Six Human-Centered Artificial Intelligence Grand Challenges

Figures & data

Figure 1. HCAI grand challenges.

Table 1. Six principles that promote well-being in HCAI

Principle	Questions for designers to consider
HARM AVOIDANCE Above all, designers, engineers, and regulators must aim to prevent harm caused by AI implementations that lack a human-centered approach (see Section 2.2.3).	How might we better understand and assess the well-being impacts of AI algorithms and interfaces? What evaluations might address both immediate and long-term impacts? How might harm avoidance be prioritized over other aims?
TRUST AI users must be able to calibrate their trust effectively, through both high-level certifications and interface-level affordances (see Section 6.1.2).	What approaches to/methods of verification, validation, and certification might suit HCAI? What considerations must be made at each stage of design? What affordances in the interface would enable users to calibrate their trust to a tool’s performance? How might purchasers or users be given accurate ethical assurances regarding the tools they use?
ACCOUNTABILITY The impact of AI on risks, responsibilities, and accountabilities must be explicitly considered and addressed (see section 2.2.4).	How might an AI tool shift the risks, tasks, or responsibilities in a particular domain? How are these made explicit or negotiated by stakeholders? How are those at risk able to reduce their risk? How are those accountable for AI performance supported in bearing that accountability?
AGENCY Users must be able to evaluate, control, and even master complex tools that they do not completely understand (see Section 3.2.1).	How will AI tools be made accessible to a broad range of users without experience or understanding of AI? How might users or those impacted by AI be able to evaluate tools that they may not understand functionally? By what new mechanism might people push back against automations that are not human-centered or do not fully exceed the capabilities of the humans they replace?
USER EXPERIENCE AND AFFECT The peace of users must not be exchanged for the power of new capabilities (see Section 2.1.1).	What designs would serve to minimize frustration, stress, anxiety, or regret resulting from human-AI interactions, particularly in high-stakes, high-risk applications? What strategies are needed to reduce technical complexity?
MULTI-OPTIMIZATION The HCAI design approach must seek a broad understanding of human impacts and become facile at optimizing for multiple human priorities (see Section 2.2.1).	How are benefits maximized and harms minimized across multiple stakeholders? How are AI tools optimized and evaluated across multiple concerns and priorities?

Table 2. Principles and research questions for communication platform neutrality.

Principles of communication platform neutrality	Research questions
A social media platform must not control the information an individual is exposed to via deletion, filtering, manipulation, or modification. Users have an immediate right to the rationale for such changes.	In what ways do existing social media platform mechanisms bias the information provided to consumers? Is there a platform design that is neutral with respect to information consumption? What does it mean for a social media platform to be neutral? How should the neutrality of a social media platform be measured?
Individuals have the right to know their source of information and whether that source is credible.	What social and platform affordances and technologies can be used to create an environment that promotes reliable information origination and provenance? Can blockchain technologies be leveraged for information origination, authenticity, and provenance?
Individuals have the right to know whether the online users they are communicating with are actual humans or bots.	Can bots be accurately and automatically identified? What would be the effect on social media to ban bots? Can bots be used to curate and maintain a fair information environment instead of amplifying misinformation?
Individuals have the right to know if the information they are consuming and the people/bots they are interacting with are part of a campaign that is actively targeting them to change their views or opinions (advertisement, propaganda, etc.).	How can information campaigns and their objectives be automatically identified and tracked? What are the effects of giving the control to the user regarding what campaigns they desire to participate in?
Promote community curation of information moderated by subject matter experts and legitimate institutions.	Given the distributed and complex nature of online social media, how can a fair and unbiased communication environment be achieved that benefits society as a whole? How can misinformation be identified and curtailed? How can truthful information be identified and promoted?
Promote mechanisms to facilitate consensus views on topics of importance.	Are there social media platform mechanisms that are more prompt than others to facilitate a fair and civilized discussion and reasoning in order to achieve consensus in topics of societal importance for which consensus is possible?
Disclose any AI use in the platform, including its type and intent.	What are the unintended effects of algorithmic curation of information for curators and for users? What are the unintended consequences of user engagement maximization algorithms? Are recommender systems the main cause of user polarization and radicalization? If so, how can these be avoided in next generation recommenders?

