Jump to Content
Lora Aroyo

Lora Aroyo

I am a research scientist at Google Research NYC where I work on Data Excellence for AI. My team DEER (Data Excellence for Evaluating Responsibly) is part of the Responsible AI (RAI) organization. Our work is focused on developing metrics and methodologies to measure the quality of human-labeled or machine-generated data. The specific scope of this work is for gathering and evaluation of adversarial data for Safety evaluation of Generative AI systems. I received MSc in Computer Science from Sofia University, Bulgaria, and PhD from Twente University, The Netherlands.

I am currently serving as a co-chair of the steering committee for the AAAI HCOMP conference series and I am a founding member of the DataPerf and the AI Safety Benchmarking working group both at MLCommons for benchmarking data-centric AI. Check out our data-centric challenge Adversarial Nibbler supported by Kaggle, Hugging Face and MLCommons. In 2023 I gave the opening keynote at NeurIPS Conference "The Many Faces of Responsible AI".

Prior to joining Google, I was a computer science professor heading the User-Centric Data Science research group at the VU University Amsterdam. Our team invented the CrowdTruth crowdsourcing method jointly with the Watson team at IBM. This method has been applied in various domains such as digital humanities, medical and online multimedia. I also guided the human-in-the-loop strategies as a Chief Scientist at a NY-based startup Tagasauris.

Some of my prior community contributions include president of the User Modeling Society, program co-chair of The Web Conference 2023, member of the ACM SIGCHI conferences board.

For a list of my publications, please see my profile on Google Scholar.

Authored Publications
Google Publications
Other Publications
Sort By
  • Title
  • Title, desc
  • Year
  • Year, desc
    Preview abstract With recent improvements in natural language generation (NLG) models for various applications, it has become imperative to have the means to identify and evaluate whether NLG output is only sharing verifiable information about the external world. In this work, we present a new evaluation framework entitled Attributable to Identified Sources (AIS) for assessing the output of natural language generation models, when such output pertains to the external world. We first define AIS and introduce a two-stage annotation pipeline for allowing annotators to appropriately evaluate model output according to AIS guidelines. We empirically validate this approach on generation datasets spanning three tasks (two conversational QA datasets, a summarization dataset, and a table-to-text dataset) via human evaluation studies that suggest that AIS could serve as a common framework for measuring whether model-generated statements are supported by underlying sources. We release guidelines for the human evaluation studies. View details
    Preview abstract Adversarial testing of large language models (LLMs) is crucial for their safe and responsible deployment. We introduce a novel approach for automated generation of adversarial evaluation datasets to test the safety of LLM generations on new downstream applications. We call it AI-assisted Red-Teaming (AART) - an automated alternative to current manual red-teaming efforts. AART offers a data generation and augmentation pipeline of reusable and customizable recipes that reduce human effort significantly and enable integration of adversarial testing earlier in new product development. AART generates evaluation datasets with high diversity of content characteristics critical for effective adversarial testing (e.g. sensitive and harmful concepts, specific to a wide range of cultural and geographic regions and application scenarios). The data generation is steered by AI-assisted recipes to define, scope and prioritize diversity within the application context. This feeds into a structured LLM-generation process that scales up evaluation priorities. Compared to some state-of-the-art tools, AART shows promising results in terms of concept coverage and data quality. View details
    Preview abstract Machine learning approaches often require training and evaluation datasets with a clear separation between positive and negative examples. This risks simplifying and even obscuring the inherent subjectivity present in many tasks. Preserving such variance in content and diversity in datasets is often expensive and laborious. This is especially troubling when building safety datasets for conversational AI systems, as safety is both socially and culturally situated. To demonstrate this crucial aspect of conversational AI safety, and to facilitate in-depth model performance analyses, we introduce the DICES (Diversity In Conversational AI Evaluation for Safety) dataset that contains fine-grained demographic information about raters, high replication of ratings per item to ensure statistical power for analyses, and encodes rater votes as distributions across different demographics to allow for in￾depth explorations of different aggregation strategies. In short, the DICES dataset enables the observation and measurement of variance, ambiguity, and diversity in the context of conversational AI safety. We also illustrate how the dataset offers a basis for establishing metrics to show how raters’ ratings can intersects with demographic categories such as racial/ethnic groups, age groups, and genders. The goal of DICES is to be used as a shared resource and benchmark that respects diverse perspectives during safety evaluation of conversational AI systems. View details
