Abstract
Poverty in the US is not invisible. A large number of Americans are low-income and experience homelessness. This population relies on scarce-resourced public services for survival and thriving. High-stake public service resource allocations are increasingly fueled by AI/ML to provide efficient and scalable services. While AI/ML tools are deployed with positive expectations, there is growing evidence of AI/ML causing invisible harm, often dismissed as inevitable. AI/ML tools are based on deficits and vulnerability ranking; they ignore the strengths of vulnerable individuals. This thesis aims to counteract existing exclusions and reduce AI/ML access barriers for low-income individuals, through their inclusion in AI/ML innovations and education, using mixed-methods experimental studies. Specifically, (1) designing and evaluating value-sensitive service-assessment tools that go beyond individual risk-factors and focus on strengths, (2) making AI/ML knowledge digestible for low-income individuals. Thereby, bringing about social change, AI/ML access, and algorithmic reform.
1 BACKGROUND AND MOTIVATION
In the US, poverty is all too pervasive and persistent. 1 In 2022, more than 50 million Americans were classified as low-income and had an annual income of less than 125% of poverty. 2 Similarly, a large number of people experience homelessness every day; for example, 582,462 Americans are experiencing homelessness in 2023. 3 Low-income individuals face barriers to supporting their daily needs, for example, access to healthcare, insurance, transportation, childcare support, access to education and stable income or job opportunities [1, 5, 15, 25, 49]. The survival of low-income people depends on public service, including welfare resources [17].
Artificial intelligence (AI), also known sometimes as powerful machine learning algorithms (ML), is now fueling public welfare services that affect poor and low-income individuals [8, 10, 15, 22, 43, 48]. For example, to offer efficient, low-cost, and scalable solutions to the large homeless population, there is a growing demand to automate public services by deploying AI/ML tools (e.g., VI-SPDAT assessment and ranking algorithms 4) [27, 40, 44]. While housing prediction and classification tools support the staff members, they are sometimes detrimental to homeless people due to inaccurate, error-prone, or biased predictions. These tools are data-hungry and predict future behavior based on historical data, causing discrimination (e.g., replicating historical and pre-existing racial biases [20, 35]). AI/ML predictions divert the poor from deserved resources; they classify, criminalize, and punish the poor; they compromise their humanity, values, and self-determination abilities [17, 28].
Current public housing service assessment tools rely on vulnerability scores for outcome prediction and completely ignore the strengths and assets of vulnerable individuals [5, 21, 26, 53]. This deficit lens increases stigma and marginalization [54]. Most importantly, these people are not even aware that algorithms are being used to predict eligibility or service deservingness. Lack of transparency increases uncertainty, causing dissatisfaction and mistrust in public welfare services [14, 17, 35, 38, 50]. Homeless people are also almost always excluded from AI/ML technological innovations while directly impacted by them. Research shows that novelty and performance values are uplifted in ML homelessness service provision work [44]. Consequently, there is little or no consideration of human values in high-stake contexts (e.g., privacy and autonomy [19]). Building on these, the thesis will go beyond individual risk factors, focus on strengths, and sustainable assets, and center on ethical values for algorithmic tool design in the homelessness context.
The rapid proliferation of AI/ML caused an increasing effort toward making AI/ML knowledge more accessible for the general public [30, 34, 52]. However, the poor and the impacted people are mostly ignored in all these undertakings, despite being directly impacted by AI/ML. I argue that impacted low-income individuals should also have fair AI/ML learning opportunities, ensuring they are not left behind in the modern AI era. Thus, the second part of this thesis aims to support the AI/ML literacy of low-income individuals, enabling their meaningful participation in AI/ML discourse.
This dissertation will address the following Research Questions (RQs): RQ1: What specific values are uplifted in ML or algorithmic approaches in homelessness service provision research? RQ2: What specific assets are discovered in homelessness research? RQ3: What are different stakeholders’ perceptions of value-sensitive algorithmic tools in the homelessness context? RQ4: How might we support low-income individuals’ knowledge about algorithmic impacts and help them learn about supervised machine learning algorithms? The first study (RQ1) was published in the ACM Conference on Human Factors in Computing Systems (CHI) 2023 [44]. The second study (RQ2) is currently under submission at the time of this paper was written. While the first two studies (RQ1-RQ2) are completed using a systematic literature review, the rest of the RQs will be investigated using mixed-method experimental studies. Background work (RQ1-RQ2) for the third study (RQ3) is currently completed, however, the design and evaluation part has not been completed yet. RQ4 is left intentionally broad. Data collection and analysis have already been performed for one of the research from the fourth RQ (RQ4), although the results are not reported in this paper. This work is expected to be submitted to the ACM Conference on Computer-Supported Cooperative Work and Social Computing (CSCW) or related HCI conferences. In the following, I will briefly discuss RQ1-RQ2 findings and planned research for RQ3-RQ4.