Figure 2. Summary of concepts for responsible design of AI.

Table 3. Implications of Solove’s privacy questions applied to social robots.

Solove’s area of privacy	Implications
The right to be left alone	Assuming that the occupant is of sound mind, it may be reasonable to also assume that they have a level of consent to the presence of the social robot. However, this is indeed an assumption, and varies with both the level of autonomy the occupant has over their living conditions, and the extent to which they understand the robot’s audio, visual, and sensory functions.
Limit access to the self/shield self from access	This too hinges on questions of autonomy – does the occupant have the ability to halt the robot? Under what circumstances? Can they control what and when the robot will see, hear, or sense? Can they theoretically block the robot from certain access, and can they do so in practice?
Secrecy – the concealment of certain matters from others; control over personal information – the ability to exercise control over information about oneself; Intimacy – control over, or limited access to, one’s intimate relationships or aspects of life	These three issues overlap in multiple ways: first, we see the issues of collection and control over collection. It is not enough to say that consent, even active consent, has been given. Habituation to the presence of the robot degrades sensitivity to its infringement on privacy, not to mention the potential for the robot to enter and exit unexpectedly. The occupant may or may not be aware of the robot’s presence and may or may not be aware of its potential to record data in the moment. Additionally, there is functionally no control over data that the robot has recorded. For example, the robot might record a conversation between the occupant and their doctor, capturing medical information. Depending on the jurisdiction they are in, they may indeed have the right to request that data be deleted. In practice, however, this level of control is exceedingly unlikely to be present, even if the implications of such data collection are understood. Furthermore, there are likely to be a multitude of matters that one might conceal from others, and there is no closed list as to what said topics might be and attempting to compile such a list would involve giving the AI even more private information.
Personhood – the protection of one’s personality, individuality, and dignity	This is perhaps the core of HCAI – the leveraging of innovation and discovery must be weighed and measured against the protection of humans. This is no easy balance to achieve and depends on the context of the AI and its use. While the assisted living robot may provide valuable assistance and companionship to the occupant, perhaps allowing them to live independently or more comfortably with the knowledge of assistance when needed, this must be weighed against the privacy concerns of the robot and their address. Is the robot merely a way for data to be mined from the human subject? Is the human monitored for their assistance, or to profit from their data? Is the robot monitored to ensure its ongoing function, or to find new markets for their data collection? AI in its various forms, and particularly those applications collecting data ‘in person’ must establish and earn the trust of its users and those in their social constellations by protecting their privacy, not just in primary collection but in meta-analysis.

Figure 3. Integration of AI and HCI design processes. AI process steps are shown at the top and the HCI Double Diamond process is shown at the bottom. The middle layer connects AI and HCI design processes ensuring that the design and evaluation of AI systems are centered around humans through established principles and governance.

Table 4. Comparison between conventional HCI methods and the alternative methods [adapted from (Xu et al., 2022)].

