Doing AI Differently
An International Initiative Integrating Humanities into the Core of AI Development
Yutong Liu & Kingston School of Art / Better Images of AI / Talking to AI 2.0 / CC-BY 4.0
About: Doing AI Differently
The Doing AI Differently initiative challenges traditional approaches to AI development by positioning humanities perspectives as integral—rather than supplemental—to technical innovation.
The initiative will build a community of scholars from diverse disciplinary backgrounds, with a special focus on bringing humanities and arts contributions further up the AI development pipeline, to the fundamental design of the technology.
Core Challenges
1. Developing Interpretive Technologies.
Leveraging current AI architectures to implement deeper interpretive capabilities, developing approaches to represent multiple perspectives and capture semantic depth while maintaining computational tractability.
2. Exploring Alternative AI Architectures.
Exploring fundamentally new approaches to AI design that move beyond current limitations in foundation models and gradient-based learning to enable more pluralistic and culturally adaptive AI systems.
3. Advancing Human-AI Ensembles
Moving beyond simple substitution or assistance models, to foster new relationships between human and artificial intelligence, each contributing unique capabilities to achieve outcomes neither could accomplish alone.
The initiative’s objectives include:
1. Developing pilots demonstrating humanities-driven technical advances in AI.
2. Fostering interdisciplinary partnerships through targeted funding mechanisms.
3. Creating pathways and scalable models for humanities scholars to participate in large-scale AI projects.
Potential Benefits:
New paradigms for complex, context-dependent tasks in AI research
Enhanced AI tools for deep contextual analysis across disciplines
AI systems better aligned with societal values and ethical considerations
The rapid deployment of increasingly sophisticated AI systems makes this timing critical, and we seek to develop inclusive, research-led ideas that will provide practical ways to address some of the recommendations set out in the recent UK Government AI Opportunities Action Plan.
To support research under this theme, an Arts & Humanities Research Council (AHRC-UKRI) International Science Partnerships Fund (ISPF) funding call will be launched in April 2025. This call will be focused on collaborations between the UK, US and Canada. This adds the opportunities for research and innovation in AI – artificial intelligence – in the UK (see Transforming our world with AI).
The Initiative is led by The Alan Turing Institute, University of Edinburgh and the UK’s Arts & Humanities Research Council (AHRC-UKRI) with partner institutions in the UK and North America.
Contact: The New Real newreal@ed.ac.uk or via Data-Centric Engineering at The Alan Turing Institute
Expressions of Interest: See below
Workshop: Doing AI Differently
Integrating Humanities into the Core of AI Development
We ask: What can AI learn from the humanities?
A workshop will convene an international community to advance this novel research theme.
Main Day: Thursday 13 March 2025, 10am-4pm
Working Group Sessions: Friday 14 March 2025, 10am-4pm
Location: The Royal Academy of Engineering, London SW1
Aligned with The Alan Turing Institute’s major annual conference (AIUK) on 17-18 March
The workshop will present and refine the research vision and topic scope for Arts & Humanities Research Council (AHRC–UKRI) International Science Partnerships Fund (ISPF) funding call launching April 2025.
With a long-term goal of establishing a new field of study, the workshop will explore what a new field would achieve and how it would change the current landscape.
To register your interest in participating in the workshop complete an Expression of Interest [here] by 17 February 2025. A limited number of bursaries are available to support travel costs for participants who do not have other institutional support. See EOI form for details.
The Urgent Need for this Work
The rapid advancement of artificial intelligence is driving an evolving consensus that insights from the humanities are more crucial than ever for understanding and shaping the future of AI. While much excellent scholarship has been done at this intersection, the new initiative responds to the need for the humanities not just to contribute to post-hoc analyses of AI outputs or provide novel data sets—but to help shape fundamental aspects of the technology. This integration is not merely beneficial but essential for addressing critical challenges in AI that cannot be solved through technical approaches alone.
