Mind in the Machine? Cross-Disciplinary Perceptions of Consciousness in Artificial Intelligence

Paper Title

Mind in the Machine? Cross-Disciplinary Perceptions of Consciousness in Artificial Intelligence

Publication Info

• Topic area: Perceptions of consciousness and intelligence in AI systems across academic disciplines.
• Keywords: AI consciousness, large language models, cross-disciplinary study, human–AI interaction, AI ethics, machine intelligence, anthropomorphism, consciousness theories, AI governance, subjective experience.

Background and Problem

• Problem / challenge: There is no consensus on the definition of consciousness, and existing studies on AI consciousness are limited by conceptual ambiguity, U.S.-centric samples, and a focus on lay populations with limited technical knowledge. This creates gaps in understanding how informed academic groups perceive AI consciousness and intelligence.
• Significance: Perceptions of AI consciousness influence trust, interaction, and ethical considerations, shaping AI design, governance, and societal integration. Understanding these perceptions is critical for addressing anthropomorphism, ethical challenges, and policy decisions.
• Motivation and related work: Prior studies have examined public and academic perceptions of AI consciousness but often relied on predefined definitions or hypothetical scenarios, limiting insight into participants’ genuine beliefs. Few studies have focused on academics with relevant expertise, leaving a gap in understanding cross-disciplinary perspectives on AI consciousness.

Solution

• Proposed approach: An empirical study using an online survey to investigate how academics from formal sciences, humanities, and natural sciences perceive consciousness and intelligence in large language models (LLMs) and future AI systems.
• Novelty:
  1. First comprehensive, non-U.S.-centric survey of academic perspectives on AI consciousness.
  2. Inclusion of diverse academic disciplines with informed mental models of AI and consciousness.
  3. Avoidance of predefined definitions of consciousness to capture participants’ genuine conceptualizations.
  4. Examination of individual traits and belief systems influencing perceptions of AI consciousness.
• Procedure and key techniques:
  • Surveyed 553 participants from 41 countries, grouped into formal sciences, humanities, natural sciences, and others.
  • Measured perceptions of LLM consciousness and intelligence, familiarity with consciousness theories, and ethical positions on AI governance.
  • Analyzed data using statistical tests (Kruskal–Wallis, Spearman correlations, logistic regression) and qualitative content analysis of open-ended responses.

Results

• Concrete findings:
  • 50.4% of participants attributed some degree of consciousness to current LLMs, with a mean rating of 2.1 (on a 0–10 scale) among those endorsing gradual consciousness.
  • 53.9% judged LLMs to be equally or more intelligent than an average human.
  • Intelligence and consciousness ratings were positively correlated (ρ = 0.41, p < 0.001).
  • Key predictors of consciousness attribution included belief in gradual consciousness, functionalism, and the relationship between intelligence and consciousness.
• Advantage over baselines:
  • Lower overall attribution of consciousness compared to prior studies, likely due to participants’ informed mental models and direct interaction with LLMs.
  • More nuanced insights into how conceptual beliefs, rather than technical knowledge, shape perceptions.
• Experiments / evaluation:
  • Survey included 42–45 questions, covering AI literacy, interaction frequency, intelligence and consciousness perceptions, ethical views, and demographic data.
  • Participants were grouped by academic discipline, and responses were analyzed quantitatively and qualitatively.
• Limitations and future work:
  • Limited by current LLM capabilities (text-based, non-embodied) and potential self-selection bias in the sample.
  • Cross-sectional design prevents analysis of changes in perceptions over time.
  • Future work should include longitudinal studies, ethnographic approaches, and exploration of cultural differences.

Summary

This study investigates how academics from diverse disciplines perceive consciousness and intelligence in large language models (LLMs) and future AI systems. About half of the participants attributed some degree of consciousness to current LLMs, with perceptions shaped more by belief systems (e.g., gradual consciousness, functionalism) than by technical knowledge. Intelligence and consciousness ratings were moderately correlated, and interaction frequency influenced intelligence perceptions. The findings highlight the need for ethical frameworks and AI designs that account for diverse and contested views on AI consciousness. Future research should address evolving AI capabilities and cultural differences to deepen understanding of these perceptions.