Virtual Workshop on Human-Centered AI Workshop at NeurIPS
Online on 9 December 2022
Call for Participation
Many institutions, researchers, and thought leaders have recently promoted the notion that AI
systems should be human-centered. Although there is no consensus definition for what
human-centered AI means, it commonly refers to a set of principles that AI systems ought to
be (a) reliable, safe, and trustworthy; (b) relatively free from bias and not promote or
reinforce existing structural inequalities; (c) empowering people by supporting their
creativity and (d) fit-for-purpose for use by specific humans, or groups of humans, to meet
human needs and self-efficacy.
Following a successful workshop in 2021, our 2022 virtual workshop aims to bring people
together across different communities that have a stake in HCAI. Broadly, these communities
encompass researchers and practitioners working across AI, machine learning, and
human-computer interaction (HCI). However, many disconnected sub-communities exist that work
across various. important topics in HCAI, including: human-centered explainable AI (XAI), AI
Fairness, human-centered data science (HCDS), human-centered machine learning, computational
creativity, and human-AI co-creation.
Building on the 2021 workshop, we will explore topics such as:
Processes, principles, and technologies to make AI systems more human-centered.
Experimental design and data collection to strengthen HCAI studies.
Explanations (XAI) that serve the needs of diverse end-users.
Human-AI frameworks for analyzing, designing, and evaluating HCAI systems.
Collaboration and (co-)creativity in HCAI systems.
Emergent questions in ethics and fairness.
Submissions to the workshop may address one or more of the following themes - or other
relevant themes of interest:
Theoretical frameworks, disciplines and disciplinarity. How we approach
AI and data science depends on the "lenses" that we bring, based in theory and in
practice. Through what perspectives do you approach this complex domain?
Experiences and cases with AI systems. Theories suggest studies and
experience reports. Studies and experience reports inform theories. What cases or
experiences of human-AI interactions can you contribute to our inter-disciplinary
knowledge and discussion?
Design frameworks for human initiative and AI initiative. Scholars have
debated the question of who should have initiative or control between human and AI for
over 70 years. What forms of discrete or shared initiative are possible now, and how can
we include these possibilities in our systems?
Experiences and cases with human-AI collaboration. Design frameworks
can inform applications. Experiences with applications can challenge frameworks, or lead
to new frameworks. What cases or experiences of human-AI collaborations can you
contribute to our inter-disciplinary knowledge and discussion?
Fairness and bias. Machine learning-based decision-making systems have
the potential to replicate or even exacerbate social inequeties and discrimination. As a
result, there is a surge of recent work on developing machine learning algorithms with
fairness constraints or guarantees. However, for these tools to have positive real-world
impact, their design and implementation should be informed by a clear understanding of
human behavior and real needs. What is the interplay between algorithmic fairness and
Privacy. In many important machine learning tasks – e.g. those related
to healthcare – there is much to be gained from training on personal information, but we
must take care to respect individuals’ privacy appropriately. In this workshop, we are
particularly interested in understanding specific use cases and considering costs and
benefits to individuals and society of making use of private data.
Transparency, explainability, interpretability, and trust. We are
interested to understand what specific types of explainability or interpretability are
helpful to whom in concrete settings, and in exploring any tradeoffs which are
User research. What do we need to know in order to create or enhance an
AI-based system? Our engineering heritage suggests that we seek user needs and resolve
user pain points. How does our user research for these concepts change with AI systems?
Are there other user research goals that are now possible with more sophisticated AI
resources and implementations?
Accountability. When people engineer (or create) an AI system and its
data, how do we hold them and ourselves accountable for design decisions and outcomes?
Automation of AI. It is tempting to apply AI to AI, in the form of
automated AI. Is this a credible approach? Does human discernment play a role in
creating AI systems? Is this a necessary role?
Evaluation. What are the appropriate measurement concepts and resulting
metrics to assess our AI systems? How do we balance among efficiency, explainability,
understandability, user satisfaction, and user hedonics?
Governance. Consequential machine learning systems impact the lives of
millions of people in areas such as criminal justice, healthcare, education, credit
scoring or hiring. Key concepts in the governance of such systems include algorithmic
discrimination, transparency, veracity, explainability and the preservation of privacy.
What is the role of HCI in relation to the governance of such systems?
Problematizing data. Data initially seem to be simple and ”objective.”
However, a growing body of evidence shows the often-hidden role of humans in shaping the
data in AI. Should we design our systems to strengthen human engagement with data? or to
reduce human impact on data?
Qualitative data in data science. Quantitative data analyses may be
powerful, but often decontextualized and potentially shallow. Qualitative data analyses
may be insightful, but often limited to a narrow sample. How can we combine the
strengths of these two approaches?
Values and ethics of AI. Values and ethics are necessarily entangled
with localized, situated, and culturally-informed human perspectives. What are useful
frameworks for a comparative analysis of values and ethics in AI?
Submissions may address one or more of the following themes - or other relevant themes of
interest - in 1-2 page position papers. All submissions will be reviewed by the workshop’s
Program Committee. Authors of the highest-rated submissions will be invited to participate
in one of the panel discussions with our invited speakers.
NeurIPS do not publish workshop papers. We plan to make a public website for the workshop,
where you will be able to list your accepted contribution in one of the following ways: (1)
title+abstract+authors; (2) the contents of item #1 plus PDF of your submission, hosted
locally; (3) the contents of item #1 plus a link to your PDF at a website of your choosing.