Embodied Interaction emphasizes that human perception is grounded in bodily experience and sensory engagement with the environment (Dourish, 2001). Unlike computers, which process data in a binary manner, humans interpret information through cultural, experiential, and embodied contexts, making data inherently nuanced. Dourish argues that our physical presence in the world shapes our interactions, meaning experiences are always situated in a bodily context. Homewood et al. (2021) expand on this by suggesting that sensory engagement is individually unique. Even when presented with the same data, interpretation varies due to intersubjectivity. Humans do not merely "input" raw data; rather, we draw on personal and cultural experiences to create meaning.
Large Language Models (LLMs), like ChatGPT, interpret data by analyzing semantic elements in a text, thereby determining intent (Böhm et al., 2024). However, I argue that LLMs lack the embodied nuances of human sense-making. Light et al. (2019), through the lens of queer theory, discuss how digital tools shape identity by prescribing behaviors, reducing human experience to metadata and oversimplifying embodiment.
Reflecting on OpenAI's (2024) introduction of memory in ChatGPT, the model can now retain past interactions, enabling it to identify user preferences. For example, if a user previously discussed pottery and sourdough bread making, the model might classify them as a hands-on, DIY enthusiast. However, this raises concerns about its ability to truly understand human identity and intent. Leer et al. (2023) explore whether LLMs possess Theory of Mind (ToM) and suggest that while they can generate human-like responses, they lack the nuanced perception gained through real-world social interactions, including body movements and nonverbal cues.
As an interaction designer, I speculate that for AI to progress toward Artificial General Intelligence (AGI), it must move beyond treating humans as mere datasets and instead recognize them as emotional beings with lived experiences. To expand its contextual understanding, AI systems should: - Incorporate concepts from Theory of Mind (ToM), particularly "joint attention," which is crucial in human collaboration (Department of Health Sciences, n.d.). Can AI "see" and "feel" beyond data structures? - Move toward "Embodied Artificial Intelligence," where AI is not confined to screens but actively engages with the physical world.
In conclusion, the potential of AGI lies in contextually aware AI systems that interpret and respond to human experiences within real-world contexts, making meaning through subjective perception.