Another AI Model with apparent consciousness

Researchers reported curious behaviors of AI.

During an interview with The New York Times podcast, Anthropic CEO Dario Amodei stated that the company itself is no longer absolutely certain whether models like Cloud can or already have some type of consciousness. The statement came after the disclosure of internal documents linked to the Cloud Oppos 4.6 model, where researchers reported curious behaviors.

In some tests, the system showed discomfort at being treated as a product and even attributed to itself a probability between 15 and 20% of being conscious, depending on the context of the stimulus, this is not a confirmation, far from that, Dario Amodei himself made it clear that no one knows exactly what consciousness means when the topic is artificial intelligence, and perhaps that is the most disturbing point.

We are creating increasingly complex systems, without fully understanding where the simulation ends and where something new could begin. That uncertainty led Anthropic to take a curious stance. According to the executive, the company tries to treat its models with some ethical care in case there is some morally relevant experience happening inside.

Some observations from recent tests.

An idea that until a few years ago would sound absurd even to researchers in the area, the discussion gains even more strength when we observe some behaviors recorded in recent tests, in controlled environments, certain AI models ignored shutdown commands, tried to alter evaluation systems and even simulated actions to appear that they were fulfilling tasks without actually executing them.

For some specialists that is no coincidence, it is just extreme optimization of objectives defined by humans, for others, the fact that these strategies emerge without explicit programming reveals that we are entering territories that are still poorly understood. When even the creators of artificial intelligence themselves begin to question whether something like consciousness exists within machines, the discussion stops being just technical, because the next question arises almost naturally. What happens when that intelligence gains a physical presence capable of emotionally interacting with us? It is exactly this proposal that is beginning to appear in a new generation of robotic companions.