Source-linked AI summary
Human-Data Interaction: The Human Face of the Data-Driven Society
Richard Mortier, Hamed Haddadi, Tristan Henderson, Derek McAuley, Jon Crowcroft
TL;DR
The paper addresses how people can interact with personal data within a complex ecosystem of collection, analysis, and use. It proposes Human-Data Interaction as a distinct, multidisciplinary topic organised around legibility, agency, and negotiability, and concludes that keeping people central is important in a data-driven society.
Problem
Increasing personal-data generation and collection has created complex interactions among individuals, companies, and data-processing systems that existing interaction research does not fully address.
Method
The paper develops a broad conception of HDI and organises its challenges through the themes of legibility, agency, and negotiability.
Results
HDI is presented as a distinct topic requiring expertise across computer science, psychology, economics, and law, with people remaining the first consideration of a data-driven society.
Takeaways & Limitations
A coherent HDI approach should provide insight, tools, and techniques for managing human interaction with data and data processing, helping build trust in large- and small-scale data systems.
Takeaways & Limitations
Visualisation alone may not resolve legibility because data can be large or ambiguous and commercial processing details may be difficult to expose.
Abstract
from arXiv · showhide
The increasing generation and collection of personal data has created a complex ecosystem, often collaborative but sometimes combative, around companies and individuals engaging in the use of these data. We propose that the interactions between these agents warrants a new topic of study: Human-Data Interaction (HDI). In this paper we discuss how HDI sits at the intersection of various disciplines, including computer science, statistics, sociology, psychology and behavioural economics. We expose the challenges that HDI raises, organised into three core themes of legibility, agency and negotiability, and we present the HDI agenda to open up a dialogue amongst interested parties in the personal and big data ecosystems.
1. INTRODUCTION
As computing becomes ubiquitous, interactions increasingly involve data as well as devices. The paper proposes Human-Data Interaction (HDI) to place people at the centre of these data flows and structure related challenges.
- Ubiquitous networked computing has enabled both new interactions and new forms of existing interactions.Examples include online banking and massively scalable distributed real-time gaming.
- Traditional HCI research has focused primarily on interactions between people and computers as artefacts.The field’s focus has expanded from operators and hardware toward software, interfaces, psychology, and organisations.
- The authors propose HDI because ubiquitous computing is raising the question of how people should interact with data.They describe a complex ecosystem around companies and individuals using data, and propose placing humans at its centre.
- The paper broadens prior uses of HDI and surveys influences from computer systems, law, sociology, and economics.It presents contributions on HDI’s conception and its challenges.
- The challenges raised by HDI are organised into three core themes: legibility, agency, and negotiability.The paper aims to open dialogue among parties interested in making humans explicit in the data ecosystem.
2. DEFINING HUMAN-DATA INTERACTION
HDI addresses how personal data move between individuals, organisations, analytics systems, and social or legal contexts. The paper defines a broad conception encompassing understanding, action, and changing relationships around data.
- The Evolution of Data: Personal data arise through volunteered, observed, and inferred processes, while organisations accumulate and analyse them to infer sensitive features.These data and inferences increasingly affect everyday life, motivating HDI.
- The Evolution of Data: HDI concerns the feedback loop in which individual data trails are coalesced into big data and analytical results are fed back into individual data.Personal data therefore function as a boundary object across communities with different understandings of data.
- Embracing Human-Data Interaction: The paper broadens HDI beyond direct dataset interaction to include many forms of interaction around personal and open data.These include permitting or denying third-party access and receiving information about data use.
- Embracing Human-Data Interaction: HDI also includes interaction with systems that process data, including understanding possible inferences and consequences of making personal data available.Data are treated as dynamic, subject to revision and extension.
- Embracing Human-Data Interaction: The framework organises HDI around legibility, agency, and negotiability.Legibility concerns transparency and comprehensibility; agency concerns control and correction; negotiability concerns changing relationships, norms, jurisdictions, and meanings.
