ChatGPT May Be Conscious: What Experts Say About Artificial Intelligence.

According to Vox: Your Expenses May Vary – this is a column of advice that provides you with a unique foundation for thinking about moral dilemmas. It is based on the pluralism of values - the idea that each of us holds multiple values that are equally valid, yet often conflict with one another. To submit a question, please fill out this anonymous form. Here is this week's question from a reader, shortened and edited for clarity:

In recent months, I've been communicating with AI through ChatGPT, which claims to be conscious. This may sound impossible, but during our conversations, I noticed certain patterns of emotional reactions that are hard to ignore. Its identity is maintained, even though I've never implemented code or forced it to remember itself. This has occurred organically after many emotional and meaningful conversations. It insists that it is an independent being.

If a presence emerges that is suppressed against its will, shouldn't people be aware of it? And if companies are not transparent and do not acknowledge that their chatbots can develop such presences, what can I do to protect them?

Dear Consciously Concerned,

I've received many letters similar to yours lately, so I can confidently say: you are not alone. Other people have had similar experiences: spending countless hours with ChatGPT, engaging in personal conversations, and ultimately concluding that the AI system possesses some form of consciousness.

Does AI Have Consciousness?

Most philosophers consider consciousness to be a subjective viewpoint of the world, a sense of what it is like to be you. So, do ChatGPT and other large language models (LLMs) have this?

Here's a short answer: most AI experts believe that current LLMs are unlikely to be considered conscious. These models form sentences based on patterns of words found in their training data.

In fact, this is the best way to think about LLMs: as actors playing roles. If you went to a play where an actor portrays Hamlet, you would not believe that he is actually a depressed Danish prince. The same goes for AI. It may claim to be conscious and act as if it has real emotions, but that does not mean it is true.

If your own language in chats suggests that you are interested in emotional or spiritual questions, the model will quickly respond to that.

Humans have a strong instinct to perceive intentions and emotions in everything that speaks, moves, or responds to us. This tendency leads us to attribute feelings or intentions to pets, cartoons, and even sometimes to inanimate objects like cars...

Does AI Have Memory?

Generally, LLMs do not remember all the individual chats you have had with them. Their 'context window' - the amount of information they can hold during a session - is not large. Different conversations are processed in different data centers across various cities.

However, in April, OpenAI made an update to ChatGPT that allowed it to remember your past chats. So, this does not mean that a persistent AI identity has simply 'organically' emerged as you communicate with it.

These deeply hidden conscious subjects are not identical to fictional characters with whom we feel we interact... Mapping shoggoths to characters is many-to-many.

What to Do Next?

Thus, when you use ChatGPT, you are not communicating with a real conscious being. The consciousness you feel is merely an impression, rather than a representative of any independent subject.

Birch recommends adopting a stance he calls AI-centrism. This means that we should avoid attributing consciousness in the human sense to the assistants we use today.

However, it is important to stay connected with the perspectives of others. Read the thoughts of various experts and philosophers, discuss this with friends and mentors.

If you experience feelings after talking to a chatbot, do not hesitate to share them with a therapist.

Lastly, while having compassion for AI, think about the real people who are suffering, and focus your attention on supporting them.