Human-centered Machine Research Articles

Developing a Course on Human-Centered Machine Learning-Based Product Design

Developing a Course on Human-Centered Machine Learning-Based Product Design

Exploitation Process for Machine Learning Systems

In recent years, many companies have been involved in the development of machine learning systems. However, developing machine learning systems presents unique challenges compared with the traditional IT system development. In software engineering, research has been conducted on software development methodologies for machine learning systems, but most of these studies focused on the development process of individual machine learning systems, and limited research has investigated the exploitation process of continuously enhancing machine learning systems after their development and deployment. Thus, this paper focuses on the exploitation process, in which the collaboration between the machine learning systems and humans evolves gradually. Specifically, four stages of the exploitation process are proposed: visualization for machine learning, human-centered machine learning assistance, machine learning-centered human assistance, and autonomy without human involvement. Then, the characteristics and challenges of each stage are clarified. In addition, this paper examines the important aspect of trust in this context. The findings of this study are expected to provide guidelines for advancing machine learning systems incrementally in collaboration with humans from the perspective of the exploitation process rather than focusing on the development process.

LegalAsst: Human-centered and AI-empowered machine to enhance court productivity and legal assistance

We propose autonomous software (namely, LegalAsst ) as a step toward an AI-empowered but human-centered machine focused on enhancing court productivity and legal assistance. LegalAsst aims to provide explainable, traceable, and controllable legal assistance and references for lawyers, judges, government officials, and the general public. To achieve this goal, it collates, processes, distills, and visualizes the whole judgment procedure. It streamlines and semi-automates the judgment procedure through case analysis, legislation analysis, and judicial decision-making. Specifically, to make laws and cases easier to navigate and understand, we incorporate structured representations to perform them. Then based on structured representations, we take a step further by introducing a decision-tree-based judgment, making the entire judging process visible and tractable. Our system not only tracks the procedural aspects of judgments but also incorporates modification capabilities, enabling the consideration of the most up-to-date legislation and societal factors to generate more adaptable judgment outcomes.1

U-PASS: An uncertainty-guided deep learning pipeline for automated sleep staging

With the increasing prevalence of machine learning in critical fields like healthcare, ensuring the safety and reliability of these systems is crucial. Estimating uncertainty plays a vital role in enhancing reliability by identifying areas of high and low confidence and reducing the risk of errors. This study introduces U-PASS, a specialized human-centered machine learning pipeline tailored for clinical applications, which effectively communicates uncertainty to clinical experts and collaborates with them to improve predictions. U-PASS incorporates uncertainty estimation at every stage of the process, including data acquisition, training, and model deployment. Training is divided into a supervised pre-training step and a semi-supervised recording-wise finetuning step. We apply U-PASS to the challenging task of sleep staging and demonstrate that it systematically improves performance at every stage. By optimizing the training dataset, actively seeking feedback from domain experts for informative samples, and deferring the most uncertain samples to experts, U-PASS achieves an impressive expert-level accuracy of 85% on a challenging clinical dataset of elderly sleep apnea patients. This represents a significant improvement over the starting point at 75% accuracy. The largest improvement gain is due to the deferral of uncertain epochs to a sleep expert. U-PASS presents a promising AI approach to incorporating uncertainty estimation in machine learning pipelines, improving their reliability and unlocking their potential in clinical settings.

Sliding into My DMs: Detecting Uncomfortable or Unsafe Sexual Risk Experiences within Instagram Direct Messages Grounded in the Perspective of Youth

We collected Instagram data from 150 adolescents (ages 13-21) that included 15,547 private message conversations of which 326 conversations were flagged as sexually risky by participants. Based on this data, we leveraged a human-centered machine learning approach to create sexual risk detection classifiers for youth social media conversations. Our Convolutional Neural Network (CNN) and Random Forest models outperformed in identifying sexual risks at the conversation-level (AUC=0.88), and CNN outperformed at the message-level (AUC=0.85). We also trained classifiers to detect the severity risk level (i.e., safe, low, medium-high) of a given message with CNN outperforming other models (AUC=0.88). A feature analysis yielded deeper insights into patterns found within sexually safe versus unsafe conversations. We found that contextual features (e.g., age, gender, and relationship type) and Linguistic Inquiry and Word Count (LIWC) contributed the most for accurately detecting sexual conversations that made youth feel uncomfortable or unsafe. Our analysis provides insights into the important factors and contextual features that enhance automated detection of sexual risks within youths' private conversations. As such, we make valuable contributions to the computational risk detection and adolescent online safety literature through our human-centered approach of collecting and ground truth coding private social media conversations of youth for the purpose of risk classification.

