Overview

Microsoft has invested heavily in Responsible AI, considering the ways that AI models and the data that drives them impacts people and society. One significant issue that has been raised and is being considered across the company is the lack of representation of people with disabilities in the datasets that power our AI-systems (R1). As a result, AI systems have difficulty recognising people with physical anomalies or recognising images taken by people who are blind.

One investment Microsoft has made is the ORBIT project (R2). A collaboration between MSRC and City University (funded by Microsoft AI4Accessibility), it aims to produce a dataset of videos of personal objects taken by people who are blind or low vision. This open-source dataset will be used to drive new methods in the emerging machine learning field of meta-learning which enables personalisation to a specific user when algorithms are given a small number of examples from the user themselves.

As part of this work, we have considered how we might extend our data collection to low and medium resource countries. Initial explorations suggest potential value tensions that we must consider before collecting data with these communities. This PhD will contribute to understanding how best to engage people with disabilities in low-resource settings in participating in data collection for creating assistive AI technologies that will work for them.

Research Questions

It is well established that the data-sets used to build and train AI systems heavily influence the usefulness and usability of the resulting systems; these concerns are amplified when considering systems for special-case user groups (R3). Furthermore, regardless of application, system development that excludes or marginalises direct user engagement will likely lead to failure or suboptimal systems. Recently, the HCI community has argued for humans at the centre of AI development and use (R4). The supervisory team’s recent work also demonstrates how user-based data-set generation with resource-constrained communities might inform and shape AI systems (R5). Furthermore, work – by the team and others – has shown the need to adapt methods to effectively engage with users in low and medium resource countries (R6).

This project will use co-creation methods, probes and prototypes to effectively engage with people who are blind or low vision in resource-constrained contexts to consider appropriate approaches to collecting data-sets that drive machine learning research and assistive systems. The probes and prototype will become a platform for contextual data capture as well as enabling novel insights into interactions with AI assistive technologies.

Supervisory Team

• Prof Matt Jones, Dr Jen Pearson, Dr Simon Robinson (Computational Foundry, Swansea University)
• Dr Cecily Morrison (Microsoft Research, Cambridge)

Exciting Environment

The successful applicant will be based at the Computational Foundry, a £32.5 million world-class facility and a beacon for research collaborations opened in 2018. The Foundry has state-of-the-art labs including machine vision and biometrics; maker labs; interaction-technologies user studies suite; visualisation lab; and IoT lab. The Foundry houses the Computer Science Department (founded over 40 years ago); in the most recent UK REF assessment (2014), the Department was deemed to be leading in many respects with for instance 100% of its “research environment” judged as world-leading or internationally leading.

The PhD Researcher will benefit from existing strong research collaborations with leading labs globally including the Industrial Design Centre in IIT-B (Mumbai) which has expertise in accessibility in the Indian context. Further, they will be affiliated with the EPSRC Centre for Doctoral Training in Enhancing Interactions and Collaborations with Data and Intelligence-Driven Systems.

Eligibility

Candidates from anywhere in the world can apply.

Application Process

The deadline for applications is 26th March 2021, with interviews on 21st and 22nd April 2021.

For an informal conversation please contact Prof. Matt Jones (matt.jones@swansea.ac.uk) or Dr Cecily Morrison (cecilym@microsoft.com).

To apply, please email your application to Swansea-MSFT-App@swansea.ac.uk. Your application must include the following attachments in PDF form:

1. Your CV
2. Degree certificates and transcripts (if you are still an undergraduate, provide a transcript of results known to date)
3. A statement no longer than 2 pages of A4 in length that explains why you want to join our Centre and what you will bring to our community. You should outline the sorts of AI and big data challenges you are interested in and which of our challenging contexts that particularly appeal to you. You do not need to specify a detailed research plan at this time though as this will be shaped during the first year.
4. Completed Scholarship Application Form
5. Academic references – all scholarship applications require two supporting references to be submitted. Please ensure that your chosen referees are aware of the funding deadline, as their references form a vital part of the evaluation process. Please either include these with your scholarship application or ask your referees to send them directly to Swansea-MSFT-App@swansea.ac.uk
6. A completed Equality, Diversion and Inclusion Monitoring Form

References

1. Guo A, Kamar E, Vaughan JW, Wallach H, Morris MR. (2019). Toward Fairness in AI for People with Disabilities: A Research Roadmap.
2. The ORBIT (Object Recognition for Blind Image Training) Dataset
3. Wang L and Wong A. Implications of Computer Vision Driven Assistive Technology Towards Individual with Visual Impairment
4. Shneiderman, B. (2020). Human-Centered Artificial Intelligence: Three Fresh Ideas. AIS Trans on Hum-Comput Inter, 12(3).
5. Reitmaier T, Robinson S, Pearson J, Raju DK, and Jones M. (2020). An Honest Conversation: Transparently Combining Machine and Human Speech Assistance in Public Spaces. CHI '20
6. Pearson J, Robinson S, Reitmaier T, Jones M, and Joshi A. (2019). Diversifying Future-Making Through Iterative Design. ACM Trans. Comput.-Hum. Interact. 26(5)