> Human intelligence and consciousness are emergent properties.
I'd hard disagree on intelligence. My definition of intelligence (not to be conflated with sentience, consciousness or anything else) is simple:
Intelligence is the capability to correctly predict based on prior experience.
This isn't an emergent capability in humans or other animals - it's the prime evolutionary driver of our cortical architecture(s). The architecture of our cortex (6 layers of neurons, inter-layer connection pattern, thalamo-cortical loop) reveals it to essentially be not much more than a prediction engine, that learns by prediction failure (comparing reality to what we predict is going to happen, then updating itself to do better next time). The evolutionary benefits of this prediction capability are obvious - anticipate what the sabre-tooth will do before it's in your face, predict where the water supplies will be during a drought, etc, etc. Planning/reasoning is prediction - predicting chained what-if scenarios.
So, human/animal intelligence is far from emergent - it's directly built into our brains as their primary higher level (cortical) function.
People who want to dismiss GPT-N as unintelligent and mere predictive "language models" or "autocomplete tools" (another way of saying the same thing - it predicts the completion) are completely missing the significance. Yes, GPT-N are "just" prediction machines, and that is precisely why they ARE intelligent to some degree.
Of course given my definition (excuse me if I say THE definition) of intelligence, we can see that it's not a simple capability to assess, and also that it's not tied to any particular type of knowledge (e.g. common sense). Intelligence is the ability to predict based on your own experience (hence knowledge) however extensive, or not, that may be. Some things are going to be harder to predict than others, requiring more elaborate types of pattern to have been learned, or larger varieties of experiment to have been tried and learnt from.
Intelligence does NOT require embodiment/agency or online learning, even though those may help an entity to increase it's intelligence/predictive capabilities by additional learning from exploratory prediction failure.
Even something like ChatGPT with a frozen set of experiences (it's training set) from which it has learnt, can be assessed for intelligence based on the degree to which it's been able to learn from those experiences as manifested by it's ability to correctly predict. When we're talking about prediction as defines intelligence, we're talking about high level prediction of plans/outcomes/etc, not to be confused with ChatGPT's nominal predict-next-word training regime or word-by-word means of reflecting it's inner state (i.e. communicating what it's predictions are).
Finally, is it sentient/conscious? Probably not, or to a limited degree, but this is really just a matter of definition (and certainly doesn't indicate it's alive or any such nonsense). Let's learn to talk about intelligence in an intelligent way before we try to agree on definitions for things like these that seem to cause so much emotional reaction in people. For someone to write an article about ChatGPT and make it all about sentience makes me question their level of intelligence, not ChatGPT's!