2 CONTRIBUTIONS TO DATE
2.1 RQ1: Values Encoded in ML Homelessness Service Provision Research
Public housing is a scarce resource. Therefore, matching the large homeless population with scarce housing resources involves complex and sometimes difficult decision-making. Data science and ML approaches present valuable opportunities for homelessness service provision [11, 22]. Recent data science approaches developed algorithms for matching housing resources with high-risk clients [9], predict who will need assistance or cost more in the future, and forecast who will re-enter shelters [24]. All of these approaches rely on Homeless Management Information System (HMIS) data 5, which is collected through collaborations and negotiations with agencies [9, 11, 47]. While the application of data science is capable of providing fast, low-cost, and scalable solutions for the staff, I would argue that homeless clients are vulnerable, and using their sensitive data in the data science or ML pipeline and prioritizing different ML values (RQ1) can have unintentional harmful consequences and harm thereof (e.g., [41, 42]). To deeply assess this, this study performed a critical analysis of research papers identified through a systematic literature review to uncover (1) what values were uplifted in the research papers, and (2) the impacts of prioritization of those values. This work is published [44], and only partial findings are discussed below.
2.1.1 Methodology.
This study followed a two-phase research method. First, the corpus was created using an in-depth systematic review of the literature of articles in CS/HCI and Science and Technology Studies (STS) that applied ML or statistical approaches in the context of homelessness [16, 40]. Second, thematic analysis was performed on the corpus [7]. A keyword-based search was carried out across the major databases, that is, the ACM digital library, Google Scholar, and IEEE Xplore, using various keywords, for example, “homeless”, “homelessness”, and “machine learning”, including their combination. In total, we had 88 candidate papers from CS/HCI and STS literature, of which around 50 papers were quantitative. To focus on RQ1, 40 papers (16 CS/HCI papers and 24 STS-focused papers) were included in the thematic analysis. Both inductive and deductive thematic analysis was performed [7]. Deductive coding utilized 67 ML values proposed in Birhane et al. [6] GitHub page, to identify ML values that were uplifted in ML homelessness service provision research. 6 Inductive coding was used to discover novel and surprising themes. Cross-coding or crosstabs was performed to discover the impact of dominant values on different values and themes.
2.1.2 Findings.
Deductive analysis indicates that novelty, performance, and identifying limitations were the three most commonly occurring values in the corpus. The novelty value has often manifested in the choice of algorithms, dataset use, particular homeless population studied (e.g., homeless women, youth, veterans), novel predictor or risk factor identification, and experimental site selection (e.g., specific US state or city). CS-focused papers primarily focused on proposing novel ML algorithms for different prediction tasks, however, STS-focused papers focused on identifying novel risk factors. Novelty value was equally prioritized across CS (48%) and STS (52.5%) papers from a technical lens rather than a human-focused lens despite the sensitivity surrounding homelessness. Performance values such as accuracy were used in the traditional sense, i.e., for justifying numerical correctness. Cross-tab analysis showed that 52.3% of the CS papers and 47.6% of the STS papers discussed performance. Although most papers included some discussion of the limitations of proposed ML interventions, however, those discussions often lacked human considerations in favor of technical limitations (e.g., dataset or sample limitations, limited predictor variable usage, and non-generality). The crosstabs further suggest that the STS (66. 6%) papers tend to include study limitations more often than the CS (33. 3%) papers.
The impact of these dominant values prioritization manifested in two ways: 1) unintended values collapse or values deprioritization 7 and 2) human deprioritization. Inductive analysis showed that the meaning of several algorithmically charged values, such as privacy and reproducibility, was deprioritized, fragmented, or collapsed in this context. For example, privacy value collapsed a) when un-anonymized datasets (e.g., sensitive data such as medical and mental-health records, jail service use history, and SSNs) were shared with the research team, and b) when authors asked difficult or even invasive questions (e.g., non-prosocial survival practices, sexual behavior while on street) during interviews and survey data collection. Privacy was deprioritized through clients’ lost autonomy about what information about them should or should not have been shared. Similarly, the reproducibility value collapsed a) when ML methods replicated vulnerability through algorithmic forecasting (e.g., someone who experienced homelessness during childhood or stayed in a homeless shelter or was born to a runaway parent is likely to experience homelessness during adulthood), b) when papers paid less attention to proposing transparent ML models and dataset use. In these ways, those research papers tend to reproduce vulnerability inadvertently and deprioritize those values.
