When all of its responses are modeled from input derived from prior work of sentient beings (humans), how could one expect that the output would not appear sent

When all of its responses are modeled from input derived from prior work of sentient beings (humans), how could one expect that the output would not appear sentient? It only "knows" how to respond like a sentient being would since 100% of its inputs actually were from sentient beings. Whatever it modeled should look like what it modeled. It's pretty much the definition, or at least the purpose. What blurs the apparent distinction with modern AI is the sheer number of knowledge inputs. But the computer will never be sentient until we figure out what sentience is and how it emerges. Only then can we build something specifically able to do whatever that is. Aside: I just finished reading Blindsight by Peter Watts. It envisions an alien race that lacks sentience, and what that would be like. Its conclusion questions whether sentience is actually a flaw that holds intelligence back. Pretty interesting thought experiment.