BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.iacapconf.org//iacap-2026//talk//MDMVBW
BEGIN:VTIMEZONE
TZID:US/Central
BEGIN:DAYLIGHT
DTSTART:20250717T000000
TZNAME:CDT
TZOFFSETFROM:-0500
TZOFFSETTO:-0500
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20251102T020000
RDATE:20261101T020000
TZNAME:CST
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20260308T030000
RDATE:20270314T030000
TZNAME:CDT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
SUMMARY:Social Triangulation and Representational Stability: A Quinean-Dav
 idsonian Framework for AI Architecture - Justin Mullins
DTSTART;TZID=US/Central:20260717T100000
DTEND;TZID=US/Central:20260717T103000
DTSTAMP:20260726T081339Z
UID:pretalx-iacap-2026-MDMVBW@pretalx.iacapconf.org
DESCRIPTION:Current AI systems (world models\, large language models\, and
  joint-embedding predictive architectures) learn representations in episte
 mic isolation. Each system interacts with its environment but never with o
 ther agents in ways that constrain its representational structure. This pa
 per argues that epistemic isolation has a specific\, measurable consequenc
 e: representational instability under distributional shift. We draw on two
  converging philosophical arguments. Quine’s inscrutability of reference
  and structural realism establish that representational stability cannot c
 ome from internal architecture alone\; proxy functions guarantee that any 
 arrangement can be systematically reinterpreted. What constrains admissibl
 e interpretations is social coordination. Davidson’s triangulation argum
 ent establishes that the concept of error\, meaning the distinction betwee
 n model failure and environmental change\, requires a second perspective. 
 We translate these insights into five architectural constraints for multi-
 agent systems: Multiple Perspectives\, Observable Reactions\, Conflict Det
 ection\, Normative Pressure\, and Structural Invariance. Each constraint i
 s traceable to a specific philosophical argument. We then propose the Gava
 gai Shift experiment\, a controlled paradigm for testing whether social co
 ordination constraints improve representational stability under distributi
 onal shift. The experiment uses a grid-world in which object appearances c
 hange while functional roles are preserved\, comparing isolated agents wit
 h socially constrained agents across three metrics. The contribution bridg
 es naturalized epistemology and AI architecture: philosophical theory moti
 vates specific engineering predictions\, and experimental design tests the
 m.
LOCATION:Apollo Auditorium
URL:https://pretalx.iacapconf.org/iacap-2026/talk/MDMVBW/
END:VEVENT
END:VCALENDAR