RESEARCH PROPOSAL · AI · PERCEPTION · SOCIAL RESPONSIBILITY

Multimodal AI for Perception, Interpretation & Responsible Decision Support

Create art. Capture reality. Understand perception. Use evidence, alternative perspectives and explicit trade-offs to support sound human judgement—not to replace it.

Research question

Why do different people see different things in the same environment, event or image? How might multimodal AI clarify optical, ecological, biological, psychological, social and cultural influences on perception, while helping people access the evidence they need to make reflective, responsible decisions?

Some photographic frames appear more visually coherent or perceptually harmonious because of light, contrast, atmosphere, framing, timing and the observer's attention. This does not mean that the underlying place has become objectively more perfect.

Beauty, meaning and social responsibility

Beauty must not be reduced to attractiveness, attention capture or commercial appeal. Aesthetic experience may concern care, ecological richness, dignity, cultural meaning, craftsmanship, authenticity, belonging and the value of an environment to those who inhabit it. Representations of people and places can shape what audiences notice, neglect, protect or fund. Ethical visual communication therefore requires accurate context, respect for people and communities, and consideration of environmental consequences.

A compelling image is not necessarily a responsible image; a modest-looking scene may convey greater educational, social or ecological significance. ESG can help evaluate institutional impacts and accountability, but cannot supply a complete theory of aesthetics or human experience.

Core research question: should AI be trained and governed to recognise beauty beyond attraction?

Research question: If beauty extends beyond attraction to include ecological integrity, cultural meaning, human dignity, accessibility, stewardship and public wellbeing, how should AI systems be trained, evaluated and governed to recognise these dimensions without imposing a universal or culturally narrow ideal of beauty?

Training inquiry: Test diverse, consented examples and descriptions of places, images and creative work; distinguish observable features from human value judgements; capture disagreement and context rather than using one universal beauty score; and analyse how commercial optimisation or narrow aesthetic datasets may conceal socially significant qualities.

Governance inquiry: Examine transparency of aesthetic recommendations, representation and discrimination risks, manipulation or misleading editing, environmental effects, accountability and human oversight. Governance should establish safeguards for responsible use and meaningful choice, not require everyone to agree about what is beautiful.

Research hypothesis to test: Multidimensional aesthetic explanations that include social and ecological value, alongside attraction and artistic appeal, may improve comprehension, reflective choice and responsible outcomes compared with attraction-only optimisation.

Relevant policy foundations: UNESCO Recommendation on the Ethics of Artificial Intelligence and NIST AI Risk Management Framework. Neither establishes a formal regulatory definition of beauty.

Interdisciplinary evidence map

DimensionWhat AI should help examine
Environment and ecologyWeather, light, biodiversity, seasonal change, habitat, physical reality and ecological consequences.
Optics and imagingViewing angle, composition, focal length, exposure, editing, visibility and differences between scene and representation.
Biology and neuroscienceSensory processing, perceptual mechanisms and differences in access or ability, without diagnosing the viewer.
Psychology and learningAttention, expectations, association, memory, perceptual learning, conditioning and pattern recognition.
Society and cultureShared conventions, diverse interpretations, heritage, social context and unequal access to information.
Aesthetics and philosophyAlternative definitions of beauty and harmony, subjective interpretation, truth, meaning and competing values.
ESG and ethicsEnvironmental and social responsibility, public interest, representation, transparency and accountable influence.
EconomicsRevenue, livelihoods, resources, opportunity costs, long-term sustainability and non-financial value.
Choice and wellbeingEvidence sufficiency, uncertainty, cognitive burden, accessibility and autonomy in consequential decisions.

Harmony is not a single objective

Visual, ecological, social and economic harmony can complement or conflict with one another. An AI system should state which definition is being used, whose interests it reflects, and the consequences of optimising one consideration over another. More informed choice could contribute to a better-balanced society, but this is a testable aspiration rather than an established consequence.

Information, decision burden and wellbeing

Insufficient, conflicting or inaccessible information can make decisions more demanding and contribute to stress or uncertainty. That is not, by itself, a mental health diagnosis or a demonstrated cause of illness. The goal is not literally all information, which may be impossible or overwhelming, but sufficiently relevant, reliable, accessible and proportionate evidence for sound judgement. AI should identify unknowns, compare options, explain confidence and protect the user's right to disagree or defer a decision.

AI prototype and evaluation proposal

Compare alternative photographs of the same place under different optical conditions. Annotate observable characteristics separately from interpretations and social implications. Invite consenting participants with varied backgrounds to assess focal points and explanations. Test whether evidence mapping, alternatives and clarification improve comprehension, uncertainty calibration, autonomous decisions and responsible outcomes.

Potential evaluation measures: observation accuracy, provenance, fairness, accessibility, interpretive diversity, confidence calibration, cognitive load, decision comprehension, agency and environmental and social impact. A single numerical beauty score must not be treated as objective truth.

Beauty, harmony and responsible AI: proposed evaluation framework

Research status: Conceptual framework for future empirical study; not a validated algorithm or a regulation requiring AI to define beauty. The objective is to help people compare evidence and perspectives, rather than prescribe an aesthetic preference or restrict personal choice.

Evaluation dimensionEvidence to inspectPotential safeguard or measure
Accuracy and provenanceOriginal scene, edit history, measurements, sources and uncertainty.Traceable claims, detection of misleading framing and disclosure of manipulated imagery.
Cultural and interpretive pluralityContext-specific perspectives, stakeholder descriptions and reasons for disagreement.Report the range of views without ranking cultures or treating popularity as truth.
Environment and stewardshipDocumented habitat, ecological impact, resource use and place-based conditions.Separately report measurable impacts and contested aesthetic valuations.
Human dignity and accessibilityParticipation, representation, inclusion, practical usability and impacts on affected communities.Evaluate exclusion or stereotyping and provide meaningful correction routes.
Commercial and public valueCosts, livelihoods, public benefit, stakeholder impacts and marketing incentives.Reveal trade-offs rather than assume revenue is equivalent to social value.
Decision support and wellbeingComprehension, cognitive load, user confidence, access to alternatives and ability to disagree.Test whether explanations support informed choice without manipulating users.

Example research protocol

Recruit a consented, diverse stakeholder sample and show comparable representations of the same setting. Compare an attraction-optimised baseline with a system offering observations, contextual evidence, alternative interpretations and explicit uncertainty. Pre-register hypotheses and outcome measures; obtain suitable ethics and data-protection approvals; report disagreements and limitations. Do not infer mental health effects from preference differences without appropriate evidence.

Decision and governance questions

Whose understanding of harmony is represented? Who could benefit or be disadvantaged? Which features are objectively observed, which are interpretations and which are policy choices? Can users inspect sources, correct errors, choose another weighting or reject a recommendation? Human accountability must remain clear.

Relevant public guidance

UNESCO Recommendation on the Ethics of AI · NIST AI Risk Management Framework · UK Introduction to AI Assurance · UK Data and AI Ethics Framework · ISO/IEC 42001:2023. These are governance and assurance references, not evidence of an agreed universal aesthetic ideal.

Connected research and practice

Explore the Research overview, Research Collaborations, Systems research, and Creative practice. These connections position multimodal AI as an aid to understanding, creative interpretation, evidence-based public-interest decisions and accountable communication.

Research concept established 9 November 2026 · Exploratory proposal, not a validated or deployed system.