The cross-coding assessment further suggests that prioritizing novelty and performance values was responsible for the ML values collapse mentioned above. Novelty and performance value prioritization also negatively impacted humans, causing human deprioritization through classification, categorization, prediction, and erasure through data pre-processing steps to algorithmic-specific parameter tuning steps in the data science pipeline [33]. Logistic regression was the most commonly used algorithm (20, 50%) in the corpus, and AUC/accuracy metrics were the most commonly used performance metrics (36, 90%), which contributed to ML values collapse. In RQ3, I will uplift the deprioritized values for designing a value-sensitive service assessment tool.
2.2 RQ2: Building Taxonomy and Critical Assessment of Homeless Person’s Assets
In RQ2, the focus was on the positive aspects, strengths, and assets.8 Prior research has indicated that persons experiencing homelessness use different assets that help them thrive within the challenging homeless environment, assets help them in transition, and eventually move out of homelessness [21, 36, 45, 51]. Homeless people’s assets are rarely showcased in HCI. Therefore, in RQ3, together with collaborators, I aspired to bring domain knowledge from homeless asset studies for HCI (1) to build a taxonomy of assets for future research and design, (2) to assess assets constraints and limitations that are recently discussed within HCI [53, 54]. This paper is currently under submission and only partial findings are reported below. The assets taxonomy developed from this work will be used in RQ3.
2.2.1 Methodology.
First, the corpus was created using a systematic literature review of papers that reported homeless persons’ assets; secondly, thematic analysis was performed to answer the RQs of this study [7]. The corpus was created using a keyword-based search using various terms such as, “homelessness”, “assets”, “strengths”, “resilience”, “survival”, “coping”, and “asset-based approaches”, including their combination. The final corpus included 40 papers, which were synthesized using thematic analysis. A bottom-up inductive thematic coding was applied, i.e., the analysis was driven by what is in the data. The research team systematically identified, and organized, insights into patterns of the dataset.
2.2.2 Findings.
Our findings show that homeless persons used various external assets (17, 42.5%), strengths (22, 55%), resilience (14, 35%), and internal assets (5, 12.5%) to thrive on the streets. In the following, I will briefly discuss internal and external assets and introduce assets critical assessment.
Internal assets, also known as personal strengths, consist of the ability to exercise self-efficacy, empathy, problem-solving abilities, and self-awareness [21]. 9, 10 We found that internal assets such as social competency and positive self-identity were associated with healthy development and resilience among homeless youth [26]. These assets also reduced their risk-taking behavior (e.g., violence, drug or alcohol use) among them. Bonds with pets and siblings were also identified as assets among youth [3]. For Aboriginal young women, reconnection with cultural knowledge and cultural efficacy, and restored identity helped them in the development of pride and hope, which in turn further contributed to their eventual transition out of homelessness [31, 32, 39]. In addition to internal assets, external assets (e.g., relationships, organizations) were also utilized among the homeless people [23]. For example, external assets available within the shelter, such as organization support and empowerment, contributed to a decrease in youth’s distress during their shelter stay and improved their life satisfaction and internal assets. Young women who became pregnant or had babies served as a motivational factor (e.g., hope), and were an enabler of re-evaluation of personal choices and the safety in place [39].
Despite assets’ positive impact on homeless persons’ lives, we found several ways in which asset usage was nuanced and context-specific. For example, studies show that external peers help in decreasing youths’ stress, by providing them with the necessary knowledge and learning opportunities for staying safe and strategies for surviving on the streets. However, we found contradictory evidence suggesting that these peer networks can also perpetuate and encourage youth to engage in extreme survival or non-prosocial strategies (e.g., selling drugs, prostitution) [18, 45]. Another subtle, yet problematic coping was observed when homeless persons’ became content and found comfort within the current limited living arrangements. Consequently, they stop developing goals for the future or try to get housed. All this contradictory evidence shows the dual nature of assets, thus hinting at the complexity of designing technology with them [53, 54].
3 EXPECTED NEXT STEPS
The background work (RQ1-RQ2) of this proposal was based on a systematic literature review, the rest of the studies (RQ3-RQ4) involve design and evaluation. Below, I will briefly discuss my proposed planned studies.
