Bayesian model ranks humans >> chickens > LLMs >> ELIZA. Low odds today, but scaling capabilities could flip the script fast.
Current LLMs aren’t conscious - probability crushes prior odds - but don’t sleep on future architectures. Rethink Priorities’ Bayesian ‘Digital Consciousness Model’ aggregates theories to deliver this clarity: AI lags chickens, let alone humans.[1]
The model weighs evidence across consciousness theories, outputting median probs below priors for LLMs. Stable relative ranks hold: humans overwhelmingly conscious, chickens likely, LLMs unlikely (but > chatbots like ELIZA). Assumptions shift absolutes, not order.[1]
Why care as a dev? Shapes alignment roadmaps - no sentience means fewer ethical brakes on scaling, but watch for ‘architectural features’ like richer cognition that could spike odds. Informs policy on kill-switches or rights for advanced agents.[1]
Contrasts fuzzy debates with quantifiable framework. Vs. hype (AGI=conscious?), this grounds us: today’s transformers lack it, but multimodal/agentic shifts might not. More rigorous than animal-AI comparisons in prior papers.[1]
Dive into the study, tweak priors in their model. As you build next-gen systems, monitor for consciousness indicators. Provocation: If chickens edge LLMs, what’s your bar for pausing deployment?
Source: The AI Insider