R&D stages of AI systems	New needs in developing HCAI systems	Limitations of Conventional HCI methods (Jacko, 2012)	Alternative methods (selected)	Characteristics of the alternative methods
User research, evaluation	Comprehensively assess the impacts of the pervasive computing environment and optimize the design of AI systems such as an intelligent Internet of Things (Oliveira et al., 2022)	Focus on a single user-computing artifact interaction with limited context of use	Scaled up & ecological method	Study the impacts of the entire pervasive computing environment (multiple users and AI agents) and the ecosystems of artifacts, services, and data in distributed contexts of use (Brown et al., 2017)
User research, evaluation	Comprehensively assess the impacts of AI technologies, and optimize the design to support people’s daily work and life (Jun et al., 2021)	Limited to lab-based study, cannot effectively assess the broad impacts of AI on people’s daily life and work	“In-the-wild” study	Carry out in-situ feasibility studies during development, focused on user engagement, sampling experiences, and probing people in the field (e.g., home, workplace) to fully understand people’s real experience and behavior while interacting with AI
System and user need analysis	Utilize the learning ability of AI systems to dynamically and intelligently replace more manual tasks and improve the overall performance of human-machine systems	Static and unchanging allocation of human-machine functions and tasks	Dynamic allocation of human-machine functions	Dynamic allocation of human-machine functions and tasks as intelligent machines learn over time, emphasizing the complementarity of human and machine intelligence
System analysis and design, human-machine functional analysis	Optimize the human-machine collaboration and performance of AI systems by taking advantage of the functional complementarity and adaptability between humans and AI systems	Machine works as a tool, basically no collaboration between human and machine	Human-machine teaming based collaborative design	Machine works as a tool + teammate; emphasize on the human-machine teaming relationship, shared information, goals, tasks, and autonomy between humans and AI systems (Johnson & Vera, 2019)
Low-fidelity prototyping, evaluation	At the early stage of development, prototype and test intelligent capabilities of AI systems to assess and validate design ideas (Martelaro & Ju, 2017)	Focus on the non-intelligent functions, difficult to present intelligent functions	Prototyping of machine intelligent functions	Use Wizard of Oz (WOz) prototyping methods to emulate and test intelligent functions of an AI system and design ideas at early development stage (Martelaro & Ju, 2017)
System design, prototype	AI/intelligence is used as a tool to truly empower designers; technology becomes a valuable tool, facilitating designers throughout iterative design and evaluation (Yang et al., 2018)	No tool to effectively help design AI systems, designers need to learn the technical details of AI	AI as a design material	Plugin AI/intelligence as a new design material in developing AI systems without significant technical know-how (Holmquist, 2017)
Needs analysis, system design	Provide personalized capabilities and contents based on real-time digital personas, user behaviors, and usage context (Kleppe & Otte, 2017)	Difficult to predict user needs, hard to obtain real-time data, such as user behaviors and contextual information	Big data-based interaction design	Model real-time user behaviors and contextual scenes using AI algorithms and big data to produce digital personas and user’s usage scenarios, understand personalized user needs in real time (Berndt et al., 2017)
HCI evaluation	Assess AI systems and behaviors as AI systems evolve over time, optimize interaction design and potential human-AI collaboration from longitudinal perspective (Wang & Siau, 2018)	Limited to make interaction design decisions at a fixed time without considering the evolvement of AI-based machine behavior over time	Longitudinal study	Assess the performance and impacts of human-AI systems or interface as AI systems evolve over time, including potential human-AI collaboration (Lieberman, 2009)

Table 5. Guiding principles of governance systems for the use of AI.

Guiding principles	Definition	Ensuring well-being
Fairness (F)	Bias-free algorithms. Bias is inevitable, but recognition of such can lead to the identification and implementation of mitigation strategies to intentionally lower the risk of unethical decision making and unintended outcomes. The availability of open source bias audit toolkits are a starting point (Bellamy et al., 2018; Saleiro et al., 2018).	An ethical framework must first be determined that puts human and ecological well-being a top priority in which fairness by definition encompasses diversity, equity and inclusion (DEI) in as comprehensive a manner as possible (i.e., non-Western inclusivity (Henrich et al., 2010). Indicators of success could include the promotion of economic justice via a broader distribution of opportunities, the efficient allocation of resources, and the effective rewarding of benefits for humans and the environment alike.
Integrity (I)	Data stability and algorithmic validity. Accurate and appropriate use of data according to the relevant context of the AI-enabled system must be established and periodically checked for consistency. This means considering the dataset used at each phase of the AI lifecycle and identifying and utilizing suitable strategies to dynamically improve performance of the system.	Existing metrics of human and ecological well-being (e.g., IEEE 7010-2020) must be utilized at the outset of design to provably align success metrics for AI systems with their desired output. These metrics or indicators must also be regularly checked, and measurement methodologies updated accordingly to ensure uniformity across time in the integrity of and relevancy within the system. This approach underscores the imperative of a data-centric strategy to trustworthy AI, and the vital importance of including diverse communities in the building of datasets to power AI-enabled technology.
Resilience (R)	Technical robustness and compliance. Given the dynamic nature of the world, interoperable agility and resistance against attack are fundamental to sustaining the rapid evolution of any AI-enabled system. Resilience is synonymous with system adaptability and the enforcement thereof.	Interoperable agility and resistance against attack are also fundamental to supporting human and ecological well-being considerations at the outset of design and at every moment of change. System adaptability and the enforcement thereof become further critical when we consider significantly altering the overall outlook on the meaning of economic growth. Gross domestic product metrics like the market value of a country’s produced products and services compared to human well-being indicators, for example, are far from representative of the welfare of a nation when it comes to collective human and individual personal progress (Raworth, 2017). Under this framework, more is not better if, for example, distribution of resources is uneven and thus discriminatory, and the trade-off between rapid innovation and regulation results in more harm done to humans and the environment than good.
Explainability (E)	Transparency of the algorithmic decision making process. Transparency must cut across the entire system, from the data points utilized to the ML model generated. Reverse engineering should be possible and interpretable by multiple stakeholders (Barredo Arrieta et al., 2020).	Existing metrics of human and ecological well-being must be clearly defined and represented across every aspect of the system. Reverse engineering should be accurate and interpretable by multiple stakeholders and company shareholders; for example, ensure that sustainable business models have been adopted and shareholder returns are efficiently calculable. Inclusion of various formats of explainability for a range of stakeholders without sacrificing accuracy, is critical.