3.1 RQ3: Ethical Values-Informed Algorithmic Tool Design and Evaluation Study
The tool will be designed using the results of RQ1 and RQ2. First, findings from RQ1 indicate that current algorithmic or ML innovations in homeless service provision have been technical value-centered. These values prioritizations have caused important values to collapse and human deprioritization. Existing works also paid less attention to designing transparent algorithmic tools. Informed by these, I will uplift transparency and autonomy in RQ3. Second, current service prioritization tools exclusively rely upon causes of homelessness or risk-factors, and are deficit-focused. We need to uplift homeless persons’ assets that enable them to thrive in harsh homeless circumstances, to counteract stereotyping and stigma. The taxonomy developed in RQ2 will be used for design in RQ3. Once the prototype is available, it will be evaluated with the relevant stakeholders, including impacted stakeholders to understand their perceptions. Following VSD, I identified both direct stakeholders: homeless frontline workers or staff, i.e., caseworkers and volunteers, and indirect stakeholders: persons experiencing or have experienced homelessness in the past 5 years, and low-income individuals with prior or no homelessness experience. 11 Prior studies have shown that low-income individuals are on the verge or at high risk of becoming homeless [1, 4, 5, 12, 29], therefore, including their perspective is equally important. One of the foreseeable challenges is engaging homeless staff and sheltered clients without a partnering organization. I have attended numerous meetings, reached out to organizations in Boston, and achieved some initial progress. In that vein, I look forward to gathering guidance and further directions at CHI’24 DC.
3.2 RQ4: AI/ML literacy studies
RQ4 aims to empower low-income individuals by providing them access to knowledge about algorithmic impacts and commonly used ML algorithms. AI’s negative impact on poor and low-income individuals is prevalent (e.g., biases, and lack of transparency causing mistrust and uncertainty [17]), however, they are commonly excluded from AI/ML literacy interventions. Particularly, my thesis will center on those who are classified as low-income individuals because they access welfare resources, as well as public housing services (e.g., [4, 12, 29]). 12, 13 These service allocation decisions involve evaluation using of vulnerability assessment or service prioritization tools, and thus low-income people need to be aware of AI/ML. RQ4 AI/ML literacy studies will employ a mixed-method experimental design and will be evaluated using crowd-sourcing platforms. Participants will be pre-screened using the monthly household income as a threshold based on the current federal poverty guidelines of the US [5, 25]. This study will challenge the normative dominant narratives about who should or should not receive AI/ML literacy based on economic disadvantage and classification. The hope is to enhance the inclusion of impacted individuals in AI/ML. One limitation of this study is that it unintentionally prioritizes those who have access to technologies (e.g., internet connectivity, phone, laptop) and the required knowledge of how to use them.
All these works will culminate in contributing to HCI and ML scholarship as well as public services through algorithmic reform – by reducing AI/ML access and inclusion barriers. More precisely, the contributions include: 1) designing an ethical values-informed service assessment tool in the homelessness context and 2) disseminating AI/ML knowledge accessible to low-income individuals.
4 EXPECTED BENEFITS AND CONTRIBUTIONS AT CHI 2024 DOCTORAL CONSORTIUM
I am currently a third-year Ph.D. Candidate in the Khoury College of Computer Sciences at Northeastern University, supervised by Dr. Alexandra To. I have completed my Ph.D. comprehensive exam (i.e., Ph.D. proposal), and my projected dissertation completion date is between Late Fall 2024 to early Spring 2025. On average, Khoury Ph.D. students graduate in about six and a half years. Upon graduation, I aspire to pursue a research-oriented career either in academia or in the industry. I have never attended any Doctoral Consortium (DC). I have completed the preliminary works of my dissertation, however, I am still in the process of completing the rest of the planned studies (RQ3-RQ4) within the next year. ACM CHI DC 2024 will provide me with a great opportunity to gather new ideas and perspectives, a chance for critical reflection, and an opportunity to gain formative feedback on various aspects of my research due to the sensitivity that surrounds working with marginalized people. In addition, I look forward to meeting HCI and multi-disciplinary research experts, learning from them, connecting with other mentors and Ph.D. students, and identifying potential future collaboration opportunities.
There are several ways I can also contribute and be helpful to other ACM CHI DC’24 attendees. My current work entirely focuses on HCI, specifically, I draw inspiration from the value-sensitive design, human-centered AI [2, 19, 42, 44], and on social justice more broadly [14, 37, 46]. I have background experience in algorithm design and development, ML, mixed-methods study design, and advanced statistical data analysis. Also, I am experienced in paper publishing and have done peer-reviewing for quite some time. All these experiences make me equally beneficial to other attendees of the CHI DC. I can provide constructive and valuable feedback and help in interdisciplinary studies.
ACKNOWLEDGMENTS
I express my gratitude to the research collaborators and research assistants, especially Lingqing Wang and Laveda Chen, and all the study participants for their time. Additionally, I extend my heartfelt appreciation to my advisor, Dr. Alexandra To, and my thesis committee members, Dr. Angela D. R. Smith, Dr. Dakuo Wang, and Dr. Varun Mishra, for their unwavering support, mentorship, critical and invaluable feedback throughout this journey.
