Investigating public discourses around decision-making algorithms using a combined computational and discursive language analysis approach
Lead Research Organisation:
University of Nottingham
Department Name: School of Computer Science
Abstract
This project examines the scope for investigating the discourse surrounding public-facing decision-making algorithms using an interdisciplinary lens. Public-facing decision-making algorithms aim to increase productivity and enable more efficient and informed decision-making. Examples include the UK's Covid-19 digital contact-tracing application and the Ofqual A Level automated algorithm used in 2020. These work without supervision and had impact on United Kingdom citizens. Investigating trust in Autonomous Systems is critical for the development of future artificial intelligence technologies as they become more relevant to our daily lives. The Trustworthy Autonomous Systems Hub adopts Devitt's view that Autonomous Systems must be trustworthy by design and perception.
The collection of opinions here will act as validation or signpost the need for alternative exploration. To analyse the views expressed, computational linguistic methods - including topic modelling, sentiment analysis and emotion detection - may be deployed. These can be non-intrusive and cost-effective, as opposed to interviews or experiments. A source of data may be social media, notably Twitter, as many who have been affected by these algorithms offer opinions on this public-access site, providing a large data set that Twitter's API can analyse in real-time.
However, these methods show limited awareness of the discursive and conversational ways in which opinions on decision-making algorithms are discussed on social media, which should be accounted for to understand this in further detail. Through initial work, popular computational linguistic methods' shortcomings include issues with the classification of negation, sarcasm and irony and difficulty in interpretation. A new approach may overcome these shortcomings by combining computational linguistic and sociolinguistic analytical methods, such as corpus linguistics and discourse analysis, where context plays an important role. Creating a hybrid approach to analysing the discourse surrounding public-facing decision-making algorithms may produce better quality insights to break down barriers to trust and adoption.
This project investigates the following questions:
- How can current forms of computational linguistic analysis support the understanding of social media public discourses around decision-making algorithms? What are the benefits and shortcomings?
- How can the discourse surrounding public-facing decision-making algorithms on social media be understood in more detail through the inclusion of discourse analysis and corpus linguistics alongside computational linguistic analysis?
- In what ways can existing computational linguistic analysis be combined with qualitative linguistic analysis to mitigate the shortcomings of existing methods?
The collection of opinions here will act as validation or signpost the need for alternative exploration. To analyse the views expressed, computational linguistic methods - including topic modelling, sentiment analysis and emotion detection - may be deployed. These can be non-intrusive and cost-effective, as opposed to interviews or experiments. A source of data may be social media, notably Twitter, as many who have been affected by these algorithms offer opinions on this public-access site, providing a large data set that Twitter's API can analyse in real-time.
However, these methods show limited awareness of the discursive and conversational ways in which opinions on decision-making algorithms are discussed on social media, which should be accounted for to understand this in further detail. Through initial work, popular computational linguistic methods' shortcomings include issues with the classification of negation, sarcasm and irony and difficulty in interpretation. A new approach may overcome these shortcomings by combining computational linguistic and sociolinguistic analytical methods, such as corpus linguistics and discourse analysis, where context plays an important role. Creating a hybrid approach to analysing the discourse surrounding public-facing decision-making algorithms may produce better quality insights to break down barriers to trust and adoption.
This project investigates the following questions:
- How can current forms of computational linguistic analysis support the understanding of social media public discourses around decision-making algorithms? What are the benefits and shortcomings?
- How can the discourse surrounding public-facing decision-making algorithms on social media be understood in more detail through the inclusion of discourse analysis and corpus linguistics alongside computational linguistic analysis?
- In what ways can existing computational linguistic analysis be combined with qualitative linguistic analysis to mitigate the shortcomings of existing methods?
Planned Impact
We will collaborate with over 40 partners drawn from across FMCG and Food; Creative Industries; Health and Wellbeing; Smart Mobility; Finance; Enabling technologies; and Policy, Law and Society. These will benefit from engagement with our CDT through the following established mechanisms:
- Training multi-disciplinary leaders. Our partners will benefit from being able to recruit highly skilled individuals who are able to work across technologies, methods and sectors and in multi-disciplinary teams. We will deliver at least 65 skilled PhD graduates into the Digital Economy.