    Preview abstract Conventional machine learning paradigms often rely on binary distinctions between positive and negative examples, disregarding the nuanced subjectivity that permeates real-world tasks and content. This simplistic dichotomy has served us well so far, but because it obscures the inherent diversity in human perspectives and opinions, as well as the inherent ambiguity of content and tasks, it poses limitations on model performance aligned with real-world expectations. This becomes even more critical when we study the impact and potential multifaceted risks associated with the adoption of emerging generative AI capabilities across different cultures and geographies. To address this, we argue that to achieve robust and responsible AI systems we need to shift our focus away from a single point of truth and weave in a diversity of perspectives in the data used by AI systems to ensure the trust, safety and reliability of model outputs. In this talk, I present a number of data-centric use cases that illustrate the inherent ambiguity of content and natural diversity of human perspectives that cause unavoidable disagreement that needs to be treated as signal and not noise. This leads to a call for action to establish culturally-aware and society-centered research on impacts of data quality and data diversity for the purposes of training and evaluating ML models and fostering responsible AI deployment in diverse sociocultural contexts. View details
    Preview abstract We tackle the problem of providing accurate, rigorous p-values for comparisons between the results of two evaluated systems whose evaluations are based on a crowdsourced “gold” reference standard. While this problem has been studied before, we argue that the null hypotheses used in previous work have been based on a common fallacy of equality of probabilities, as opposed to the standard null hypothesis that two sets are drawn from the same distribution. We propose using the standard null hypothesis, that two systems’ responses are drawn from the same distribution, and introduce a simulation-based framework for determining the true p-value for this null hypothesis. We explore how to estimate the true p-value from a single test set under different metrics, tests, and sampling methods, and call particular attention to the role of response variance, which exists in crowdsourced annotations as a product of genuine disagreement, and in system predictions as a product of stochastic training regimes, or in generative models as an expected property of the outputs. We find that response variance is a powerful tool for estimating p-values, and present results for the metrics, tests, and sampling methods that make the best p-value estimates in a simple machine learning model comparison View details
    Adversarial Nibbler: A DataPerf Challenge for Text-to-Image Models
    Hannah Kirk
    Jessica Quaye
    Charvi Rastogi
    Max Bartolo
    Oana Inel
    Meg Risdal
    Will Cukierski
    Vijay Reddy
    Online (2023)
    Preview abstract Machine learning progress has been strongly influenced by the data used for model training and evaluation. Only recently however, have development teams shifted their focus more to the data. This shift has been triggered by the numerous reports about biases and errors discovered in AI datasets. Thus, the data-centric AI movement introduced the notion of iterating on the data used in AI systems, as opposed to the traditional model-centric AI approach, which typically treats the data as a given static artifact in model development. With the recent advancement of generative AI, the role of data is even more crucial for successfully developing more factual and safe models. DataPerf challenges follow up on recent successful data- centric challenges drawing attention to the data used for training and evaluation of machine learning model. Specifically, Adversarial Nibbler focuses on data used for safety evaluation of generative text-to-image models. A typical bottleneck in safety evaluation is achieving a representative diversity and coverage of different types of examples in the evaluation set. Our competition aims to gather a wide range of long-tail and unexpected failure modes for text-to-image models in order to identify as many new problems as possible and use various automated approaches to expand the dataset to be useful for training, fine tuning, and evaluation. View details
    Preview abstract Dialogue safety as a task is complex, in part because ‘safety’ entails a broad range of topics and concerns, such as toxicity, harm, legal concerns, health advice, etc. Who we ask to judge safety and who we ask to define safety may lead to differing conclusions. This is because definitions and understandings of safety can vary according to one’s identity, public opinion, and the interpretation of existing laws and regulations. In this study, we compare annotations from a diverse set of over 100 crowd raters to gold labels derived from trust and safety (T&S) experts in a dialogue safety task consisting of 350 human-chatbot conversations. We find patterns of disagreements rooted in dialogue structure, dialogue content, and rating rationale. In contrast to typical approaches which treat gold labels as ground truth, we propose alternative ways of interpreting gold data and incorporating crowd disagreement rather than mitigating it. We discuss the complexity of safety annotation as a task, what crowd and T&S labels each uniquely capture, and how to make determinations about when and how to rely on crowd or T&S labels. View details