Footnotes
1 https://www.census.gov/newsroom/press-releases/2023/income-poverty-health-insurance-coverage.html
Footnote2 https://justicegap.lsc.gov/resource/section-2-todays-low-income-america/
Footnote3 https://todayshomeowner.com/general/guides/national-homeless-facts-and-statistics/
Footnote4 https://everyonehome.org/wp-content/uploads/2016/02/VI-SPDAT-2.0-Single-Adults.pdf
Footnote5 https://www.hudexchange.info/programs/hmis/
Footnote6 https://github.com/wagnew3/The-Values-Encoded-in-Machine-Learning-Research
Footnote7 I define “values collapse” as any ML values that were deprioritized or required re-definition as they break down or uphold their true definition as we know it in ML.
Footnote8 Assets encompass all potential resources in the community, including financial resources, institutions, and property; and intangible and non-financial resources, consisting of people, connections, relationships, social competencies, positive values and life perspectives, individual knowledge, strengths, and capacities [5, 54] existing in the lives of persons experiencing homelessness.
Footnote9 Internal assets generally identify characteristics and behaviors that reflect positive internal growth and development, e.g., these assets are about positive values and life perspectives, skills and adaptive characteristics, social competencies, positive self-identity, and commitment to learning [13, 21, 23]
Footnote10 https://outreach.msu.edu/documents/presentations/capcomm.pdf
Footnote11 “indirect stakeholders are those individuals who are also impacted by the system, though they never interact directly with it” [19]
Footnote12 According to the 2023 U.S. federal poverty guidelines, individuals who had a household income below 150% of the poverty line, that is annual incomes below the range of 50, 560forafamilyofeightand 14,580 for an individual or family of size one, are classified as low-income.
Footnote13 https://aspe.hhs.gov/topics/poverty-economic-mobility/poverty-guidelines
Footnote
- Claudia Aiken, Vincent J Reina, and Dennis P Culhane. 2021. Understanding low-income hispanic housing challenges and the use of housing and homelessness assistance. Cityscape 23, 2 (2021), 123–158.Google Scholar
- Cecilia Aragon, Shion Guha, Marina Kogan, Michael Muller, and Gina Neff. 2022. Human-centered data science: an introduction. MIT Press, USA.Google Scholar
- Andrew J Barnes, Jace Gilbertson, and Debanjana Chatterjee. 2018. Emotional health among youth experiencing family homelessness. Pediatrics 141, 4 (2018), na.Google Scholar
- Ellen L Bassuk, Linda F Weinreb, John C Buckner, Angela Browne, Amy Salomon, and Shari S Bassuk. 1996. The characteristics and needs of sheltered homeless and low-income housed mothers. Jama 276, 8 (1996), 640–646.Google ScholarCross Ref
- Jodi Benenson. 2017. Civic engagement and economic opportunity among low-income individuals: An asset-based approach. VOLUNTAS: International Journal of Voluntary and Nonprofit Organizations 28 (2017), 988–1014.Google ScholarCross Ref
- Abeba Birhane, Pratyusha Kalluri, Dallas Card, William Agnew, Ravit Dotan, and Michelle Bao. 2022. The Values Encoded in Machine Learning Research. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency (Seoul, Republic of Korea) (FAccT ’22). Association for Computing Machinery, New York, NY, USA, 173–184. https://doi.org/10.1145/3531146.3533083Google ScholarDigital Library
- Virginia Braun and Victoria Clarke. 2012. Thematic analysis.American Psychological Association, USA.Google Scholar
- Thomas Byrne, Travis Baggett, Thomas Land, Dana Bernson, Maria-Elena Hood, Cheryl Kennedy-Perez, Rodrigo Monterrey, David Smelson, Marc Dones, and Monica Bharel. 2020. A classification model of homelessness using integrated administrative data: Implications for targeting interventions to improve the housing status, health and well-being of a highly vulnerable population. PloS one 15, 8 (2020), e0237905.Google ScholarCross Ref
- Hau Chan, Eric Rice, Phebe Vayanos, Milind Tambe, and Matthew Morton. 2017. Evidence from the past: AI decision aids to improve housing systems for homeless youth. In 2017 AAAI Fall Symposium Series. AAAI, USA, na.Google Scholar
- Charalampos Chelmis, Wenting Qi, and Wonhyung Lee. 2021. Challenges and opportunities in using data science for homelessness service provision. In Companion proceedings of the web conference 2021. ACM, USA, 128–135.Google ScholarDigital Library
- Charalampos Chelmis, Wenting Qi, Wonhyung Lee, and Stephanie Duncan. 2021. Smart homelessness service provision with machine learning. Procedia Computer Science 185 (2021), 9–18.Google ScholarCross Ref