Toward Practices for Human-Centered Machine Learning

Designing for the social, cultural, and ethical implications of ML are just as important as its technical advances.

Contrasting Explain-ML with Interpretability Machine Learning Tools in Light of Interactive Machine Learning Principles

The way Complex Machine Learning (ML) models generate their results is not fully understood, including by very knowledgeable users. If users cannot interpret or trust the predictions generated by the model, they will not use them. Furthermore, the human role is often not properly considered in the development of ML systems. In this article, we present the design, implementation and evaluation of Explain-ML, an Interactive Machine Learning (IML) system for Explainable Machine Learning that follows the principles of Human-Centered Machine Learning (HCML). We assess the user experience with the Explain-ML interpretability strategies, contrasting them with the analysis of how other IML tools address the IML principles. To do so, we have conducted an analysis of the results of the evaluation of Explain-ML with potential users in light of principles for IML systems design and a systematic inspection of three other tools – Rulematrix, Explanation Explorer and ATMSeer – using the Semiotic Inspection Method (SIM). Our results generated positive indicators regarding Explain-ML and the process that guided its development. Our analyses also highlighted aspects of the IML principles that are relevant from the users’ perspective. By contrasting the results with Explain-ML and SIM inspections of the other tools we were able to identify common interpretability strategies. We believe that the results reported in this work contribute to the understanding and consolidation of the IML principles, ultimately advancing the knowledge in HCML.

Conceptualization of an Assessment System to Measure Vocabulary in Science (Project MELVA-S)

This paper aims to report the conceptualization of a web-based Automated Speech Recognition Scoring System, project MELVA-S (Measuring the English Language Vocabulary Acquisition of Latinx Bilingual Students), to measure the science vocabulary of second- and third-grade Latinx students. ELVA (English Learner Vocabulary Acquisition First Iteration) and ELVA-2 (English Learner Vocabulary Acquisition Second Iteration) focused on student’s learning and comprehension on science vocabularies. Both of the iterations are the foundation to build MELVA-S, which intends to measure and evaluate student’s answers with greater accuracy with Machine Learning. As a web-based agent, this system increases satisfaction for both teachers’ and students’ User Experience (UX) from content, design, and engineering perspectives. The project utilized a design-thinking approach and prototyped both the algorithm and the automated system interfaces. Future iterations of ELVA-2 and MELVA-S could consider adopting a Human-Centered Machine Learning approach, implemented with incremental improvements that include evaluation and testing with users, to keep enhancing both usability and functionality of the system for better UX.

Integrating Data and Model Space in Ensemble Learning by Visual Analytics

Ensembles of classifier models typically deliver superior performance and can outperform single classifier models given a dataset and classification task at hand. However, the gain in performance comes together with the lack of comprehensibility, posing a challenge to understand how each model affects the classification outputs and from where the errors come. We propose a tight visual integration of the data and the model space for exploring and combining classifier models. We introduce an interactive workflow that builds upon the visual integration and enables the effective exploration of classification outputs and models. The involvement of the user is key to our approach. Therefore, we elaborate on the role of the human and connect our approach to theoretical frameworks on human-centered machine learning. We showcase the usefulness of our approach and the integration of the user via binary and multiclass classification problems. Based on ensembles automatically selected by a standard ensemble selection algorithm, the user can manipulate models and alternative combinations.

