Jeremy Owen Turner
Jeremy’s work uses philosophical knowledge to help improve metrics for achieving AGI (Artificial General Intelligence) systems and agents. Specifically, Jeremy is researching the degree to which an artificially intelligent agent can deeply contemplate an imagined and/or virtual world-model. Jeremy asks questions like, “what does it mean for an autonomous virtual-character (NPC/Bot) to be a ‘deep-thinker'”?
This interdisciplinary dissertation identifies contemplative intensity in cognitive architectures for virtual-agent minds. This research semantically interprets contemplative intensity through rubrics derived from Kantian Aesthetic Philosophy and Crawfordian Interaction Design. This dissertation adopted mentalist rather than behaviorist epistemological approaches towards exploring the artificial psyche and the interactive capabilities of a virtual-agent’s mind (from mechanistic finite-state machines to “deep thinkers”).
A qualitative methodology was selected to triangulate hermeneutic exegesis alongside an expert interview validation process. These methods scrutinized the: state-of-the-art; architectural cognitive components, interconnections; and their relationship to five speculative virtual-agent contemplation scenarios. This hermeneutic exegesis was conducted with graph-based flow-chart representations of these scenarios. These graphs were created with the PyPhi calculator which normally deals with quantitative measurements of consciousness (i.e. Integrated Information Theory, aka. IIT) rather than the qualitative assessment of contemplative intensity. These flow-chart examples represent the component configurations of three established cognitive architectures: ACT-R, SOAR, and CLARION.
The purpose of this study was to determine whether contemplative intensity could be detectable through IIT-graph visualization at an architectural level in a virtual-mind and which cognitive architecture (and/or architectural components) was most likely to intensify contemplation for this mind in a virtual world. These flow-chart illustrations representing architecturally enhanced minds helped categorize virtual-agents according to their overall cognitive capabilities and provided discussion inspiration for the experts during the interviewing process.
A combination of solo hermeneutic exegesis and expert opinions determined that the IIT lens could not sufficiently qualify contemplative intensity for virtual-agent minds at an architectural resolution level of flow-chart analysis. On a positive note, the experts confirmed the validity of the mentalist epistemological perspective as well as the initial assumption that CLARION was the best architecture to intensify contemplation. However, these experts attributed their CLARION preference to architecturally-agnostic metacognitive properties rather than to CLARION’s idiosyncratic architectural configuration. These experts also insisted on the value of extra-architectural mechanisms and algorithms as well as the semantics from the virtual environment.
These additional findings further confirm the qualitative limitations of the IIT lens for architecturally identifying contemplative intensity in virtual-agent minds. This dissertation concluded with proposed heuristics and suggestions for future research.
Keywords: NPCs; Virtual-Agents; Artificial General Intelligence; Cognitive Architectures; Video Games; Arrabales; Consscale; Kant; Tononi; IIT; Cerullo; CCMT; Hermeneutic Exegesis; Creative Introspection; Textual Analysis; Crawford;
Position: PhD Researcher