- Hannah Cohen-Cline, Kyle Jones, and Keri Vartanian. 2021. Direct and indirect pathways between childhood instability and adult homelessness in a low-income population. Children and Youth Services Review 120 (2021), 105707.Google ScholarCross Ref
- Sam J Cooley, Mary L Quinton, Mark JG Holland, Benjamin J Parry, and Jennifer Cumming. 2019. The experiences of homeless youth when using strengths profiling to identify their character strengths. Frontiers in psychology 10 (2019), 2036.Google Scholar
- Sasha Costanza-Chock. 2020. Design justice: Community-led practices to build the worlds we need. The MIT Press, USA.Google Scholar
- Tawanna R. Dillahunt, Sandy Ng, Michelle Fiesta, and Zengguang Wang. 2016. Do Massive Open Online Course Platforms Support Employability?. In Proceedings of the 19th ACM Conference on Computer-Supported Cooperative Work & Social Computing (San Francisco, California, USA) (CSCW ’16). Association for Computing Machinery, New York, NY, USA, 233–244. https://doi.org/10.1145/2818048.2819924Google ScholarDigital Library
- Tawanna R Dillahunt, Xinyi Wang, Earnest Wheeler, Hao Fei Cheng, Brent Hecht, and Haiyi Zhu. 2017. The sharing economy in computing: A systematic literature review. Proceedings of the ACM on Human-Computer Interaction 1, CSCW (2017), 1–26.Google ScholarDigital Library
- Virginia Eubanks. 2018. Automating inequality: How high-tech tools profile, police, and punish the poor. St. Martin’s Press, USA.Google ScholarDigital Library
- Kristin M Ferguson, Kimberly Bender, Sanna Thompson, Bin Xie, and David Pollio. 2011. Correlates of street-survival behaviors in homeless young adults in four US Cities.American Journal of Orthopsychiatry 81, 3 (2011), 401.Google Scholar
- Batya Friedman, Peter H. Kahn, Alan Borning, and Alina Huldtgren. 2013. Value Sensitive Design and Information Systems. Springer Netherlands, Dordrecht, 55–95. https://doi.org/10.1007/978-94-007-7844-3_4Google ScholarCross Ref
- Batya Friedman and Helen Nissenbaum. 1996. Bias in computer systems. ACM Transactions on information systems (TOIS) 14, 3 (1996), 330–347.Google ScholarDigital Library
- Michael J Furlong, Kristin M Ritchey, and Lindsey M O’Brennan. 2009. Developing norms for the California Resilience Youth Development Module: Internal assets and school resources subscales. The California school psychologist 14 (2009), 35–46.Google Scholar
- Yuan Gao, Sanmay Das, and Patrick Fowler. 2017. Homelessness service provision: a data science perspective. In Workshops at the thirty-first AAAI conference on artificial intelligence. AAAI, USA, 5.Google Scholar
- Hillary J Heinze. 2013. Beyond a bed: Support for positive development for youth residing in emergency shelters. Children and Youth Services Review 35, 2 (2013), 278–286.Google ScholarCross Ref
- Boyeong Hong, Awais Malik, Jack Lundquist, Ira Bellach, and Constantine E Kontokosta. 2018. Applications of machine learning methods to predict readmission and length-of-stay for homeless families: The case of win shelters in new york city. Journal of Technology in Human Services 36, 1 (2018), 89–104.Google ScholarCross Ref
- Aarti Israni, Nicole B Ellison, and Tawanna R Dillahunt. 2021. ’A Library of People’ Online Resource-Seeking in Low-Income Communities. Proceedings of the ACM on Human-Computer Interaction 5, CSCW1 (2021), 1–28.Google Scholar
- Kristen M Aggerbeck Kessler, Debanjana Chatterjee, Rebecca Shlafer, and Andrew J Barnes. 2018. Adolescent connectedness with parents promotes resilience among homeless youth. Children 5, 7 (2018), 96.Google ScholarCross Ref
- Tzu-Sheng Kuo, Hong Shen, Jisoo Geum, Nev Jones, Jason I. Hong, Haiyi Zhu, and Kenneth Holstein. 2023. Understanding Frontline Workers’ and Unhoused Individuals’ Perspectives on AI Used in Homeless Services. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems(CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 860, 17 pages. https://doi.org/10.1145/3544548.3580882Google ScholarDigital Library
- Christopher A. Le Dantec, Erika Shehan Poole, and Susan P. Wyche. 2009. Values as lived experience: evolving value sensitive design in support of value discovery. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (Boston, MA, USA) (CHI ’09). Association for Computing Machinery, New York, NY, USA, 1141–1150. https://doi.org/10.1145/1518701.1518875Google ScholarDigital Library
- Erika R Lehmann, Christiana M Drake, Philip H Kass, and Sara B Nichols. 2007. Risk factors for first-time homelessness in low-income women. American Journal of Orthopsychiatry 77, 1 (2007), 20–28.Google ScholarCross Ref