- Internships. Each Horizon student undertakes at least one industry internship or exchange at an external partner. These internships have a benefit to the student in developing their appreciation of the relevance of their PhD to the external societal and industrial context, and have a benefit to the external partner through engagement with our students and their multidisciplinary skill sets combined with an ability to help innovate new ideas and approaches with minimal long-term risk. Internships are a compulsory part of our programme, taking place in the summer of the first year. We will deliver at least 65 internships with partners.
- Industry-led challenge projects. Each student participates in an industry-led group project in their second year. Our partners benefit from being able to commission focused research projects to help them answer a challenge that they could not normally fund from their core resources. We will deliver at least 15 such projects (3 a year) throughout the lifetime of the CDT.
- Industry-relevant PhD projects. Each student delivers a PhD thesis project in collaboration with at least one external partner who benefits from being able to engage in longer-term and deeper research that they would not normally be able to undertake, especially for those who do not have their own dedicated R&D labs. We will deliver at least 65 such PhDs over the lifetime of this CDT renewal.
- Public engagement. All students receive training in public engagement and learn to communicate their findings through press releases, media coverage.
This proposal introduces two new impact channels in order to further the impact of our students' work and help widen our network of partners.
- The Horizon Impact Fund. Final year students can apply for support to undertake short impact projects. This benefits industry partners, public and third sector partners, academic partners and the wider public benefit from targeted activities that deepen the impact of individual students' PhD work. This will support activities such as developing plans for spin-outs and commercialization; establishing an IP position; preparing and documenting open-source software or datasets; and developing tourable public experiences.
- ORBIT as an impact partner for RRI. Students will embed findings and methods for Responsible Research Innovation into the national training programme that is delivered by ORBIT, the Observatory for Responsible Research and Innovation in ICT (www.orbit-rri.org). Through our direct partnership with ORBIT all Horizon CDT students will be encouraged to write up their experience of RRI as contributions to ORBIT so as to ensure that their PhD research will not only gain visibility but also inform future RRI training and education. PhD projects that are predominantly in the area of RRI are expected to contribute to new training modules, online tools or other ORBIT services.
- Training multi-disciplinary leaders. Our partners will benefit from being able to recruit highly skilled individuals who are able to work across technologies, methods and sectors and in multi-disciplinary teams. We will deliver at least 65 skilled PhD graduates into the Digital Economy.
- Internships. Each Horizon student undertakes at least one industry internship or exchange at an external partner. These internships have a benefit to the student in developing their appreciation of the relevance of their PhD to the external societal and industrial context, and have a benefit to the external partner through engagement with our students and their multidisciplinary skill sets combined with an ability to help innovate new ideas and approaches with minimal long-term risk. Internships are a compulsory part of our programme, taking place in the summer of the first year. We will deliver at least 65 internships with partners.
- Industry-led challenge projects. Each student participates in an industry-led group project in their second year. Our partners benefit from being able to commission focused research projects to help them answer a challenge that they could not normally fund from their core resources. We will deliver at least 15 such projects (3 a year) throughout the lifetime of the CDT.
- Industry-relevant PhD projects. Each student delivers a PhD thesis project in collaboration with at least one external partner who benefits from being able to engage in longer-term and deeper research that they would not normally be able to undertake, especially for those who do not have their own dedicated R&D labs. We will deliver at least 65 such PhDs over the lifetime of this CDT renewal.
- Public engagement. All students receive training in public engagement and learn to communicate their findings through press releases, media coverage.
This proposal introduces two new impact channels in order to further the impact of our students' work and help widen our network of partners.
- The Horizon Impact Fund. Final year students can apply for support to undertake short impact projects. This benefits industry partners, public and third sector partners, academic partners and the wider public benefit from targeted activities that deepen the impact of individual students' PhD work. This will support activities such as developing plans for spin-outs and commercialization; establishing an IP position; preparing and documenting open-source software or datasets; and developing tourable public experiences.
- ORBIT as an impact partner for RRI. Students will embed findings and methods for Responsible Research Innovation into the national training programme that is delivered by ORBIT, the Observatory for Responsible Research and Innovation in ICT (www.orbit-rri.org). Through our direct partnership with ORBIT all Horizon CDT students will be encouraged to write up their experience of RRI as contributions to ORBIT so as to ensure that their PhD research will not only gain visibility but also inform future RRI training and education. PhD projects that are predominantly in the area of RRI are expected to contribute to new training modules, online tools or other ORBIT services.