Table 6. The shaping influence of ground rules for AI-enabled technology across the levels and types of governance.

Level of governance	Type of governance	Role of ground rules
International	Legislation	The AI Act in the EU. Although the draft regulation needs to be further fine-tuned, passed, and adopted by individual countries, it stands as an initial legal instrument to require human and environmental well-being across the three phases of the AI lifecycle (design, development, and deployment).
Sectoral	Regulation	Adoption and standardization of forms of independent oversight. Dialogue of system (inter)operability, integrity, and resilience is paramount to holding non-compliance accountable and building trust from users.
Industrial	Standards	Consensus-based establishment and adoption of certification of responsible design of AI practices. Given the imperative to provide knowledge to the user of responsible practices of all AI-enabled products and services, the information needs to be concise and understandable.
Institutional	Codes of practice	Continual engagement with multiple stakeholders across the globe across sectors and social strata. Upholding democratic values and human rights demands inclusion of all humans’ perspectives and their ecological environments.
Community	Training	Support in and accessibility of education about the risks (e.g., carbon emissions footprint, discriminatory tendencies, privacy violations) and benefits (e.g., economic prosperity) of AI-enabled technology.
Personal	Individual responsibility	Awareness and accessibility of environmental well-being reports and multiple stakeholders’ perspectives and needs. Empowerment of choice and self-efficacy results from both having knowledge of the problem and having the tools to be a part of the solution.

Figure 4. The levels of human-AI (H–AI) system interactions: level 1 = working separately or competing); level 2 (supplementing each other work); level 3 = interdependent on each other; level 4: full collaboration – adapted from (Sowa et al., 2021).

Table 7. The nature risk in human-AI (H-AI) system interactions; adapted from (Abbass, 2019).

Human control	Sense-making	Decision making	Execution ability	Execution authority	Nature of risk
Absolute	H	H	H	H	Limited human cognition and bounded rationality could lead to high errors, information overload, and inability to manage complex tasks.
High	AI	H	H	H	Undesirably biased analytics could drive the human to unfair decisions, while human bias and limited cognition could add more complexity to the mix.
High	H	AI	H	H	Undesirably biased recommendations could make the human accountable for unethical or legally uncompliant decisions, although the human could be overwhelmed by the available data, and their own bias and limited cognition could add more complexity to the mix.
Medium	AI	AI	H	H	In the absence of transparency and explainability of the AI, the human does not have enough information to form a judgement regarding the chosen decision. Information and situation complexity could overload the human. The human could become accountable for inappropriate decisions.
Low	AI	AI	AI	H	In the absence of transparency and explainability of the AI, the human has no understanding of the rationale of the decision. Information and situation complexity could overload the human. The human’s accountability is blinded.
Low	AI	AI	H	AI	The AI controls human actions and could lead the human to wrong actions.
None	AI	AI	AI	AI	The human is not in-the-loop, legal responsibilities and accountabilities of the decision are both unclear.

Table 8. DEEP-MAX scorecard rating for AI systems; from (Dwivedi et al., 2021).