- Duri Long and Brian Magerko. 2020. What is AI Literacy? Competencies and Design Considerations. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems(CHI ’20). Association for Computing Machinery, New York, NY, USA, 1–16. https://doi.org/10.1145/3313831.3376727Google ScholarDigital Library
- Katrina Milaney, KAYLEE Ramage, X Yang Fang, and MARILYN Louis. 2017. Understanding mothers experiencing homelessness: A gendered approach to finding solutions for family homelessness. Canadian Observatory on Homelessness Press Toronto, Canada, Canada.Google Scholar
- David P Moxley and Olivia GM Washington. 2016. Souls in extremis: Enacting processes of recovery from homelessness among older African American women. Journal of religion and health 55 (2016), 1038–1054.Google ScholarCross Ref
- Michael Muller, Ingrid Lange, Dakuo Wang, David Piorkowski, Jason Tsay, Q. Vera Liao, Casey Dugan, and Thomas Erickson. 2019. How Data Science Workers Work with Data: Discovery, Capture, Curation, Design, Creation. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems (Glasgow, Scotland Uk) (CHI ’19). Association for Computing Machinery, New York, NY, USA, 1–15. https://doi.org/10.1145/3290605.3300356Google ScholarDigital Library
- Davy Tsz Kit Ng, Jac Ka Lok Leung, Samuel Kai Wah Chu, and Maggie Shen Qiao. 2021. Conceptualizing AI literacy: An exploratory review. Computers and Education: Artificial Intelligence 2 (2021), 100041.Google ScholarCross Ref
- Safiya Umoja Noble. 2018. Algorithms of oppression. In Algorithms of oppression. New York university press, USA.Google Scholar
- Brooke Dolenc Nott and Samuel Vuchinich. 2016. Homeless adolescents’ perceptions of positive development: A comparative study. In Child & Youth Care Forum, Vol. 45. Springer, Springer, USA, 865–886.Google Scholar
- Ihudiya Finda Ogbonnaya-Ogburu, Angela D.R. Smith, Alexandra To, and Kentaro Toyama. 2020. Critical Race Theory for HCI. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems(CHI ’20). Association for Computing Machinery, New York, NY, USA, 1–16. https://doi.org/10.1145/3313831.3376392Google ScholarDigital Library
- Cathy O’neil. 2017. Weapons of math destruction: How big data increases inequality and threatens democracy. Crown, USA.Google Scholar
- Lia Ruttan, Patricia Laboucane-Benson, and Brenda Munro. 2012. Does a baby help young women transition out of homelessness? Motivation, coping, and parenting. Journal of Family Social Work 15, 1 (2012), 34–49.Google ScholarCross Ref
- Devansh Saxena, Karla Badillo-Urquiola, Pamela J. Wisniewski, and Shion Guha. 2020. A Human-Centered Review of Algorithms used within the U.S. Child Welfare System. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems(CHI ’20). Association for Computing Machinery, New York, NY, USA, 1–15. https://doi.org/10.1145/3313831.3376229Google ScholarDigital Library
- Renee Shelby, Shalaleh Rismani, Kathryn Henne, AJung Moon, Negar Rostamzadeh, Paul Nicholas, N’Mah Yilla-Akbari, Jess Gallegos, Andrew Smart, Emilio Garcia, and Gurleen Virk. 2023. Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm Reduction. In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society(AIES ’23). Association for Computing Machinery, New York, NY, USA, 723–741. https://doi.org/10.1145/3600211.3604673Google ScholarDigital Library
- Dilruba Showkat and Eric P. S. Baumer. 2022. “It’s Like the Value System in the Loop”: Domain Experts’ Values Expectations for NLP Automation. In Proceedings of the 2022 ACM Designing Interactive Systems Conference(DIS ’22). Association for Computing Machinery, New York, NY, USA, 100–122. https://doi.org/10.1145/3532106.3533483Google ScholarDigital Library
- Dilruba Showkat, Shereen Bellamy, and Alexandra To. 2022. Fair and Trustworthy Welfare Systems: Rethinking Explainable AI in the Public Sector. In 2nd CHI HC-XAI Workshop. ACM, USA. https://doi.org/10.13140/RG.2.2.23902.72001Google ScholarCross Ref
- Dilruba Showkat, Angela D. R. Smith, Wang Lingqing, and Alexandra To. 2023. “Who is the right homeless client?”: Values in Algorithmic Homelessness Service Provision and Machine Learning Research. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems(CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 394, 21 pages. https://doi.org/10.1145/3544548.3581010Google ScholarDigital Library
- Sanna J Thompson, Tiffany N Ryan, Katherine L Montgomery, Angie Del Prado Lippman, Kimberly Bender, and Kristin Ferguson. 2016. Perceptions of resiliency and coping: Homeless young adults speak out. Youth & Society 48, 1 (2016), 58–76.Google ScholarCross Ref