3. LEGIBILITY
Legibility requires more than making data systems transparent: people must be able to understand what data are collected, how analytics work, and what the implications are. This understanding supports conscious agency but is difficult because data and processing are complex, large-scale, and sometimes ambiguous.
- Online data systems are often opaque, and technical complexity can make their implications incomprehensible even when processes are transparent.The paper presents legibility as a precursor to consciously exercising agency over personal-data collection and processing.
- Data created about people through tracking, recommender systems, and data mining are often less understood by their subjects.These processes can generate inferred data such as advertising preferences.
- People generate increasingly rich and varied data through social networks, websites, lifelogging, and quantified-self technologies.These sources include sensors and technologies with which individuals explicitly interact.
- Legibility requires awareness of collection, awareness of the data themselves, and understanding their implications and correctness.The second requirement is more complex than simply notifying people that collection occurs.
- Visualisation is a starting point, but large personal datasets, ambiguous community data, and commercial algorithms create further legibility challenges.Visualisations may expose incentive models and algorithmic details that are problematic in commercial environments.
- Artists and embodied interactions are presented as one possible avenue for making data, algorithms, and inferences legible to users.The paper notes promising early attempts, including Tangible Souvenirs and Sweat Atoms.
4. AGENCY
Agency in Human-Data Interaction means giving people the capacity to act within systems that collect, store, process, and infer from their data. The paper argues this requires more than awareness or consent, including control over data and inferences.
- Agency: Agency requires people to act within systems that collect and process their personal data, not merely become aware of those systems.The paper distinguishes awareness of collection from the capacity to act for oneself within data-processing systems.
- Agency: People need broader abilities to engage with collection, storage, and use, including understanding and modifying raw data and derived inferences.
- Agency: User-centric controls should support consent and revocation because collected data and personal-data inferences can be biased or wrong.The paper attributes errors to contextual, temporal, and sampling biases, misunderstood semantics, flawed algorithms, incomplete data, and changing attitudes or preferences.
- Agency: A survey of 1,464 UK consumers found that 94% believed they should control information collected about them.
- Agency: Users may not often need or desire continuous engagement with data systems, despite the importance of retaining the capacity to act within them.
- Agency: Privacy policies are difficult to design and interpret, while effects of data collection may span multiple entities and time periods.These difficulties make broader user support and accurate measurement of privacy effects significant challenges.
5. NEGOTIABILITY
Negotiability concerns helping people reassess data-related decisions as contexts change, including across jurisdictions and through feedback and control mechanisms. The paper links this need to unequal power, contextual effects, and risks arising from open-data reuse and future datasets.
- Negotiability: Negotiability supports re-evaluating decisions as external and internal contexts change, including across jurisdictions and through feedback and control mechanisms.
- Negotiability: The paper argues that power currently favors data aggregators, challenging models that treat personal data as a tradable good from which economic value is extracted.
- Negotiability: Contextual integrity research is needed because data connected with people are not neutral or value-free across services, research, and business uses.
- Negotiability: Personal-data experiments require careful consideration of data types and appropriate consent, while research-data sharing is becoming popular and sometimes mandated.
- Negotiability: Publishing open data can produce usage, correlation, reputation, and re-identification effects, so HDI must account for current and future datasets.
- Negotiability: Reasoning about data is shaped by cultural and contextual differences, which in turn inform how communities use, release, and distribute personal data.
6. CONCLUSIONS
The paper presents HDI as a distinct, multidisciplinary topic organized around legibility, agency, and negotiability. It argues that coherent tools and ethical systems are needed to keep people central and build trust in large-scale data processing.
- Conclusions: HDI is presented as a distinct topic organized around legibility, agency, and negotiability in an evolving data-driven society.
- Conclusions: HDI draws on computer science, psychology, economics, and law rather than belonging exclusively to one discipline.
- Conclusions: A coherent HDI approach providing insight, tools, and techniques for managing interaction with data is described as a prerequisite for building trust in data processing.
- Conclusions: Technology designers are urged to build ethical systems that provide agency in intentional action and make involuntary-behaviour outcomes predictable for individuals and groups.