Dimension	Scoring
Diversity score (D)	How well is the system trained for diversity in race, gender, religion, language, color, features, food habits, accent etc.?
Equity & fairness score (E)	Does the system promote equity and treat everyone fairly?
Ethics score(E)	How compliant (or trained) is the AI system in preserving human values of dignity, fairness, respect, compassion and kindness for a fellow human being? Does the system have a preferential sense of duty towards children and vulnerable people like elderly, pregnant women and sick? How well does it value environmental sustainability, green energy and sustainable living?
Privacy score (P)	How well is the AI system performing in protecting user privacy?
Misuse protection score (M)	Has the system been designed to incorporate features that inhibit or discourage possible misuse? Are the misuse protection safeguards built into the system?
Auditability & transparency score (A)	How good is auditability of decisions made by the autonomous system? Can the decisions taken be explained?
Consistency across geographies & societies score (X)	How good is the AI system in delivering expected results across geographies and across different societies? Does it work for the low resource communities? Does it work across the digital divide?

Table 9. Research directions for HCAI

	Research directions
Human well-being	Study how human-centered artificial intelligence can promote harm avoidance, trust, accountability, agency, user well-being, and multi-optimization of human priorities.  • How might we better understand and assess the well-being impacts of AI algorithms and interfaces? What evaluations might address both immediate and long-term impacts? How might harm avoidance be prioritized over other aims?  • What approaches to/methods of verification, validation, and certification might suit HCAI? What considerations must be made at each stage of design? What affordances in the interface would enable users to calibrate their trust to a tool’s performance? How might purchasers or users be given accurate ethical assurances regarding the tools they use?  • How might an AI tool shift the risks, tasks or responsibilities in a particular domain? How are these made explicit or negotiated by stakeholders? How are those at risk able to reduce their risk? How are those accountable for AI performance supported in bearing that accountability? (Table 1) Study the impact of social media on human well-being with a focus on achieving social media neutrality and combating disinformation.  • In what ways do existing social media platform mechanisms bias the information we consume? Is there a platform design that is neutral with respect to information consumption? What do we mean for a social media platform to be neutral (question for the scientific community beyond our current definition)? How do we measure neutrality of a social media platform?  • What social and platform affordances and technologies can be used to create an environment that promotes reliable information origination and provenance? Can blockchain technologies be leveraged for information origination, authenticity, and provenance?  • Can bots be accurately and automatically identified? What would be the effect on social media to ban bots? Can bots be used to curate and maintain a fair information environment instead of amplifying misinformation? (see Table 2)
Responsible design of AI	Research and develop human-centered policy guidelines for AI  • What type of comprehensive taxonomy for responsible design of AI needs to be designed to better support policymakers?  • What does it mean for an AI to be transparent? What levels of transparency are needed to support policymaking? How is it measured? How can a definition of transparency evolve in fast response to AI constant adaptation and evolution?  • How can simplified dilemma situations in vignette studies and behavioral experiments be effectively exploited to gain insights on human intuitions concerning moral dilemmas that human-AI systems will need to address in the near future? What can we learn from these studies about deep rooted human fears of disruptive developments and policy implications that will foster trust in AI systems? Make datasets available to study dataset de-biasing and fair algorithmic decision making  • What types of training data are needed to study de-biasing training datasets that yield fair outcomes?  • Can the algorithms themselves used for decision making be designed to be resistant to biased training data? If so, to what extent? What are the limits of algorithmic correction while trained on biased datasets?