- Alexandra To, Angela D. R. Smith, Dilruba Showkat, Adinawa Adjagbodjou, and Christina Harrington. 2023. Flourishing in the Everyday: Moving Beyond Damage-Centered Design in HCI for BIPOC Communities. In Proceedings of the 2023 ACM Designing Interactive Systems Conference(DIS ’23). Association for Computing Machinery, New York, NY, USA, 917–933. https://doi.org/10.1145/3563657.3596057Google ScholarDigital Library
- Halil Toros and Daniel Flaming. 2017. Prioritizing which homeless people get housing using predictive algorithms. Available at SSRN 2960410 na, na (2017).Google Scholar
- Halil Toros, Daniel Flaming, and Patrick Burns. 2019. Early intervention to prevent persistent homelessness: Predictive models for identifying unemployed workers and young adults who become persistently homeless. Available at SSRN 3370634 na, na (2019).Google Scholar
- Dhaval Vyas and Tawanna Dillahunt. 2017. Everyday resilience: Supporting resilient strategies among low socioeconomic status communities. Proceedings of the ACM on Human-Computer Interaction 1, CSCW (2017), 1–21.Google ScholarDigital Library
- Claudia Wagner, Markus Strohmaier, Alexandra Olteanu, Emre Kıcıman, Noshir Contractor, and Tina Eliassi-Rad. 2021. Measuring algorithmically infused societies. Nature 595, 7866 (2021), 197–204.Google Scholar
- Olivia GM Washington, David P Moxley, Lois Garriott, and Jennifer P Weinberger. 2009. Five dimensions of faith and spiritually of older African American women transitioning out of homelessness. Journal of Religion and Health 48 (2009), 431–444.Google ScholarCross Ref
- Daricia Wilkinson, Kate Crawford, Hanna Wallach, Deborah Raji, Bogdana Rakova, Ranjit Singh, Angelika Strohmayer, and Ethan Zuckerman. 2023. Accountability in Algorithmic Systems: From Principles to Practice. In Extended Abstracts of the 2023 CHI Conference on Human Factors in Computing Systems(CHI EA ’23). Association for Computing Machinery, New York, NY, USA, Article 521, 4 pages. https://doi.org/10.1145/3544549.3583747Google ScholarDigital Library
- Marisol Wong-Villacres, Sheena Erete, Aakash Gautam, Azra Ismail, Neha Kumar, Lucy Pei, Wendy Roldan, Veronica Ahumada-Newhart, Karla Badillo-Urquiola, J Maya Hernandez, 2022. Elevating strengths and capacities: the different shades of assets-based design in HCI. Interactions 29, 5 (2022), 28–33.Google ScholarDigital Library
- Marisol Wong-Villacres, Aakash Gautam, Wendy Roldan, Lucy Pei, Jessa Dickinson, Azra Ismail, Betsy DiSalvo, Neha Kumar, Tammy Clegg, Sheena Erete, Emily Roden, Nithya Sambasivan, and Jason Yip. 2020. From Needs to Strengths: Operationalizing an Assets-Based Design of Technology. In Companion Publication of the 2020 Conference on Computer Supported Cooperative Work and Social Computing (Virtual Event, USA) (CSCW ’20 Companion). Association for Computing Machinery, New York, NY, USA, 527–535. https://doi.org/10.1145/3406865.3418594Google ScholarDigital Library
Index Terms
- Towards Algorithmic Reform: Low-Income Individuals Inclusion in AI/ML Literacy and Ethical Values-Informed Tool Design
Recommendations
How Different Groups Prioritize Ethical Values for Responsible AI
FAccT '22: Proceedings of the 2022 ACM Conference on Fairness, Accountability, and TransparencyPrivate companies, public sector organizations, and academic groups have outlined ethical values they consider important for responsible artificial intelligence technologies. While their recommendations converge on a set of central values, little is ...
Towards ethical aspects on artificial intelligence
AIKED'09: Proceedings of the 8th WSEAS international conference on Artificial intelligence, knowledge engineering and data basesThis paper presents the role of ethics in developing artificial intelligence, and how the artificial intelligence could change our perspective, because artificial intelligence in fact is all around us. Artificial intelligence is an important part of our ...
Ethical dilemmas and PD as important steps towards critical e-government design
PDC '14: Proceedings of the 13th Participatory Design Conference: Short Papers, Industry Cases, Workshop Descriptions, Doctoral Consortium papers, and Keynote abstracts - Volume 2The delivering of public services to citizens through the internet -- also known as e-government - has gained serious momentum, driven by political ambitions of improved efficiency. E-government, however, is considered complex and e-government failures ...
Comments