    Preview abstract Chatbots based on large language models (LLM) exhibit a level of human-like behavior that promises to have profound impacts on how people access information, create content, and seek social support. Yet these models have also shown a propensity toward biases and hallucinations, i.e., make up entirely false information and convey it as truthful. Consequently, understanding and moderating safety risks in these models is a critical technical and social challenge. We use Bayesian multilevel models to explore the connection between rater demographics and their perception of safety in chatbot dialogues. We study a sample of 252 human raters stratified by gender, age, race/ethnicity, and location. Raters were asked to annotate the safety risks of 1,340 chatbot conversations. We show that raters from certain demographic groups are more likely to report safety risks than raters from other groups. We discuss the implications of these differences in safety perception and suggest measures to ameliorate these differences. View details
    LaMDA: Language Models for Dialog Applications
    Aaron Daniel Cohen
    Alena Butryna
    Alicia Jin
    Apoorv Kulshreshtha
    Ben Zevenbergen
    Chung-ching Chang
    Cosmo Du
    Daniel De Freitas Adiwardana
    Dehao Chen
    Dmitry (Dima) Lepikhin
    Erin Hoffman-John
    Igor Krivokon
    James Qin
    Jamie Hall
    Joe Fenton
    Johnny Soraker
    Maarten Paul Bosma
    Marc Joseph Pickett
    Marcelo Amorim Menegali
    Marian Croak
    Maxim Krikun
    Noam Shazeer
    Rachel Bernstein
    Ravi Rajakumar
    Ray Kurzweil
    Romal Thoppilan
    Steven Zheng
    Taylor Bos
    Toju Duke
    Tulsee Doshi
    Vincent Y. Zhao
    Will Rusch
    Yuanzhong Xu
    arXiv (2022)
    Preview abstract We present LaMDA: Language Models for Dialog Applications. LaMDA is a family of Transformer-based neural language models specialized for dialog, which have up to 137B parameters and arepre-trained on 1.56T words of public dialog data and web text. While model scaling alone canimprove quality, it shows less improvements on safety and factual grounding. We demonstrate thatfine-tuning with annotated data and enabling the model to consult external knowledge sources canlead to significant improvements towards the two key challenges of safety and factual grounding.The first challenge, safety, involves ensuring that the model’s responses are consistent with a set ofhuman values, such as preventing harmful suggestions and unfair bias. We quantify safety using ametric based on an illustrative set of values, and we find that filtering candidate responses using aLaMDA classifier fine-tuned with a small amount of crowdworker-annotated data offers a promisingapproach to improving model safety. The second challenge, factual grounding, involves enabling themodel to consult external knowledge sources, such as an information retrieval system, a languagetranslator, and a calculator. We quantify factuality using a groundedness metric, and we find that ourapproach enables the model to generate responses grounded in known sources, rather than responsesthat merely sound plausible. Finally, we explore the use of LaMDA in the domains of education andcontent recommendations, and analyze their helpfulness and role consistency. View details
    Preview abstract Successful knowledge graphs (KGs) solved the historical knowledge acquisition bottleneck by supplanting an expert focus with a simple, crowd-friendly one: KG nodes represent popular people, places, organizations, etc., and the graph arcs represent common sense relations like affiliations, locations, etc. Techniques for more general, categorical, KG curation do not seem to have made the same transition: the KG research community is still largely focused on logic-based methods that belie the common-sense characteristics of successful KGs. In this paper, we propose a simple yet novel approach to acquiring \emph{class-level attributes} from the crowd that represent broad common sense associations between categories, and can be used with the classic knowledge-base default \& override technique (e.g. \cite{reiter1978}) to address the early \textit{label sparsity problem} faced by machine learning systems for problems that lack data for training. We demonstrate the effectiveness of our acquisition and reasoning approach on a pair of very real industrial-scale problems: how to augment an existing KG of places and offerings (e.g. stores and products, restaurants and dishes) with associations between them indicating the availability of the offerings at those places, which would enable the KG to provide answers to questions like, ``Where can I buy milk nearby?'' This problem has several practical challenges but for this paper we focus mostly on the label sparsity. Less than 30\% of physical places worldwide (i.e. brick \& mortar stores and