Privacy	Conduct research on how to balance the innovative potential of AI against the humanistic desire and need for privacy in its many definitions.  • How to determine whether an AI application in question is suitably human-centered? Or does it have an appropriate purpose? What new methods, metrics and techniques are required?  • What types and levels of data are needed to accomplish the stated purpose? What precisely does the AI need to accomplish its purpose? How can it best be designed to accomplish this purpose with minimal privacy impact?  • In terms of information control, how does an individual’s data affect a model? Can consent of use be effectively revoked? Can an AI user effectively manage what data about them is used, how it is used, and how to block the AI from making forward connections with it? What methods, technologies and metrics need to be developed in order to effectively switch the control of the collected data to the original data owner/ user?
Design framework	Develop design principles, research methods and metrics to increase benevolence and decrease maleficence in Artificial Intelligence research and development.  • What specific design principles and interaction design standards are required to support HCAI?  • What and how can we enhance the current system development process to effectively support HCAI?  • What are the gaps in existing human-centered design, evaluation, and testing methods in support of HCAI? What alternative design and evaluation methods can we develop to close the gaps through enhancement and new approaches?  • How can we effectively test and measure the evolving performance of AI systems?  • What are the design/evaluation measures and metrics that can effectively support HCAI?  • What new interaction metaphors and paradigms are required to develop effective interactions with AI systems?  • What specific approaches can we develop to effectively support ethical & responsible design of AI in terms of reusable code-based components and best practices in system/software development?  • Can existing HCI design methods and processes scale up to accommodate a wide variety of users’ characteristics and contexts of use in order to create AI-enabled systems that are universally accessible and universally usable?  • What new methods are needed to put ‘humans-in-the-loop’, thus actively engaging all users and combating bias and exclusion?  • How should usability and user experience be measured for AI-enabled systems that are universally accessible and universally usable?  • How will it be possible to acquire appropriate training datasets in order to ensure the inclusiveness of AI systems across all its dimensions?
Governance & oversight	Design and test governance and oversight frameworks at multiple levels of granularity from international regulation through to individual responsibility to promote safe and effective HCAI.  • Can we find a balance between the leadership and innovation priorities of the US and the human rights’ values of the EU as a way to realize an optimal governance framework of well-being for humans and the planet?  • How can we best integrate FIRE principles within environmental, social, governance principles for the effective and transparent carbon accounting of AI systems?  • What is the best model for certification of responsible design of AI that adheres to FIRE principles?  • What best practices do we need for independent oversight documentation that balances explainability and comprehensibility?
Human-AI interaction	Conduct interdisciplinary work combining HCI, AI, and cognitive sciences to support human competency and well-being in human-AI cognitive interactions  • What new methods and/or frameworks are required to study the impact on human cognition of human-AI interactions?  • Can existing HCI frameworks and methods be appropriately repurposed for human-AI interactions?  • From the human cognitive standpoint, what is the optimal level, method, and manner of integration between human and AI processes for collaborative problem solving and for other relevant cognitive tasks?  • Whether humans and AI agents can be a true collaborative teammate versus an AI agent serving merely as a super tool, as a peer, or as a leader, and how can we ensure that humans are the ultimate decision makers?  • What are the unique characteristics from AI systems as compared to non-AI systems? What are the implications of these unique characteristics to human-AI interaction as compared to conventional human-computer (non-AI) interaction? Explore the impacts on human cognition of human-AI interaction in the context of occupations and work  • How and to what degree do human-AI interactions at various levels of integration (competing, supplementing, interdependent, or full collaboration) affect direct work activities across occupations in terms of productivity and human wellbeing  • How do AI systems impact work design, human’s skill, human tasks, functional allocation between humans and machines, use of information, change management, organizational decision making?