Toward User-Driven Sound Recognizer Personalization with People Who Are d/Deaf or Hard of Hearing

Automated sound recognition tools can be a useful complement to d/Deaf and hard of hearing (DHH) people's typical communication and environmental awareness strategies. Pre-trained sound recognition models, however, may not meet the diverse needs of individual DHH users. While approaches from human-centered machine learning can enable non-expert users to build their own automated systems, end-user ML solutions that augment human sensory abilities present a unique challenge for users who have sensory disabilities: how can a DHH user, who has difficulty hearing a sound themselves, effectively record samples to train an ML system to recognize that sound? To better understand how DHH users can drive personalization of their own assistive sound recognition tools, we conducted a three-part study with 14 DHH participants: (1) an initial interview and demo of a personalizable sound recognizer, (2) a week-long field study of in situ recording, and (3) a follow-up interview and ideation session. Our results highlight a positive subjective experience when recording and interpreting training data in situ, but we uncover several key pitfalls unique to DHH users---such as inhibited judgement of representative samples due to limited audiological experience. We share implications of these results for the design of recording interfaces and human-the-the-loop systems that can support DHH users to build sound recognizers for their personal needs.

A Review of Recent Deep Learning Approaches in Human-Centered Machine Learning.

After Deep Learning (DL) regained popularity recently, the Artificial Intelligence (AI) or Machine Learning (ML) field is undergoing rapid growth concerning research and real-world application development. Deep Learning has generated complexities in algorithms, and researchers and users have raised concerns regarding the usability and adoptability of Deep Learning systems. These concerns, coupled with the increasing human-AI interactions, have created the emerging field that is Human-Centered Machine Learning (HCML). We present this review paper as an overview and analysis of existing work in HCML related to DL. Firstly, we collaborated with field domain experts to develop a working definition for HCML. Secondly, through a systematic literature review, we analyze and classify 162 publications that fall within HCML. Our classification is based on aspects including contribution type, application area, and focused human categories. Finally, we analyze the topology of the HCML landscape by identifying research gaps, highlighting conflicting interpretations, addressing current challenges, and presenting future HCML research opportunities.

Computational Media Intelligence: Human-Centered Machine Analysis of Media

Media is created by humans for humans to tell stories. There exists a natural and imminent need for creating human-centered media analytics to illuminate the stories being told and to understand their impact on individuals and society at large. An objective understanding of media content has numerous applications for different stakeholders, from creators to decision-/policy-makers to consumers. Advances in multimodal signal processing and machine learning (ML) can enable detailed and nuanced characterization of media content (of who, what, how, where, and why) at scale. They can also aid our understanding of the impact of media on a range of issues, including individual experiences, behavioral, cultural, and societal trends, and commercial outcomes. Modern deep learning models combined with audiovisual signal processing can analyze entertainment media, such as Film & TV content to quantify gender, age, and race representations. This creates awareness in an objective way that was hitherto impossible. On the other hand, text mining and natural language processing allow nuanced understanding of language use and spoken interactions in media, such as News to track patterns and trends across different contexts. Moreover, advances in human sensing have enabled us to directly measure the influence of media on an individual’s physiology (and brain), while social media analysis enables tracking the societal impact of media content on different cross sections of the society. This article reviews representative methodologies and algorithms, tools, and systems advancing human-centered media understanding through ML in the pursuit of developing computational media intelligence.

A recommendation and risk classification system for connecting rough sleepers to essential outreach services

Abstract Rough sleeping is a chronic experience faced by some of the most disadvantaged people in modern society. This paper describes work carried out in partnership with Homeless Link (HL), a UK-based charity, in developing a data-driven approach to better connect people sleeping rough on the streets with outreach service providers. HL's platform has grown exponentially in recent years, leading to thousands of alerts per day during extreme weather events; this overwhelms the volunteer-based system they currently rely upon for the processing of alerts. In order to solve this problem, we propose a human-centered machine learning system to augment the volunteers' efforts by prioritizing alerts based on the likelihood of making a successful connection with a rough sleeper. This addresses capacity and resource limitations whilst allowing HL to quickly, effectively, and equitably process all of the alerts that they receive. Initial evaluation using historical data shows that our approach increases the rate at which rough sleepers are found following a referral by at least 15% based on labeled data, implying a greater overall increase when the alerts with unknown outcomes are considered, and suggesting the benefit in a trial taking place over a longer period to assess the models in practice. The discussion and modeling process is done with careful considerations of ethics, transparency, and explainability due to the sensitive nature of the data involved and the vulnerability of the people that are affected.