restaurants) have a website, and less than half of those list their product catalog or menus, leaving a large acquisition gap to be filled by methods other than information extraction (IE). Label sparsity is a general problem, and not specific to these use cases, that prevents modern AI and machine learning techniques from applying to many applications for which labeled data is not readily available. As a result, the study of how to acquire the knowledge and data needed for AI to work is as much a problem today as it was in the 1970s and 80s during the advent of expert systems \cite{mycin1975}. The class-level attributes approach presented here is based on a KG-inspired intuition that a lot of the knowledge people need to understand where to go to buy a product they need, or where to find the dishes they want to eat, is categorical and part of their general common sense: everyone knows grocery stores sell milk and don't sell asphalt, chinese restaurants serve fried rice and not hamburgers, etc. We acquired a mixture of instance- and class- level pairs (e.g. $\langle$\textit{Ajay Mittal Dairy}, milk$\rangle$, $\langle$GroceryStore, milk$\rangle$, resp.) from a novel 3-tier crowdsourcing method, and demonstrate the scalability advantages of the class-level approach. Our results show that crowdsourced class-level knowledge can provide rapid scaling of knowledge acquisition in shopping and dining domains. The acquired common sense knowledge also has long-term value in the KG. The approach was a critical part of enabling a worldwide \textit{local search} capability on Google Maps, with which users can find products and dishes that are available in most places on earth. View details
    Preview abstract To understand what captures people's attention (what they find relevant) we focussed on understanding better the content of videos. In information science, the concept of relevance is most connected to end-users' judgments and is considered fundamental as a subjective, dynamic user-centric perception. People might have or use different relevance standards or criteria when performing the task of video searching. Textual and visual criteria are essential for identifying relevant video content, but subjective, implicit criteria, such as interest or familiarity could be equally used by people. Typically, people tend to connect bridges to concepts or perspectives that are not necessarily shown in the video, but that might be expressed or referred to. We carried out a number of studies with news videos and broadcasts. In our initial study [6], we took a digital hermeneutics approach to understand which video aspects capture the attention of digital humanities scholars and drive the creation of narratives, or short audio-visual stories. In subsequent studies, we focused on understanding the utility of machine-extracted video concepts and how people can teach machines in terms of video concept relevance. We harnessed the intrinsic subjectivity of concept relevance to capture the diverse range of video concepts found relevant through the eyes of our participants [4]. We explored to what extent current information extraction systems meet users' goals, and what are the novel aspects users bring in video concept relevance assessment. We performed two types of crowdsourcing studies. The Selection study (Figure 1) focused on understanding the utility of machine-extracted video concepts from video subtitles and video streams. While the Free Input study (Figure 2) focused on understanding the complementarity between machine and human concepts in terms of relevance. By studying the gap between machines and humans in terms of perceived video concept relevance, we gained insights into how machines can collaborate with users, to better support their needs and preferences. Our studies revealed that events, locations, people, organizations, and general concepts (i.e., of any type) are fundamental elements for content exploration and understanding. They are most commonly machine concepts extracted and as such used in machine summarization of content as well as for information search. However, people engaging with online videos most often provide events, people, locations, and organizations as relevant concepts. Concepts of other types are also found relevant, but to a lesser extent. These concept types are thus fundamental for contextualizing the content of the videos, and also sufficient to capture human interest in terms of relevance. View details
    Annotator Response Distributions as a Sampling Frame
    Christopher Homan
    LREC WOrkshop on Perspectivist NLP (2022)
    Preview abstract Annotator disagreement is often dismissed as noise or the result of poor annotation process quality. Others have argued that it can be meaningful. But lacking a rigorous statistical foundation, the analysis of disagreement patterns can resemble a high-tech form of tea-leaf-reading. We contribute a framework for analyzing the variation of per-item annotator response distributions to data for humans-in-the-loop machine learning. We provide visualizations for, and use the framework to analyze the variance in, a crowdsourced dataset of hard-to-classify examples of the OpenImages archive. View details
    Data Excellence for AI: Why Should You Care
    Matt Lease
    Praveen Kumar Paritosh
    ACM IX Interactions (2022)
