TrustLAB: Translates – Is it time for a unified approach to trust and AI?

Artificial Intelligence (AI) is no longer confined to the realm of science fiction or academic speculation – it’s embedded in everyday life, shaping how we work, communicate, and make decisions. As AI systems become more capable and more deeply integrated into society, the
demand for systems that are ethical, secure, and worthy of human trust has intensified. Within this realm three interrelated, but distinct, concepts have emerged: responsible AI, trustworthy AI, and trust in AI.

Figure 1 Google Relative interest worldwide in time

At first glance, these terms might seem interchangeable. But in practice, they are quite distinct. Understanding how these concepts differ – and where they overlap – is essential for researchers, developers, policymakers, and end users alike. It is my hope that this blog may stimulate some ideas for some adventurous researcher moving forward. I should add the disclaimer that I am neither an ethics researcher nor a cybersecurity researcher so I also apologise in advance to those much more knowledgeable on these subjects.

Responsible AI: Ethics, Values, and Governance

Responsible AI refers to the design, development, deployment, and oversight of AI systems in ways that align with ethical principles, human rights, democratic values, and legal standards. It emerged from the broader field of tech ethics and gained momentum from 2016 onwards.

Figure 2 Justice meets AI: When courtroom decisions depend on code, who holds the gavel?

As a response to high profile failures such as algorithmic bias in criminal justice (e.g., COMPAS) and data misuse (e.g., Cambridge Analytica), organisations like the OECD, UNESCO, and the European Commission began issuing guidelines for ethical AI. These guidelines stressed that AI must not only be technologically efficient but also socially and morally accountable.

Responsible AI responds to questions like:
● Is this AI system fair?
● Does it respect autonomy and dignity?
● Who is accountable if it causes harm?

Trustworthy AI: Technical Reliability and System Integrity

Trustworthy AI refers to AI systems that are reliable, secure, and robust against manipulation, failure, or adversarial threats. In contrast to responsible AI, trustworthy AI traces its lineage back to reliable software systems and secure computing environments, but it gained sharper focus with the rise of machine learning, particularly deep learning. Unlike traditional software, ML systems often behave like black boxes. This opacity introduced a new kind of risk: not just coding bugs, but misbehaviour arising from biased data, unintended correlations, or adversarial examples.

Figure 3 In a world of black-box machines, trust may be built on verifiable security and design.

Reflecting its foundational base in computer science, software engineering, and cybersecurity, Responsible AI focuses on building systems that deserve trust from a technical standpoint.

It seeks to answer questions like:
● Does this AI system do what it says it does?
● Is this AI system secure against attacks?
● Can this AI system explain its outputs?

The EU High Level Group’s Ethics Guidelines for Trustworthy AI:

The EU’s Ethics Guidelines for Trustworthy AI, published in 2019 by the High-Level Expert Group on AI, are conceptually grounded in a human-centric approach that integrates ethical principles, fundamental rights, and technical robustness. The guidelines emphasise that AI systems must be lawful (complying with all applicable laws), ethical (respecting principles like autonomy, justice, beneficence, and explicability), and robust (both technically and socially). In many ways it marries Responsible AI and Trustworthy AI.

What these guidelines don’t do is define trust or explicitly ground trustworthiness in any psychological basis. There is no attempt to unpack what trust is, how it functions psychologically, or how it might or is measured. It is a goal rather than a dynamic process rooted in cognition and emotion.

Psychological Trust in AI: A Tale of Two Flavours

Trust in AI suggests the study of the psychological state or attitude of a human user toward an AI system. To paraphrase Rousseau et al (1998), it is characterised by the human’s willingness to accept vulnerability based upon the positive expectations of the intentions or behaviour of the AI system. The literature on psychological trust would lead us to believe the expectations are based on the AI’s ability, benevolence, or integrity (the so-called ABI dimensions) or to package neatly, their trustworthiness (Mayer et al. 1995). Tomlinson et al (2014), in discussing organisational trust, suggest that value congruence between trustors and trustees is a necessary component of trustworthiness, and to some extent this may capture the essence of responsible AI at a micro-level.

Figure 4 Who is working for whom?

This psychological perspective raises many new questions not addressed by those studying responsible or trustworthy AI:

● Do users feel vulnerable when interacting with this AI system, and if so, why are they still willing to engage with it?
● Do users perceive the AI as having sufficient ability to perform its intended function competently?
● Do users perceive the AI as acting in their best interests? Whose interests are they acting in?
● Do users trust the AI system more when they anthropomorphise it, or perceive it as a social agent?