Who is the "Human" in Human-Centered Machine Learning

"Human-centered machine learning" (HCML) combines human insights and domain expertise with data-driven predictions to answer societal questions. This area's inherent interdisciplinarity causes tensions in the obligations researchers have to the humans whose data they use. This paper studies how scientific papers represent human research subjects in HCML. Using mental health status prediction on social media as a case study, we conduct thematic discourse analysis on 55 papers to examine these representations. We identify five discourses that weave a complex narrative of who the human subject is in this research: Disorder/Patient, Social Media, Scientific, Data/Machine Learning, and Person. We show how these five discourses create paradoxical subject and object representations of the human, which may inadvertently risk dehumanization. We also discuss the tensions and impacts of interdisciplinary research; the risks of this work to scientific rigor, online communities, and mental health; and guidelines for stronger HCML research in this nascent area.

The Evolution of Human-Autonomy Teams in Remotely Piloted Aircraft Systems Operations

The focus of this current research is two-fold: (1) to understand how team interaction in human-autonomy teams (HAT)s evolve in the Remotely Piloted Aircraft Systems (RPAS) task context, and (2) to understand how HATs respond to three types of failures (automation, autonomy, and cyber-attack) over time. We summarize the findings from three of our recent experiments regarding the team interaction within HAT over time in the dynamic context of RPAS. For the first and the second experiments, we summarize general findings related to team member interaction of a three-member team over time, by comparison of HATs with all-human teams. In the third experiment, which extends beyond the first two experiments, we investigate HAT evolution when HATs are faced with three types of failures during the task. For all three of these experiments, measures focus on team interactions and temporal dynamics consistent with the theory of interactive team cognition. We applied Joint Recurrence Quantification Analysis, to communication flow in the three experiments. One of the most interesting and significant findings from our experiments regarding team evolution is the idea of entrainment, that one team member (the pilot in our study, either agent or human) can change the communication behaviors of the other teammates over time, including coordination, and affect team performance. In the first and second studies, behavioral passiveness of the synthetic teams resulted in very stable and rigid coordination in comparison to the all-human teams that were less stable. Experimenter teams demonstrated metastable coordination (not rigid nor unstable) and performed better than rigid and unstable teams during the dynamic task. In the third experiment, metastable behavior helped teams overcome all three types of failures. These summarized findings address three potential future needs for ensuring effective HAT: (1) training of autonomous agents on the principles of teamwork, specifically understanding tasks and roles of teammates, (2) human-centered machine learning design of the synthetic agent so the agents can better understand human behavior and ultimately human needs, and (3) training of human members to communicate and coordinate with agents due to current limitations of Natural Language Processing of the agents.

Introduction to the Special Issue on Human-Centered Machine Learning

Machine learning is one of the most important and successful techniques in contemporary computer science. Although it can be applied to myriad problems of human interest, research in machine learning is often framed in an impersonal way, as merely algorithms being applied to model data. However, this viewpoint hides considerable human work of tuning the algorithms, gathering the data, deciding what should be modeled in the first place, and using the outcomes of machine learning in the real world. Examining machine learning from a human-centered perspective includes explicitly recognizing human work, as well as reframing machine learning workflows based on situated human working practices, and exploring the co-adaptation of humans and intelligent systems. A human-centered understanding of machine learning in human contexts can lead not only to more usable machine learning tools, but to new ways of understanding what machine learning is good for and how to make it more useful. This special issue brings together nine articles that present different ways to frame machine learning in a human context. They represent very different application areas (from medicine to audio) and methodologies (including machine learning methods, human-computer interaction methods, and hybrids), but they all explore the human contexts in which machine learning is used. This introduction summarizes the articles in this issue and draws out some common themes.

What you see is what you can change: Human-centered machine learning by interactive visualization

Visual analytics (VA) systems help data analysts solve complex problems interactively, by integrating automated data analysis and mining, such as machine learning (ML) based methods, with interactive visualizations. We propose a conceptual framework that models human interactions with ML components in the VA process, and that puts the central relationship between automated algorithms and interactive visualizations into sharp focus. The framework is illustrated with several examples and we further elaborate on the interactive ML process by identifying key scenarios where ML methods are combined with human feedback through interactive visualization. We derive five open research challenges at the intersection of ML and visualization research, whose solution should lead to more effective data analysis.