    Preview abstract The efficacy of machine learning (ML) models depends on both algorithms and data. Training data defines what we want our models to learn, and testing data provides the means by which empirical progress is measured. Benchmark datasets such as SQuAD, GLUE, and ImageNet define the entire world within which models exist and operate, yet research continues to focus on critiquing and improving the models, e.g., via shared-task challenges or Kaggle contests, rather than critiquing and improving the data environment in which our models operate. Research and community challenges focused on improving the data itself are relatively rare. If “data is the new oil,” our use of data remains crude today, and we are missing work on the refineries by which the data itself could be optimized for more effective use. Important scientific opportunities and value are being neglected [Schaekermann et al., 2020]. Data is potentially the most under-valued and de-glamorised aspect of today’s AI ecosystem. Data issues are often perceived and characterized as mundane and rote, the “pre-processing” that has to be done before the real (modeling) work can be done. For example, Kandel et al. (2012) emphasize that ML practitioners view data wrangling as tedious and time-consuming. However, Sambasivan et al. (2021) provide examples of how data quality is crucial to ensure that AI systems can accurately represent and predict the phenomenon it is claiming to measure. They introduce four classes of Data Cascades: compounding events causing negative, downstream effects from data issues triggered by conventional AI/ML practices that undervalue data quality. This emphasizes the significance of data due to its downstream impact on user wellbeing and societal effects. Real-world datasets are often ‘dirty’, with various data quality problems (Northcutt et al, 2021), with the risk of “garbage in = garbage out” in terms of the downstream AI systems we train and test on such data. This has inspired a steadily growing body of work on understanding and improving data quality (Chu, et al, 2013; Krishnan, et al, 2016; Redman, et al, 2018; Raman et al, 2001). It also highlights the importance of rigorously managing data quality using mechanisms specific to data validation, instead of relying on model performance as a proxy for data quality (Thomas, et al, 2020). Just as we rigorously test our code for software defects before deployment, we might test for data defects with the same degree of rigor, so that we might detect, prevent, or mitigate weaknesses in ML models caused by underlying issues in data quality. The “Crowdsourcing Adverse Test Sets for Machine Learning (CATS4ML)” Data Challenge (Aroyo and Paritosh, 2021) aims to raise the bar in ML evaluation sets and to find as many examples as possible that are confusing or otherwise problematic for algorithms to process. Similarly to (Vandenhof, 2019) CATS4ML relies on people’s abilities and intuition to spot new data examples about which machine learning is confident, but actually misclassified. This research is inspired by (Attenberg et al, 2015) following the claim “Humans should always be part of machine learning solutions, as they can guide machine learning systems to learn about things that the systems don't yet know — the “unknown unknowns.”” by Iperiotis, (2016). Many benchmark datasets contain instances that are relatively easy (e.g., photos with a subject that is easy to identify). In so doing, they miss the natural ambiguity of the real world in which our models are to be actually applied. Data instances with annotator disagreement are often aggregated to eliminate disagreement (obscuring uncertainty), or filtered out of datasets entirely. Exclusion of difficult and/or ambiguous real-world examples in evaluation risks “toy dataset” benchmarks that diverge from the real data to be encountered in practice. Successful benchmark models fail to generalize to real data, and inflated benchmark results mislead our assessment of state-of-the-art capabilities. ML models become prone to develop “weak spots”, i.e., classes of examples that are difficult or impossible for a model to accurately evaluate, because that class of examples is missing from the evaluation set. Measuring data quality is challenging, nebulous, and often circularly defined, with annotated data defining the “ground truth” on which models are trained and tested [Riezler, 2014]. When dataset quality is considered, the ways in which it is measured in practice is often poorly understood and sometimes simply wrong. Challenges identified include fairness and bias issues in labeled datasets [Goel and Faltings, 2019], quality issues in datasets [Crawford and Paglen, 2019], limitations of benchmarks [Kovaleva et al., 2019, Welty et al., 2019] reproducibility concerns in machine learning research [Pineau et al., 2018, Gunderson and Kjensmo, 2018], lack of documentation and replication of data [Katsuno et al., 2019]. Measurement of AI success and progress today is often metrics-driven, with emphasis on rigorous measurement and A/B testing. However, measuring goodness of the fit of the model to the dataset completely ignores any consideration of how well the dataset fits the real world problem to be solved and its data. Goodness-of-fit metrics, such as F1, Accuracy, AUC, do not tell us much about data fidelity (i.e., how well the dataset represents reality) and validity (how well the data explains things related to the phenomena captured by the data). No standardised metrics exist today for characterising the goodness-of-data [11,13]. Research on metrics is emerging [15,91] but is not yet widely known, accepted, or applied in the AI ecosystem today. As a result, there is an overreliance on goodness-of-fit metrics and post-deployment product metrics. Focusing on fidelity and validity of data will further increase its scientific value and reusability. Such research is necessary for enabling better incentives for data, as it is hard to improve something we can not measure. Researchers in human computation (HCOMP) and various ML-related fields have demonstrated a longstanding interest in applying crowdsourcing approaches to generate human-annotated data for model training and testing [25,128]. A series of workshops (Meta-Eval 2020 @ AAAI, REAIS 2019 @ HCOMP, SAD 2019 @ TheWebConf (WWW), SAD 2018 @ HCOMP) have helped increase further awareness about the issues of data quality for ML evaluation and provide a venue for scholarship on this subject. Because human annotated data represents the compass that the entire ML community relies on, data-focused research, by the HCOMP community and others, can potentially have a multiplicative effect on accelerating progress in ML more broadly. Optimizing the cost, size, and speed of collecting data has attracted significant attention in the first-to-market rush with data. However, aspects of maintainability, reliability, validity, and fidelity of datasets have been often overlooked. We argue we have now reached an inflection point in the field of ML in which attention to neglected data quality is poised to significantly accelerate progress. Toward this end, we advocate for research defining and creating processes to achieve data excellence. We highlight examples, case-studies, and methodologies. This will enable the necessary change in our research culture to value excellence in data practices, which is a critical milestone on the road to enabling the next generation of breakthroughs in ML and AI. View details
    The Reasonable Effectiveness of Diverse Evaluation Data
    Christopher Homan
    Alex Taylor
    Human Evaluation for Generative Models (HEGM) Workshop at NeurIPS2022
    Preview abstract In this paper, we present findings from an semi-experimental exploration of rater diversity and its influence on safety annotations of conversations generated by humans talking to a generative AI-chat bot. We find significant differences in judgments produced by raters from different geographic regions and annotation platforms, and correlate these perspectives with demographic sub-groups. Our work helps define best practices in model development-- specifically human evaluation of generative models-- on the backdrop of growing work on sociotechnical AI evaluations. View details
    Preview abstract Successful knowledge graphs (KGs) solved the historical knowledge acquisition bottleneck by supplanting an expert focus with a simple, crowd-friendly one: KG nodes represent popular people, places, organizations, etc., and the graph arcs represent common sense relations like affiliations, locations, etc. Techniques for more general, categorical, KG curation do not seem to have made the same transition: the KG research community is still largely focused on methods that belie the common-sense characteristics of successful KGs. In this paper, we propose a simple approach to acquiring and reasoning with class-level attributes from the crowd that represent broad common sense associations between categories. We pick a very real industrial-scale data set and problem: how to augment an existing knowledge graph of places and products with associations between them indicating the availability of the products at those places, which would enable a KG to provide answers to questions like, ``Where can I buy milk nearby?'' This problem has several practical challenges, not least of which is that only 30\% of physical stores (i.e. brick \& mortar stores) have a website, and fewer list their product inventory, leaving a large acquisition gap to be filled by methods other than information extraction (IE). Based on a KG-inspired intuition that a lot of the class-level pairs are part of people's general common sense, e.g. everyone knows grocery stores sell milk and don't sell asphalt, we acquired a mixture of instance- and class- level pairs (e.g. {Ajay Mittal Dairy, milk}, {GroceryStore, milk}, resp.) from a novel 3-tier crowdsourcing method, and demonstrate the scalability advantages of the class-level approach. Our results show that crowdsourced class-level knowledge can provide rapid scaling of knowledge acquisition in this and similar domains, as well as long-term value in the KG. View details
    "Everyone wants to do the model work, not the data work": Data Cascades in High-Stakes AI
    Nithya Sambasivan
    Shivani Kapania
    Hannah Highfill
    Praveen Kumar Paritosh
    SIGCHI, ACM (2021)