Publications
Heaton D
(2023)
Critical reflections on three popular computational linguistic approaches to examine Twitter discourses.
in PeerJ. Computer science
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| EP/S023305/1 | 01/10/2019 | 31/03/2028 | |||
| 2445668 | Studentship | EP/S023305/1 | 01/10/2020 | 30/09/2024 | Daniel Heaton |
| Description | I have published a paper related to the PhD in PeerJ Computer Science. Although computational linguistic methods -- such as topic modelling, sentiment analysis and emotion detection -- can provide social media researchers with insights into online public discourses, it is not inherent as to how these methods should be used, with a lack of transparent instructions on how to apply them in a critical way. There is a growing body of work focusing on the strengths and shortcomings of these methods. Through applying best practices for using these methods within the literature, we focus on setting expectations, presenting trajectories, examining with context and critically reflecting on the diachronic Twitter discourse of two case studies: the longitudinal discourse of the NHS Covid-19 digital contact-tracing app and the snapshot discourse of the Ofqual A Level grade calculation algorithm, both related to the UK. We identified difficulties in interpretation and potential application in all three of the approaches. Other shortcomings, such the detection of negation and sarcasm, were also found. We discuss the need for further transparency of these methods for diachronic social media researchers, including the potential for combining these approaches with qualitative ones -- such as corpus linguistics and critical discourse analysis -- in a more formal framework. I have submitted a paper (currently in review) related to the PhD to SAGE Open. In August 2020, the UK government and Ofqual replaced A Level examinations with automatically-computed grades. After public outcry, teacher assessments were used instead. Views concerning who to blame were expressed on Twitter. While research of this scale usually requires NLP-based computational linguistics, shortcomings include difficulties in interpretation and limited conclusions concerning blame. Thus, we complement this by analysing 18,239 tweets using Corpus Linguistics (CL) and Critical Discourse Analysis (CDA), underpinned by Social Actor Representation (SAR). Through analysing transitivity and blame attribution, we found the algorithm, the UK government and Ofqual were all implicated through active agency, agency metaphor and possession. Accordingly, students were found to have limited blame. We discuss how this builds upon existing research where the algorithm is implicated and how constructions obscure blame. Methodologically, we demonstrate that CL and CDA complement sentiment analysis; however, there is further scope to use these approaches in an iterative manner. I have submitted a paper (currently in review) related to the award to The Information Society. Since September 2020, the NHS Covid-19 App has been used to mitigate the spread of Covid-19 in the UK. Since its launch, this app has been part of the research discussion regarding the perceived social agency of decision-making algorithms. On the social media website Twitter, for instance, a plethora of views about the app have been found. To investigate how the app was portrayed publicly, we took advantage of the strong link between grammatical and social agency and analysed a corpus of 118,316 tweets from September 2020 to July 2021 to see whether the app featured as a social actor. Using Corpus Linguistics and Critical Discourse Analysis, underpinned by Social Actor Representation, we found that active presentations of the app - seen mainly through personalisation and agency metaphor - dominated the discourse and grew in proportion to passive presentations as time moved on. These active presentations showed the app to be a social actor in five main ways: informing, instructing, providing permission, disrupting and functioning. We also found occasions where the app was presented passively, through backgrounding and exclusion. In this paper, we discuss how this builds upon existing research about the app specifically and agency and decision-making algorithms more generally. I have submitted a paper (currently in review) related to the award to TAS '23. Despite the ever-growing use of decision-making algorithms in daily life, there has been limited examination of how the supposed agency, responsibility and accountability of these algorithms can have impact on whether users trust them or blame them for failures. Therefore, this contribution reviews current literature relating to these concepts in order to synthesise present ideas around the social impact of these systems. We highlight the challenges of defining and operationalising these concepts in the context of algorithmic governance and discuss the need for more empirical research on how decision-making algorithms impact trust and blame in practice. We also foreground the importance of presumed agency and whether human agency is mitigated by increased algorithmic agency. After this, we use the AREA 4P responsible research and innovation framework to reflect on the findings of the literature review, which emphasises the need for a more nuanced understanding of the impact of agency, responsibility and accountability on trust and blame in algorithmic decision-making. By addressing these concerns and gaps in research, the authors argue that scholars can develop more effective strategies for ensuring responsible and ethical governance of decision-making algorithms. |
| Exploitation Route | These outcomes will influence how developers and promoters of decision-making algorithms overcome barriers to adoption. They will also influence how computational and discursive linguistic methodological approaches can be used together for more valuable insights into social media analytics. |
| Sectors | Digital/Communication/Information Technologies (including Software),Education |