OpenAI showed it in 2017 with the sentiment neuron (https://openai.com/research/unsupervised-sentiment-neuron https://openai.com/research/unsupervised-sentiment

OpenAI showed it in 2017 with the sentiment neuron (https://openai.com/research/unsupervised-sentiment-neuron https://openai.com/research/unsupervised-sentiment-neuron). Basically, the model learned to classify the sentiment of a text which I would agree is a general principle, so the model learned a generalized representation based on the data. Having said that, the real question is what percentage of the learned representations do generalize. For a perfect model, it would learn only representations that generalize and none that overfit. But, that's unreasonable to expect for a machine *and* even for a human. Maybe we just don't know. We are staring at a black box and doing some statistical tests, but actually don't know whether the current AI architecture is capable enough to get to some kind of human intelligence equivalent.