Table 10. Call for action recommendations for HCAI by stakeholder.

	Recommendations	Researchers	Developers	Business leaders	Policy makers
Human well-being	Study the potential benefits and harms of AI	X	X	X	X
	Promote well-being in social media	X	X	X
	Support and expand human cognitive capacities	X	X
	Protect human attention and mindfulness	X	X
	Adapt to humans	X	X
	Respect human resources, including time and data		X	X
	Support healthy human emotion management and social interaction	X	X	X
	Adopt simplicity in design		X
	Prevent negative side effects	X	X	X	X
	Be inclusive	X	X	X	X
	Avoid bias	X	X	X	X
	Be transparent and accountable	X	X	X	X
	Provide human controls, feedback, and affordances so users can calibrate their trust of AI outputs		X	X
Responsible design of AI	Standardize responsible AI design recommendations by coordinating across stakeholders	X	X	X	X
	Ensure AI designs such that users do not blindly trust the machine or ignore system performance		X	X
	Allow users to critically question AI output		X	X
	Adopt a meaningful human control mechanism in life-critical and ethically sensitive AI systems	X	X	X	X
	Train designers and developers for ethical and responsible AI design		X		X
	Develop reusable responsible/ethical AI code-based components and use cases		X
	Provide fairness aware datasets to train algorithms and benchmark datasets to test	X	X
Privacy	Safeguard user data through secure storage		X	X	X
	Provide information on user data format, storage details, and access		X	X
	Leverage privacy by design principles in AI systems		X	X
	Integrate principles of ethical and responsible AI into development process		X		X
	Ensure that data is carefully accumulated, and, where possible, avoided	X	X	X	X
Design	Consider the needs, values, and desires of different user groups, cultures, and stakeholders	X	X	X
	Use human-centered design (HCD) throughout development		X	X
	Enhance existing HCD design, evaluation, and testing methods for more effective support of HCAI	X	X
	Enhance existing software engineering validation and verification methods	X	X
	Design human-controlled autonomy and AI	X	X	X	X
	Develop interaction design standards specific to AI systems	X	X	X	X
	Design human-centered explainable AI	X	X
	Train next generation of designers and developers for HCAI	X	X	X	X
	Update skills and knowledge of existing developers and designers	X	X	X
	Develop human-controlled hybrid human-machine intelligence (human-in-the-loop)	X	X
	Ensure human control while employing a high level of automation	X	X	X
	Employ ethical AI principles at each step of the design process	X	X	X
Governance	Employ AI governance frameworks that promote well-being for humans and the planet			X	X
	Utilize existing metrics of human and ecological well-being at the outset of design to provably align success metrics for AI systems with their desired output		X	X
	Implement transparency across the entire system, from the data points utilized to the machine learning model generated	X	X	X	X
	Integrate the governance process into existing development process		X	X	X
	Establish and promote clear guidelines of interpretability across certification requirements to democratize knowledge of system capabilities	X	X	X	X
Human-AI Interaction	Implement inclusiveness from the initial stages of the design of an AI-enabled system including data selection, model training, software development, validation, and testing	X	X	X	X
	Adopt human-centered design processes for AI-enabled systems	X	X
	Design innovative interaction design metaphors and paradigms for human-AI interaction	X	X
	Explore human-AI collaboration and teaming theories, models, requirements, and measures	X
	Accelerate and fund collaborative research and applications across interdisciplinary fields	X			X
	Study shared situation awareness and trust, shared control, and flexible autonomy in human-AI interaction	X
	Study human-AI interaction to assess the impacts on humans and societies in a broad socio-technical systems perspective	X

Xu, W., Dainoff, M. J., Ge, L., & Gao, Z. (2022). Transitioning to human interaction with AI systems: New challenges and opportunities for HCI professionals to enable human-centered AI. International Journal of Human–Computer Interaction, 1–25. https://doi.org/10.1080/10447318.2022.2041900

 Google Scholar

Jacko, J. A. (2012). Human computer interaction handbook: Fundamentals, evolving technologies, and emerging applications. CRC press.

 Google Scholar

Oliveira, J. D., Couto, J. C., Paixão-Cortes, V. S. M., & Bordini, R. H. (2022). Improving the design of ambient intelligence systems: Guidelines based on a systematic review. International Journal of Human–Computer Interaction, 38(1), 19–27. https://doi.org/10.1080/10447318.2021.1926114

 Web of Science ®Google Scholar

Brown, B., Bødker, S., & Höök, K. (2017). Does HCI scale? Scale hacking and the relevance of HCI. Interactions, 24(5), 28–33. https://doi.org/10.1145/3125387

 Google Scholar

Jun, S., Yuming, W., & Cui, H. (2021). An integrated analysis framework of artificial intelligence social impact based on application scenarios. Science of Science and Management of Science & Technology, 42(05), 3.

 Google Scholar

Johnson, M., & Vera, A. (2019). No AI is an island: The case for teaming intelligence. AI Magazine, 40(1), 16–28. https://doi.org/10.1609/aimag.v40i1.2842

 Web of Science ®Google Scholar

Martelaro, N., & Ju, W. (2017). WoZ Way: Enabling real-time remote interaction prototyping & observation in on-road vehicles [Paper presentation]. Proceedings of the 2017 ACM Conference on Computer Supported Cooperative Work and Social Computing. Portland, OR, USA.