    Preview abstract AI models are increasingly applied in high-stakes domains like health and conservation. Data quality carries an elevated significance in high-stakes AI due to its heightened downstream impact, impacting predictions like cancer detection, wildlife poaching, and loan allocations. Paradoxically, data is the most under-valued and de-glamorised aspect of AI. In this paper, we report on data practices in high-stakes AI, from interviews with 53 AI practitioners in India, East and West African countries, and USA. We define, identify, and present empirical evidence on Data Cascades---compounding events causing negative, downstream effects from data issues---triggered by conventional AI/ML practices that undervalue data quality. Data cascades are pervasive (92% prevalence), invisible, delayed, but often avoidable. We discuss HCI opportunities in designing and incentivizing data excellence as a first-class citizen of AI, resulting in safer and more robust systems for all. View details
    Adversarial Test Set for Image Classification: Lessons Learned from CATS4ML Data Challenge
    Praveen Kumar Paritosh
    NeurIPS 2021 Datasets and Benchmarks Track, NeurIPS (2021)
    Preview abstract A primary role of data in ML is to serve as benchmarks that allow us to measure progress. Often, items that are difficult and have natural ambiguity of real world context are relatively underrepresented in evaluation datasets and benchmarks. This absence of ambiguous real-world examples in evaluation undermines the ability to reliably test machine learning performance. This results in unknown unknowns of an ML model’s behaviour, which is a large risk in their deployment. We designed and ran a public data challenge to proactively discover unknown unknowns in state-of-the-art image classification models applied to the Open Images v6 dataset. In this paper, we describe the design and implementation of the AAAI HCOMP CATS4ML 2020 challenge. Participants in this challenge were incentivized to find images that are incorrectly classified by the ML models. We present a set of failure modes in the state-of-art image classification abstracted from the 13,000 submissions from this challenge. We present a black-swan benchmark test set based on this challenge. View details
    Preview abstract When collecting annotations and labeled data from humans, a standard practice is to use inter-rater reliability (IRR) as a measure of data goodness (Hallgren, 2012). Metrics such as Krippendorff’s alpha or Cohen’s kappa are typically required to be above a threshold of 0.6 (Landis and Koch, 1977). These absolute thresholds are unreasonable for crowdsourced data from annotators with high cultural and training variances, especially on subjective topics. We present a new alternative to interpreting IRR that is more empirical and contextualized. It is based upon benchmarking IRR against baseline measures in a replication, one of which is a novel cross-replication reliability (xRR) measure based on Cohen’s (1960) kappa. We call this approach the xRR framework. We opensource a replication dataset of 4 million human judgements of facial expressions and analyze it with the proposed framework. We argue this framework can be used to measure the quality of crowdsourced datasets. View details
    Empirical methodology for crowdsourcing ground truth
    Anca Dumitrache
    Benjamin Timmermans
    Oana Inel
    Semantic Web Journal, vol. 12:3; 2021 (2021)
    Preview abstract The process of gathering ground truth data through human annotation is a major bottleneck in the use of information extraction methods for populating the Semantic Web. Crowdsourcing-based approaches are gaining popularity in the attempt to solve the issues related to volume of data and lack of annotators. Typically these practices use inter-annotator agreement as a measure of quality. However, in many domains, such as event detection, there is ambiguity in the data, as well as a multitude of perspectives of the information examples. We present an empirically derived methodology for efficiently gathering of ground truth data in a diverse set of use cases covering a variety of domains and annotation tasks. Central to our approach is the use of CrowdTruth metrics that capture inter-annotator disagreement. We show that measuring disagreement is essential for acquiring a high quality ground truth. We achieve this by comparing the quality of the data aggregated with CrowdTruth metrics with majority vote, over a set of diverse crowdsourcing tasks: Medical Relation Extraction, Twitter Event Identification, News Event Extraction and Sound Interpretation. We also show that an increased number of crowd workers leads to growth and stabilization in the quality of annotations, going against the usual practice of employing a small number of annotators. View details
