The concept is highly flawed, but today's 'social scientists' are often idiots.
Why do I say flawed? Let's examine the tool and the approach.
A chatbot is trained from a mass of diverse sources. That means that the data comes from a mass of different people. Sane people, weird people, abnormal people, even bots in the many internet sources. For example, Reddit is hardly wholly sane and reliable or even human in the posts.
Would one seriously want to prepare a formal research paper based on inputs from possibly fake people? Given a conclusion based on data from sources you cannot be certain were actual people? Of course not. Using chatbots as stand-ins for real people is a big stupid approach.
It is like doing a poll of political preferences and accepting some results like 75% of responders favored Hillary Clinton for POTUS, but not allowing that half of them might be bots run by the DNC.
The concept is highly flawed, but today's 'social scientists' are often idiots.
Why do I say flawed? Let's examine the tool and the approach.
A chatbot is trained from a mass of diverse sources. That means that the data comes from a mass of different people. Sane people, weird people, abnormal people, even bots in the many internet sources. For example, Reddit is hardly wholly sane and reliable or even human in the posts.
Would one seriously want to prepare a formal research paper based on inputs from possibly fake people? Given a conclusion based on data from sources you cannot be certain were actual people? Of course not. Using chatbots as stand-ins for real people is a big stupid approach.
It is like doing a poll of political preferences and accepting some results like 75% of responders favored Hillary Clinton for POTUS, but not allowing that half of them might be bots run by the DNC.
So when I say flawed I mean flawed.