Italics indicate terms with glossary entries of their own.
Algorithm: A set of instructions for solving a problem or performing a task, which can be as simple as a recipe for meal. In computer science, algorithms are the essential building blocks of artificial intelligence applications. Algorithms “instruct” AI applications on how to interact with training material to achieve the desired outcome, including, for instance, how to collect, process, and analyze data. One everyday example would be how internet search engines use algorithms to rank websites that provide the best answers to user queries.
Algorithm audit: A method of repeatedly varying inputs to observe an algorithm’s outputs, often in order to draw conclusions about its inner workings and/or its potentially problematic outputs. See Conducting a DIY Algorithm Audit for more detail.
Algorithmic accountability reporting: Reporting that adapts journalism’s traditional watchdog function to investigate algorithms, their power, and their potential for producing bias and other social risks. Nicholas Diakopoulos introduced the term in a 2014 article and developed it in greater detail in his 2019 book, Automating the News.
Algorithmic bias: Errors in a computer system that create unfair outcomes, such as privileging one category over another, for example, in terms of race and ethnicity, gender, or sexuality. Algorithmic bias can emerge from pre-existing biases embedded in the data used to train the algorithm, from developers’ selection, cataloging, or coding of that data, or from unintended or unanticipated uses of the algorithm. When biases are built into algorithms, whether intentionally or by mistake, use of algorithms can both perpetuate existing injustices and mask those injustices, due to common assumptions about the objectivity of AI applications
Artificial intelligence (AI): An umbrella term for technologies that use machine learning to make decisions without direct human instruction. AI enables computers and machines to simulate human learning, comprehension, problem solving, decision making, creativity, and autonomy. Artificial intelligence technologies use algorithms to perform these functions, often in remarkably less time than it would take a human to do the same. Many proponents argue that AI technologies reduce the margin of human error in complex tasks and calculations; but critics note their potential for producing algorithmic bias.
Algorithmic literacy: The ability to understand and evaluate algorithmic systems, and the consequences of their use. As one specific type of media literacy, algorithmic literacy emphasizes a critical perspective on the power of algorithms to influence what information reaches people (including journalists and the public) and how they respond to it.
Blocklists: A control mechanism used by advertisers to protect their brands from negative reputation impacts. Blocklists can restrict news publishers’ advertising revenues, and even pressure newsrooms to avoid covering certain types of stories, a form of self-censorship.
Content reduction: See reduction.
Frame, or News Frame: The central idea or primary storyline that organizes the information in a news story and gives meaning to it.
Gatekeeping: The process by which media filter information that reaches the public. First developed by communications researchers in the mid-20th century, the concept has been updated by contemporary scholars to address algorithmic gatekeeping, especially by Big Tech companies and social media platforms, and its impact on the distribution of news stories.
Guardrails: Safeguards to prevent AI technologies from causing harm by ensuring that they operate within ethical and legal boundaries. Without guardrails, AI technology can, for example, perpetuate existing biases, make erroneous or unethical decisions, and introduce privacy concerns. Guardrails include controls embedded in the technology itself and overarching policies and laws.
Keyword blocklists: See blocklists.
Machine learning: A subset of artificial intelligence focused on enabling computers and machines to improve their performance and accuracy through exposure to more data. Machine learning algorithms are frequently used to make classifications or predictions.
Natural language processing: A subfield of artificial intelligence that uses machine learning to allow humans to communicate with artificial intelligence systems using human language. NLP can give human users of AI systems the feeling that they are interacting with another sentient, human being.
News frame: See frame.
Predictive modeling: Use of artificial intelligence to identify patterns that can then be used to predict specific future outcomes, in fields such as health care, criminal sentencing, or customer service. Many critics of AI, including computer scientists Arvind Narayanan and Sayash Kapoor, focus on pitfalls in predictive modeling, which they note often has harmful consequences due to algorithmic bias.
Reduction: An umbrella term developed by Tarleton Gillespie to describe a range of content moderation practices, including shadow banning, used by social media platforms to reduce the visibility or reach of user-produced content deemed problematic, rather than simply removing that content. Gillespie and others argue that reduction on social media platforms can “further scuttle marginalized communities” that have historically been relegated insignificant or invisible by traditional media.
Shadow banning: A technique used by social media platforms to reduce the visibility of posted content without the content creator’s knowledge, and without officially blocking or removing that content. Shadow bans and other forms of content reduction can render newsworthy reporting on controversial topics invisible.