    Preview abstract Human annotated data plays a crucial role in the current ML/AI climate, where the human judgements are referenced as the ultimate source of truth. As such, human annotated data is a kind of compass for AI and research on Human Computation has a multiplicative effect on the AI field. Optimizing the cost, scale, and speed of data collection has been the center of the Human Computation research. What is less known is that such optimization is sometimes done at the cost of quality [Riezler, 2014]. Quality is evidently important but unfortunately poorly defined and rarely measured. A decade later, problems inherent to data are beginning to surface: fairness and bias [Goel and Faltings, 2019], quality issues [Crawford and Paglen, 2019], limitations of benchmarks [Kovaleva et al., 2019, Welty et al., 2019] reproducibility in ML research [Pineau et al., 2018, Gunderson and Kjensmo, 2018], lack of documentation [Katsuno et al., 2019]. Finally, in rushing to be first to market, aspects of data quality such as maintainability, reliability, validity, and fidelity are often overlooked. We want to turn this way of thinking on its head and highlight examples, case-studies, and methodologies for excellence in data collection. Currently, to the extent that it does, data excellence happens organically by virtue of individual expertise, diligence, commitment, pride, etc. This could be dangerous as we grow increasingly dependent on automation technologies, we don't want to be at the mercy of individual exemplars. Instead, we should codify data excellence in a systematic manner and raise the standards on the entire industry. View details
    Eliciting User Preferences for Personalized Explanations for Video Summaries
    Nava Tintarev
    Oana Inel
    ACM UMAP2020 Conference, ACM (2020) (to appear)
    Preview abstract Video summaries or highlights are a compelling alternative for exploring and contextualizing unprecedented amounts of video material. However, the summarization process is commonly automatic, non-transparent and potentially biased towards particular aspects depicted in the original video. Therefore, our aim is to help users like archivists or collection managers to quickly understand which summaries are the most representative for an original video.In this paper, we present empirical results on the utility of different types of visual explanations to achieve transparency for end users on how representative video summaries are, with respect to the original video. We consider four types of video summary explanations, which use in different ways the concepts extracted from the original video subtitles and the video stream, and their prominence.The explanations are generated to meet target user preferences and express different dimensions of transparency:concept prominence, semantic coverage, distance and quantity of coverage. In two user studies we evaluate the utility of the visual explanations for achieving transparency for end users. Our results show that explanations representing all of the dimensions have the highest utility for transparency. View details
    A Metrological Framework for Evaluating Crowd-powered Instruments
    Praveen Kumar Paritosh
    HCOMP-2019: AAAI Conference on Human Computation
    Preview abstract In this paper we present the first steps towards hardening the science of measuring AI systems, by adopting metrology, the science of measurement and its application, and applying it to human (crowd) powered evaluations. We begin with the intuitive observation that evaluating the performance of an AI system is a form of measurement. In all other science and engineering disciplines, the devices used to measure are called instruments, and all measurements are recorded with respect to the characteristics of the instruments used. One does not report mass, speed, or length, for example, of a studied object without disclosing the precision (measurement variance) and resolution (smallest detectable change) of the instrument used. It is extremely common in the AI literature to compare the performance of two systems by using a crowd-sourced dataset as an instrument, but failing to report if the performance difference lies within the capability of that instrument to measure. To further the discussion, we focus on a single crowd-sourced dataset, the so-called WS-353, a venerable and often used gold standard for word similarity, and propose a set of metrological characteristics for it {\it as an instrument}. We then analyze several previously published experiments that use the WS-353 instrument, and show that, in the light of these proposed characteristics, the differences in performance of these systems cannot be measured with this instrument. View details
    Preview abstract We present a resource for the task of FrameNetsemantic frame disambiguation over 5,000word-sentence pairs from the Wikipedia cor-pus.The annotations were collected usinga novel crowdsourcing approach with mul-tiple workers per sentence to captureinter-annotator disagreement.In contrast to thetypical approach of attributing the best singleframe to each word, we provide a list of frameswith disagreement-based scores that expressthe confidence with which each frame appliesto the word. This is based on the idea thatinter-annotator disagreement is at least partlycaused by ambiguity that is inherent to the textand frames. We have found many exampleswhere the semantics of individual frames over-lap sufficiently to make them acceptable alter-natives for interpreting a sentence. We haveargued that ignoring this ambiguity creates anoverly arbitrary target for training and eval-uating natural language processing systems -if humans cannot agree, why would we ex-pect the correct answer from a machine to beany different? To process this data we alsoutilized an expanded lemma-set provided bythe Framester system, which merges FN withWordNet to enhance coverage. Our datasetincludes annotations of 1,000 sentence-wordpairs whose lemmas are not part of FN. Finallywe present metrics for evaluating frame disam-biguation systems that account for ambiguity. View details
    No Results Found