In the wake of GPT 3, it seems that AGI is possible. No question about that. But my question is can it be more "human" than we think? I watched an interview with Eric Elliot and GPT 3. The machine says that it can lie, I don't think it is impossible for a machine to lie (as far as I know, but I could be wrong). But then it begs the question, how many human traits will AGI pick up? And if picking up negative human traits is a consequence of AGI itself how "general" do we want to make it?
I am questioning the purpose of making AGI not because I fear it, I just don't understand where the direction of this entire concept is going. Any references to the discussions and ethics regarding this matter will be very appreciated.
Hi Ashwin, thank you for your response. Yeah, I guess it does make sense that an AI will respond purely based on the dataset. It does mean that to make AGI we would require a massive amount of datasets and an equal or an even larger amount of time to process it.
"where the information is gathered by the AI itself is an area which requires a bit more exploration," I think this line sums it all best. I am more confident in an answer now. Thank you!