Speculation. If we assume, for the sake of argument (it is a big assumption), that human and artificial intelligence both are neural networks and that if trained on the same data, will both produce the same intelligence, we still are left with the question of whether humans and machines can be trained on the same data sets.
Humans have bodies, exist in ecosystems and complex biological contexts and this, all of this, is by necessity a part of the data that we train on. Machines just train on language. images and other things that can be captured digitally. Sure, we could also capture all the possible movements of a body and deploy sensors to start to try to collect a data set that would mirror what it means to be embodied in the world but we don’t have an inkling of an idea of how much data that would be or exactly how it would look, or how to wire those sensors to capture embodiment in some sense.
That would suggest that the data set that humans train on is radically different from that of machines, at least for now.
But is that true? We could also argue that embodiment, evolutionary context and embeddedness are all captured in language. This argument would amount to an argument that says that language contains a compressed version of the whole of the evolutionary and biological experience – and that metaphors code for embodied existence, for example. If this is the case, language is all you need to replicate intelligence (all other assumptions here equal, including that we are also neural networks).
The question of the equivalence between the data sets L (language) and B (embodied existence) asks a fundamental question about the nature of language, and one that is intriguing to explore. We can cut it in a multitude of different ways and try to figure out if the embodied existence we lead can be captured in other ways – but at some point this will be relevant to the recreation or mimicry of human intelligence.
There are arguments here to be made about limits – such as: there is no way to collect data that corresponds to the evolutionary experience and history without rerunning evolution at the pace of evolution, an argument about the incompressibility of human experience as biological beings, and hence about the impossibility of a replication of human intelligence.
All of this is very speculative, of course, but there is something here worth exploring more, a set of philosophical problems and models that may matter. Note to self: language as coding human experience effectively and efficiently would be a semiotic field of exploration – the idea of a fourth function of the Peircean sign.