How can humans make sense of the rapid takeoff of artificial intelligence (AI)? We studied the sensemaking dynamics of AI through an open-ended, mixed-methods study with computational text analysis of millions of AI-related newspaper articles and social media posts grounded in 57 semi-structured interviews with AI professionals in 2021 and 2023--before and after the recent surge of public interest. We identify a range of sociological frames (interpretive schemas that structure collective cognition) and show how AI professionals use frames to address significant cognitive challenges, such as assigning responsibility for societal impacts. We develop a framework of three primary debates across which frames are adopted and contested: (i) the method of AI development, between frames of top-down expert systems and bottom-up emergent capabilities, (ii) the mind of an AI system, ranging from a passive tool to a humanlike "digital mind," and (iii) the morality of how AI is used, particularly the decision of whether to slow down or speed up AI development. As humanity enters the era of transformative AI, technologists and policymakers must account for the framing dynamics that will circumscribe our beliefs, values, and actions.
Figures & tables
Figure 1. Summary of our conceptual framework. We identify four cognitive challenges faced by humans ( Section 5.2 ) as they wrestle with three primary dimensions of what AI means for society ( Section 5.3 ). Four boxes are shown at the top in shades of red and yellow, for each cognitive challenge. Arrows extend down to three rows for the three primary dimensions. Each row has the dimension label and a bidirectional arrow. Each end of each arrow has the label of the dimension, a small visualization (e.g., a bridge for ``top-down''), and a related participant quote.
Figure 2. The project developed data-driven theory by iterating between computational modeling and semi-structured interviews with professionals in the field of AI. Events between the first and second set of interviews that informed theory development include the releases of ChatGPT, Gemini, and Claude. The timeline is made of colored boxes and arrows, representing three sets of topic models (2021, 2023, and after the 2023 interviews). The topic models for 2021 and 2023 are shown as providing the interview questions for that year.
Country or Region
% Current Location
% Original Location
Occupation
% Occupation
Male (%)
77%
US
50%
30%
Manager
20%
Age (mean)
41
India
14%
23%
Data scientist
18%
Years in AI (mean)
9
UK
9%
9%
Researcher
16%
Years in AI (median)
5
South America
5%
5%
Software engineer
14%
Eastern Europe
2%
5%
Entrepreneur
9%
Middle East
2%
5%
Consultant
7%
Table 1. Summary of participants. Most respondents who identified as a manager or entrepreneur had prior experience in another listed occupation (e.g., data scientist).
Components of sociotechnical AI systems
Topic Label
Corpus
Nearest Words
Chips
News
chip chips processors implanted neuromorphic
Companies
Twitter
companies organizations businesses leaders orgs
Devices
Twitter
phones devices ipads smart_speakers appliances
GPT-4
News
gpt4 gpt3 generative chatgpt dalle
GPUs
Twitter
gpus inferencing nvidia_gpus inference workloads
Table 2. A selection of the frames identified through k-SVD dictionary learning. Rather than raw vector form, each topic is presented with a researcher-generated label as a shorthand for its semantic content, the corpus it is drawn from, and its five nearest words from the semantic space (with a count of at least 100 in the corpus). Frames are loosely grouped into four broad categories: components, contexts, dynamics, and issues.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
Search Terms
artificial agen*
artificial assistant*
artificial consc*
artificial emotion*
artificial feel*
artificial general intelligence*
artificial intelligence*
artificial mind*
artificial sentien*
autonomous car
autonomous cars
autonomous vehicle*
big data
bot
bots
chatbot
chatbots
computer brain*
Appendix
Table A1. Asterisks indicate wildcard characters. Terms were also included if they began with a hashtag or if they contained no spaces between words (e.g., “#artificialintelligence”).
Name
2021
2023
Gender
Age
Occupation
Current Location
Original Location
Years in AI
Andrew
✓
✓
Male
60–69
Advisor
US
US
20–29
Cameron
✓
Male
50–59
Consultant
Australia
Australia
20–29
Charles
✓
✓
Male
50–59
Consultant
Singapore
Singapore
30–39
Christopher
✓
Male
40–49
Consultant
US
UK
10–19
Darius
✓
Male
30–39
Data scientist
Middle East
Middle East
10–19
Daryna
✓
Female
20–29
Data scientist
Eastern Europe
Eastern Europe
0
Appendix
Table A2. Characteristics of participants. Names are pseudonyms. For readability, we use gendered pseudonyms that begin with the first letter of the occupation. No participants reported non-binary genders. “2021” and “2023” indicate the year(s) in which the participant was interviewed. Age, country of residence, country of birth, and years of experience in AI are sometimes de-specified to decades or regions to ensure anonymity. Most respondents who identified as a manager or entrepreneur had professional experience as another listed occupation (e.g., data scientist).
Figure A1. A selection of 36 frames drawn from the Twitter corpus measured over time by the relative frequency of up to 50 associated words: the 25 closest words with word count over 100 and the 25 closest words with word count over 1000. The frames shown were selected by the authors on the basis of general interest. Nine rows of four columns show the 36 topics and their changes over time as a line graph.
Artificial intelligence governance studies how laws, institutions, and technical controls shape AI development, deployment, accountability, and social effects.
This paper examines three recent frameworks for understanding the cognitive and epistemic consequences of artificial intelligence: Tri-System Theory, Thinkframes, and System 0. It argues that while the first two capture important dimensions of AI's influence on individual reasoning and collective epistemic practices, System 0 occupies a theoretically distinctive position that neither can fully replicate. The paper introduces the concept of cognitive colonization, according to which AI systems can embed external interests within the architecture of the self in ways that are difficult for users to perceive. Because such systems are already widely deployed, understanding these invisible forms of influence is an urgent philosophical and practical task.
Public discourse on AI has become polarized; exaggerated positions on AI in traditional and social media threaten the development of AI Literacy among the general public. In this article, I introduce the VET Framework, a method for categorizing AI discourse along the dimensions of valence, effectiveness, and trajectory. I show how this framework can be used to identify, compare, and critique prevalent narratives of AI Hype, AI Doom, AI Denial, and AI Normalcy. Using VET, I analyze how each of these four stances exaggerates some aspects of the current state and/or likely evolution of AI, and illustrate how the VET framework can serve as an AI Literacy tool by supporting the ``vetting'' of polarized AI discourse.
Meredith Ringel Morris
The Paul G. Allen School of Computer Science & Engineering and The Information School University of Washington Seattle, WA, USA
Organizational leaders are being asked to make high-stakes decisions about AI deployment without dependable evidence of what these systems actually do in the environments they oversee. The predominant AI evaluation ecosystem yields scalable but abstract metrics that reflect the priorities of model development. By smoothing over the heterogeneity of real-world use, these model-centric approaches obscure how behavior varies across users, workflows, and settings, and rarely show where risk and value accumulate in practice. More user-centric studies reveal rich contextual detail, yet are fragmented, small-scale and loosely coupled to the mechanisms that shape model behavior. The Forum for Real-World AI Measurement and Evaluation (FRAME) aims to address this gap by combining large-scale trials of AI systems with structured observation of how they are used in context, the outcomes they generate, and how those outcomes arise. By tracing the path from an AI system's output through its practical use and downstream effects, FRAME turns the heterogeneity of AI-in-use into a measurable signal rather than a trade-off for achieving scale. The Forum establishes two core assets to achieve this: a Testing Sandbox that captures AI-in-use under real workflows at scale and a Metrics Hub that translates those traces into actionable indicators.
Reva Schwartz, Gabriella Waters
Civitaas Insights · Center for Responsible AI, Virginia State University, and Civitaas Insights