These questions are deeply relational, experiential, and cognitive-emotional in orientation, staying focused on the dyadic trustor–trustee relationship, where the trustor is a human and the trustee is an AI.

Information Systems (IS) researchers have translated the ABI dimensions to automated systems where each dimension has its own technological corollary e.g., ability (performance, functionality), benevolence (purpose/helpfulness), and integrity (process/predictability/reliability) (McKnight et al. 2011; Sollner et al. 2013). The integration of moral or ethical values into trust models for automated systems remains relatively underexplored. McKnight, Sollner and others have successfully integrated aspects of psychological trust and technological trust.

Applying this perspective is not dissimilar to proponents of Trustworthy AI but infused with more than a hint of psychological trust:

● Do users perceive the AI system as helping them achieve their goals? Does the user believe that the AI system has the capability, functionality or features to do what needs to be done?
● Do users believe that the AI system or its designer provides adequate and responsive support and help to the users?
● Do users perceive the AI system functionality and features have been designed appropriately, and are dependable, understandable, controllable, and predictable?
● Do users believe that the AI system will consistently operate properly and predictably? What system-level indicators most influence user trust in AI across different contexts?

One can easily see that these questions are technologically grounded yet psychologically nuanced, centering on the trustor–trustee relationship where the AI is assessed not only for its functional competence but also for its perceived intentions and design integrity, reflecting a synthesis of technological and psychological trust dimensions. While such theorising is useful in the context of non-intelligent systems, AI is different from mere automated systems. AI exhibits both cognitive and system-level capabilities. If an AI were to behave too much like a human, it might be less trustworthy due to the uncanny valley phenomenon and yet if it behaved too much like an automated system, its trust potential might be limited. In many respects we want it all, promethean in our ambition – we want AI to think and act like humans but also think and act rationally.

Trust and AI
Trust and AI is different. It is a more holistic and inclusive perspective on trust and AI that considers not only trust between humans and AI, but AI and other AI, and indeed AI and other non-intelligent systems. It distinguishes between different types of AI – Artificial Narrow
Intelligence (Narrow AI, weak AI, or ANI), Artificial General Intelligence (General AI, strong AI, or AGI) and Artificial Super Intelligence (Super AI or ASI).

Figure 5 When we conflate different types of AI, obscuring nuanced discourse on the differentiated capabilities, risks, and ethical implications inherent to each AI paradigm.

Trust and AI is multi-levelled moving from the micro to the macro, from considering individual perspectives to that of humanity. It is not just the study of a trustor and a trustee. The AI ecosystem is complex and multi-layered, comprising referents that may be interdependent, embedded, and in some cases, both cognitive and autonomous, and in others pre-programmed, rule-based and bounded. It includes end users, technology vendors, regulators, AI, and other non-intelligent systems.

I think this perspective asks us as researchers to respond to new questions:

● What does trust look like between two AI systems, and does it resemble human models of trust?
● Can trust exist between autonomous systems and rule-based systems, and if so, what mediates that trust?
● How do our perceptions of trust in different referents and interdependencies within the AI ecosystem shape the formation or breakdown of trust in AI?
● How does the type of AI (narrow, general, super) change the nature or necessity of trust?
● What does it mean to scale trust from the individual to the societal or even global level, especially across divergent political or cultural systems?
● How can we design AI ecosystems that are not only fair, secure, and interpretable, but also capable of sustaining trust across diverse agents and scales – from individual human users to complex, interdependent machine networks?

In contrast to responsible AI, trustworthy AI and psychological trust in AI (in both the human and IS flavours), these questions are expansive, systemic, and deeply interdisciplinary.

Final Thoughts
At its core, trust and AI requires us to unpack not only the ethical, legal, and technological aspects of trust but the psychological and social aspects too. Trust and AI is a multi-disciplined endeavour. Computer scientists must work with ethicists. Psychologists must collaborate with
engineers. Policymakers must engage with everyone. Only then can we develop AI users and build AI systems that can be trusted.

AI is shaping the future and we are shaping AI. We need to design AI that is responsible in intent, trustworthy in design, and trusted in practice. Each of these pillars matters; understanding the distinctions between them is the first step toward building a better AI-
powered society.

As we move forward, the challenge will be to harmonise these three dimensions—not to blur them, but to recognise their unique contributions to a common goal: an AI ecosystem that earns and deserves trust, whether human or otherwise, while advancing the common good and protecting our humanity.

Author: Professor Theo Lynn, (Full) Professor of Digital Business and Associate Dean for Research at Dublin City University Business School.