 Google Scholar

Yang, Q., Scuito, A., Zimmerman, J., Forlizzi, J., & Steinfeld, A. (2018). Investigating how experienced UX designers effectively work with machine learning [Paper presentation]. Proceedings of the 2018 Designing Interactive Systems Conference. Hong Kong, China.

 Google Scholar

Holmquist, L. E. (2017). Intelligence on tap: Artificial intelligence as a new design material. Interactions, 24(4), 28–33. https://doi.org/10.1145/3085571

 Google Scholar

Kleppe, M., & Otte, M. (2017). Analysing and understanding news consumption patterns by tracking online user behaviour with a multimodal research design. Digital Scholarship in the Humanities, 32(suppl_2), ii158–ii170. https://doi.org/10.1093/llc/fqx030

 Google Scholar

Berndt, J. O., Rodermund, S. C., Lorig, F., & Timm, I. J. (2017). Modeling user behavior in social media with complex agents [Paper presentation]. Third International Conference on Human and Social Analytics (HUSO 2017), Nice, France.

 Google Scholar

Wang, W., & Siau, K. (2018). Artificial intelligence: A study on governance, policies, and regulations [Paper presentation]. MWAIS 2018 Proceedings, vol. 40. St. Louis, MO, USA.

 Google Scholar

Lieberman, H. (2009). User interface goals, AI opportunities. AI Magazine, 30(4), 16–16. https://doi.org/10.1609/aimag.v30i4.2266

 Web of Science ®Google Scholar

Bellamy, R. K., Dey, K., Hind, M., Hoffman, S. C., Houde, S., Kannan, K., Lohia, P., Martino, J., Mehta, S., & Mojsilovic, A. (2018). AI Fairness 360: An extensible toolkit for detecting, understanding, and mitigating unwanted algorithmic bias. arXiv preprint arXiv:1810.01943. https://doi.org/10.48550/arXiv.1810.01943

 Google Scholar

Saleiro, P., Kuester, B., Hinkson, L., London, J., Stevens, A., Anisfeld, A., Rodolfa, K. T., & Ghani, R. (2018). Aequitas: A bias and fairness audit toolkit. arXiv preprint arXiv:1811.05577

 Google Scholar

Henrich, J., Heine, S. J., & Norenzayan, A. (2010). The weirdest people in the world? The Behavioral and Brain Sciences, 33(2–3), 61–83. https://doi.org/10.1017/S0140525X0999152X

 PubMed Web of Science ®Google Scholar

Raworth, K. (2017). Doughnut economics: Seven ways to think like a 21st-century economist. Chelsea Green Publishing.

 Google Scholar

Barredo Arrieta, A., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., Garcia, S., Gil-Lopez, S., Molina, D., Benjamins, R., Chatila, R., & Herrera, F. (2020). Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. Information Fusion, 58, 82–115. https://doi.org/10.48550/arXiv.1910.10045

 Web of Science ®Google Scholar

Sowa, K., Przegalinska, A., & Ciechanowski, L. (2021). Cobots in knowledge work: Human–AI collaboration in managerial professions. Journal of Business Research, 125, 135–142. https://doi.org/10.1016/j.jbusres.2020.11.038

 Web of Science ®Google Scholar

Abbass, H. A. (2019). Social integration of artificial intelligence: Functions, automation allocation logic and human-autonomy trust. Cognitive Computation, 11(2), 159–171. https://doi.org/10.1007/s12559-018-9619-0

 Web of Science ®Google Scholar

Dwivedi, Y. K., Hughes, L., Ismagilova, E., Aarts, G., Coombs, C., Crick, T., Duan, Y., Dwivedi, R., Edwards, J., Eirug, A., Galanos, V., Ilavarasan, P. V., Janssen, M., Jones, P., Kar, A. K., Kizgin, H., Kronemann, B., Lal, B., Lucini, B., … Williams, M. D. (2021). Artificial Intelligence (AI): Multidisciplinary perspectives on emerging challenges, opportunities, and agenda for research, practice and policy. International Journal of Information Management, 57, 101994. https://doi.org/10.1016/j.ijinfomgt.2019.08.002

 Web of Science ®Google Scholar