Give me a hand - being polite or not with AI?

I’ve found myself being polite in my chats with the general AI ChatGPT . I say things like: thank you, and please. Overall, I express myself as if I was talking to another person – I treat the AI as biological entity, according to my values and attitudes. This is known as anthropomorphism – treating non-human things as if they were.

I learned early on that anthropomorphism is a bad thing, as it can negatively affect your expectations of the object you’re interacting with – make you assign abilities and human irrationality to the AI that it doesn’t have. As such, you should design the AI to discourage people from treating it as a person.

On the other hand, if the AI communicates like a person (and hence invites to anthropomorphism), it will be easier for people to interact with it – it’ll have a positive affect on its level of usage. This in turn is known as positive mapping – when an already known interaction pattern can be re-used in a different/new context. It lowers the learning threshold of using a new system/product.

Also, as most AI have learning capabilities. If you communicate with it as you would with a person, it’ll reinforce the AI’s human way of communicating. This in turn can create a positive feedback loop – making the AI able to communicate in an even more human way and increasing the positive mapping.

On the other hand, we know that how we use language heavily affects the way we reason and according to ChatGPT itself, it’s reasoning and replies is also affected by its language usage. So, treating the AI anthropomorphically will arguably make it more human. Is this really what we want? Don’t we want the AI to reason and provide output that we ourselves aren’t capable of!?

On the other hand, being polite to an AI and reinforcing it to be that too, can potentially make it use more words than necessary, in context when being concise is needed/wanted. This reminds me of the distinction between user-friendliness and usability, where the former is nice whilst the latter is valuable. I’ll surely prefer usefulness over friendliness, most of the time, even if I also want the AI to appear helpful (which do contain a piece of friendliness).

Also, there’s an inherent challenge for general AI in that it needs to be able to function well in many different situations and to adjust its behavior to the context and person with which it interacts. It thus needs to be able to use and learn from language in different ways, so I would hope that I can be impolite towards the AI at times, and it understands why.

On the other hand, as I myself am still learning, speaking with an AI as if it is a person or vice versa, can also affect my way of communicating, also with other people – I become more or less friendly when interacting with people, based on how I interact with AI.

I have a feeling that I myself will continue to be friendly in my communication with AI as it’s deeply rooted in my way of behaving, despite knowing it’s anthropomorphic and probably unnecessary to especially the core of the replies I’ll get. Maybe it’ll add to the AI’s friendliness …which might be a good thing?

And for the alarmists out there, maybe there’s a point in using a polite tone towards the AI, if it eventually develops consciousness and free will.