A significant driver for this need (though by no means the only one) is the rise of large language models, marking what we describe as a qualitative turn in AI. For most of its history, AI was a primarily quantitative discipline. Its main outputs and interfaces took the form of numbers. That is no longer true: they now take the form of texts, more akin to the documents generated and studied by humanists than equations devised by mathematicians and engineers. Contemporary AI is now qualitative in a way it was not before. And so it is crucial that qualitative disciplines make fundamental contributions to the way this technology is developed and designed.
A key contention is that qualitative disciplines can help contribute to solving one of the flagship problems in contemporary AI: homogeneity. The vast majority of contemporary AI models are based on a very small set of basic architectural designs (specifically, some combination of deep neural networks and reinforcement learning). This places significant limitations on the kind of outcomes and experience AI models can support—as well as the populations of people for whom they are designed to work.
There is furthermore a homogeneity of implementation as well. Current development is focused on single-user, general-purpose AI. There is a widespread framework of human-AI interchangeability: the tacit assumption is that if an AI agent is introduced to a system, a human agent must be removed. Ultimately, we view the various insights and perspectives offered by the humanities—including critique, creativity, and interpretation—as central to addressing the question: How do we create AI that enhances, rather than replaces, human capabilities?
There is a need for sustained collaboration between the humanities and engineering to integrate sophisticated humanities frameworks with scalable computational approaches, and our twin base in The Alan Turing Institute's Data-Centric Engineering programme and The New Real provides a foundation for this mission. And yet—the perspectives of the humanities are now endangered, at a time when they are needed more than ever. A core intention of this initiative is to articulate how a range of humanistic perspectives—including the humanities, arts, and qualitative social sciences—can contribute to burgeoning efforts in AI, specifically to foundational aspects of how contemporary AI models are developed, designed, and deployed.
Transformative Shifts in AI: Challenges and Opportunities
The current wave of AI development presents both unprecedented opportunities and complex challenges. These challenges demand a fundamental shift in how we approach AI, moving beyond purely technical solutions to embrace the insights and methodologies of the humanities. To date, humanist scholars have made significant contributions to AI in terms of post hoc analysis and novel data sets. While these approaches are vitally necessary to the AI ecosystem, the aim of our initiative is to integrate humanities perspectives into the very foundation of AI’s technological infrastructure. We identify three primary challenges which, if addressed constructively, present significant opportunities in the societal impact of AI development.
The Qualitative Turn
The field of AI has undergone a qualitative turn. Early AI systems primarily dealt with quantitative data, producing numerical outputs, and were best understood by the mathematical logic governing these systems. By contrast, modern AI generates outputs—text, images, and even code—which resemble human cultural artefacts. This shift necessitates a new set of skills for analyzing and interpreting these outputs, skills traditionally cultivated within the humanities.
The Homogenization Problem
There is notable homogeneity among the key contemporary AI’s foundation models. When a few dominant models underpin the AI ecosystem, any biases or limitations within those models are amplified and propagated across countless applications. Likewise, certain kinds of architectures support some possible applications and outcomes while excluding others. This can lead to the reinforcement of existing social inequalities and the exclusion of diverse cultural perspectives. There is a long history of feminist, ecological, and indigenous critiques of the limitations of traditional computational frameworks. Integrating these critical perspectives into AI development is crucial for creating systems that are truly inclusive and equitable.
The Transformation of Human Cognition and Capabilities
As people increasingly collaborate with AI systems, the nature of human thought itself is likely to be transformed. AI is not merely a tool; it is a partner in our cognitive processes, shaping how we access information, make decisions, and even understand ourselves. This co-evolution of human and artificial intelligence presents both possibilities and challenges. The humanities are essential for navigating this uncharted territory. We must develop frameworks for understanding how AI is changing human cognition and design sociotechnical systems that enhance, rather than diminish, our uniquely human capabilities.
Doing AI Differently is led by The Alan Turing Institute, University of Edinburgh and the UK’s Arts & Humanities Research Council (AHRC–UKRI) with partner institutions in the UK and North America.
Data-Centric Engineering
The Alan Turing Institute
The New Real
University of Edinburgh and The Alan Turing Institute
