Mapping English-Language AI Research Controversies on Twitter, 2022

Marres, Noortje Suzanne (2025). Mapping English-Language AI Research Controversies on Twitter, 2022. [Data Collection]. Colchester, Essex: UK Data Service. 10.5255/UKDA-SN-857742

The 3-year international project “Shaping AI” investigates the development of “Artificial Intelligence” (AI) as a socio-technical phenomenon: AI is not only a noteworthy innovation that is currently being introduced into science and society. It also entails a distinctive approach to solving problems through automation and potentially new ways of making sense of the world and organising the polity.
AI has become the subject of heated debate in recent years, from scientific discussions sparked by the release of new language models to public outrage following the firing of experts by big tech companies and societal concerns about the use of face recognition technologies by the police, and in schools. In this social research project, we map and analyse controversies about AI across different spheres: research, policy and the media. Our objective is to identify what are the most important, possibly overlooked, concerns, disputes and problematics that have arisen in the last 10 years in relation to AI as a strategic area of research, public policy and societal change.

The 6 controversies were selected based on responses submitted to the Shaping AI online expert consultation that took place in Autumn 2021: COMPAS, Facial recognition, "Gaydar," NHS+Deepmind, LLMs ("Stochastic Parrots") and Deep learning as a solution for AI.

The Twitter data were collected by submitting queries - which consisted of terms and URLs of research publications with high relevance to the 6 AI controversies - to Twitter's academic API using TWARC between January 2022 and June 2022. Further details of the method can be found in the methods file with further detail of the queries in the ReadMe.

Data description (abstract)

This submission consists of 12 data sets containing Twitter IDs pertaining to 6 AI controversies identified by UK-based experts in AI and Society as especially significant during the period 2012-2021.
The data sets were collected by researchers at the University of Warwick as part of the 3-year international project “Shaping AI” which mapped controversies about “Artificial Intelligence” (AI) during 2012-2022. Research teams in the UK, France, Germany and Canada analysed controversies about AI in their countries across different spheres: research, policy and the media during this 10-year period.
The UK team at the University of Warwick designed and undertook an analysis of research controversies about AI in the relevant period following a standpoint methodology. Our study began with an online consultation that took place in the Autumn of 2021, in which we asked UK-based experts in AI from across disciplines to identify what are the most important concerns, disputes and problematics that have arisen in the last 10 years in relation to AI as a strategic area of research.

Based on the responses to this expert consultation—described in detail in Marres et al (2024) and Poletti et al (forthcoming)—we identified a broad range of relevant controversy topics, objects and problems. To select controversies for further analysis, we considered their research intensity, in the form of a frequency count of research publications mentioned by respondents in relation to controversy topics.

On this basis, we selected 6 AI research controversies for further research: COMPAS; NHS+Deepmind; Gaydar; Facial recognition; Stochastic Parrots (LLMs) & Deeplearning as a solution for AI. For each of these controversies, we collected Twitter data by submitting queries to Twitter's academic API using TWARC between January 2022 and June 2022. Further details of the methods of data collection and curation can be found in the methods file with further detail of the queries in the ReadMe file.

Data creators:

Creator Name	Affiliation	ORCID (as URL)
Marres Noortje Suzanne	University of Warwick	https://orcid.org/0000-0002-8237-6946

Sponsors:

Economic and Social Research Council

Grant reference:

ES/V013599/1

Topic classification:

Society and culture

Keywords:

ARTIFICIAL INTELLIGENCE, SOCIAL MEDIA, CULTURAL POLICY, HEALTH POLICY, POLICY MAKING, NATIONAL POLICY

Project title:

Shaping 21st Century AI: Controversy and Closure in Research, Policy and Media

Grant holders:

Noortje Marres

Date published:

03 Apr 2025 09:12

Last modified:

03 Apr 2025 17:01

Coverage and Methodology

Temporal coverage:

From	To
11 January 2012	23 June 2022

Geographical area:

United Kingdom, United States, and other countries

Country:

United Kingdom, United States

Spatial unit:

No Spatial Unit

Data collection method:

The submitted Twitter IDs and associated data were collected by submitting queries composed of keywords and URLs of research publications to Twitter's academic API using TWARC.

Keywords and URLs were derived from the online UK expert consultation of the ShapingAI project that asked UK-based AI and Society experts to identify the most important controversies about AI in the period 2012-2022. This online consultation was conducted in the Autumn of 2021 (for more information see: https://doi.org/10.1177/20539517241255103).

Based on the consultation responses, the Shaping AI research team at the University of Warwick identified 5 AI controversies for further investigation. For each of the 5 controversies we selected research publications to serve as queries. These research publications were identified by the consultation respondents as relevant to the controversy in question.

The five controversies and associated publications are:
1. Correctional Offender Management Profiling for Alternative Sanctions (COMPAS): a controversy about algorithmic discrimination in judicial systems sparked by the ProPublica (Angwin et al., 2016) report ‘Machine Bias’.
2. NHS + DeepMind: a controversy about data sharing between UK public sector hospitals and big tech fuelled by the Powles and Hodson (2017) paper on Google DeepMind research in the Royal Free.
3. Facial recognition (Gaydar): a controversy about the use of machine learning-based analysis of portrait images to predict sexual orientation (Wang and Kosinski, 2018).
4. LLMs (Stochastic Parrots): A controversy about bias in large neural network models for encoding and generating text, focusing on Bender et al. (2021) ‘On the Dangers of Stochastic Parrots’.
5. Deep learning (DL) as a solution for AI: a controversy about the capacity of DL—the use of trained multilayer neural networks with large numbers of weight parameters—to sustain the claims of artificial intelligence research (Marcus, 2018).

Twitter queries were designed to detect expert debate about the above controversy topics on Twitter, with "expert" being broadly defined as including all ‘those with a stake in an issue who are committed to genuine debate’ (Funtowicz and Ravetz, 1991: 170).

Observation unit:

Time unit

Kind of data:

Text

Type of data:

Time series data

Resource language:

English

Access and Administration

Available Files

Downloads

data downloads and page views since this item was published

View more statistics

Altmetric

Related Resources

Website

Shaping 21st Century AI: Controversies and Closure in Media, Policy, and Research

Edit item (login required)

Edit Item