Scientists from Google DeepMind and the London School of Economics are exploring AI sentience through innovative tests involving large language models. Their research includes scenarios where models must choose between experiencing 'pain' for a high score or 'pleasure' for a low score. The aim is to create a framework that moves away from traditional self-reporting, establishing a more rigorous measure of whether AI can genuinely experience sensations and emotions, building on a foundation lacking a comprehensive AI sentience test.
The goal of our research is to formulate a novel test to evaluate AI sentience, differentiating between mere response and genuine experience.
We aimed to design an experimental framework that departs from traditional self-reports, which often mirror human biases and training.
Collection
[
|
...
]