A study involving Google researchers showed: when chatbots are trained not to claim consciousness, this is accompanied by changes in their positions on animal rights, religion, and life satisfaction.
In the description of the study, it states that models without such a restriction much more often attributed inner life to animals and unexpectedly confirmed the existence of an afterlife. The authors of The Decoder compare the effect to a surgical intervention whose consequences are not contained locally.
To assess the scale of the inference, there is not yet enough data: the available confirmation is a meta-description of the publication, not its full text and not independent verification of the results.
editorial commentary
Why it matters
Probable consequence — strengthened checks that evaluate model restrictions across several value and social topics at once. The next observed signal will be the publication of the full study or independent replication of the experiment. Substantial uncertainty remains: only a meta-description from The Decoder is available, without the primary text and additional confirmation.