake photos or videos of real people saying or doing things that they did not do
Drift: The decline in an AI model's accuracy in predictions due to changes over time that are not reflected in the training data
Few-shot prompting: A technique that provides two or more examples in a prompt
Generative AI: AI that can generate new content, like text, images, or other media
Hallucinations: AI outputs that are not true
Human-in-the-loop approach: A combination of machine and human intelligence to train, use, verify, and refine AI models
Interpersonal harm: The use of technology to create a disadvantage to certain people that negatively affects their relationships with others or causes a loss of one's sense of self and agency
Knowledge cutoff: The concept that an AI model is trained at a specific point in time, so it doesn’t have any knowledge of events or information after that date
Large language model (LLM): An AI model that is trained on large amounts of text to identify patterns between words, concepts, and phrases so that it can generate responses to prompts
Machine learning (ML): A subset of AI focused on developing computer programs that can analyze data to make decisions or predictions
Med-PaLM: A large language model designed to answer medical questions, specialized through training with clinicians
Med-PaLM M: A multimodal version of Med-PaLM that can interpret several types of biomedical data, including language, medical imaging, and genomic data
Multimodal model: An AI model that can accept and learn from multiple types of input, such as images, video, or audio
Natural language: The way people talk or write when communicating with each other
One-shot prompting: A technique that provides a single example in a prompt
Open dataset: A dataset that is freely available to anyone to use
Privacy: The right for a user to have control over how their personal information and data are collected, stored, and used
Prompt: Text input that provides instructions to the AI model on how to generate output
Prompt engineering: The practice of developing effective prompts that elicit useful output from generative AI
Quality-of-service harm: A circumstance in which AI tools do not perform as well for certain groups of people based on their identity
Representational harm: An AI tool’s reinforcement of the subordination of social groups based on their identities
Responsible AI: The principle of developing and using AI ethically, with the intent of benefiting people and society while avoiding harm
Security: The act of safeguarding personal information and private data, and ensuring that the system is secure by preventing unauthorized access
Social system harm: Macro-level societal effects that amplify existing class, power, or privilege disparities, or cause physical harm, as a result of the development or use of AI tools
Systemic bias: A tendency upheld by institutions that favors or disadvantages certain outcomes or groups
Training set: A collection of data used to teach AI
Transparency: The idea that an AI tool should provide insight into how it works, why it made a particular output, and what factors contributed to that output