While artificial intelligence (AI) design continues to aspire towards romanticised visions of intelligence and human-likeness, the increasing dichotomy between its technological and social facets has given rise to tangible evidence of surveillance, inequality and discrimination. This evidence emphasises a pressing need for systems more aligned with human expectations and values, but computationally, it remains extremely difficult to embed concepts as fundamentally fluid and situational as value and meaning. Such difficulty has led present-day research to formalise these intricate principles of human behaviour, in all their complexity, through predictive machine learning models built on data extracted from how people behave not in relationship to AI but in the absence of it.

This research proposes that the inseparability of the AI social artefact from the Human entails, foremost, an understanding of the interaction between concept (humanlike design) and percept (anthropomorphism). Consequently, the effect of these variables on each other is explored in two stages: first by reviewing whether our mental models of humanlike AI are built on pre-existing knowledge and schemas of other people (Just Like Me), and second, by probing their homogeneity with respect to their corresponding human counterparts (But Not Exactly). Building on the results of these initial investigations, this study demonstrates that the locus of the interaction, in practice, cannot be neatly extracted from manipulating the correlation of these variables, but rather by looking at how it is continuously constituted in the dynamic between the two. With combinations of effects born in the fleeting concept-perception interaction, it is concluded that very few behavioural patterns and mental models can truly be universalised, and for the most part, one context cannot be neatly applied to the next.

To address this epistemological limitation, the present study proposes a novel methodological approach – taking influence from cognitive dissonance theory – that requires the designer/researcher to locate themselves in the complexity of this socio-technical world and to reflect on how their interventions impact that which they are observing. In addition to providing a visual method of investigating the interaction, the proposed approach is ultimately intended as a design strategy for addressing the nonlinear and often chaotic dynamics of an emergent field of study: the Human-AI.