I've been following AI for over 40 years at this point, and to be honest, your article leaves me sort of baffled.
Your claim above is particularly strange, because any program that can really converse with us will actually understand language, in any practical way.
Even a child can give others access to a consistent representation of their own internal mental states, and has models of the mental states of others. Even a child can answer the question, "Why?" about most of their behavior and their internal mental states. Even a child can learn a fact and fit it into their mental framework from a single encounter with it. Even a child can do simple symbolic processing - "there is poop on the kitchen floor, it must be the dog."
And no AI program has made even the first steps along the way to these goals.
GPT-2 or -3 shows no sign at all of intelligence. It is a sophisticated form of Mad Libs from a huge corpus of existing text and speech utterances from humans. You can't ask, "Why did you say that?" or "What are you thinking?" or any questions at all about its current state of its symbolic processing, because there isn't any.
If you had a program where you could interrogate its internal states, ask why, ask it to explain its reasoning or make an argument for its beliefs, it would be intelligent, by exactly the same argument Turing made. We simply have made no progress in this direction at all.
Do I think this is impossible? Heck, no. We know that intelligent machines are possible, because almost 8 billion of us exist.
So why have we made so little progress on Strong AI? Is it some magic?
No, it's two very regular issues.
The first is that Strong AI is probably very hard - or perhaps there's some conceptual breakthrough that hasn't yet come. An example of this is fusion power - we're almost sure it's possible (look at the Sun!), we've made significant progress, probably we will get there, but it's just a hard problem.
The second is money. Machine learning models offer immediately financially valuable results. But a machine that could communicate to you with the same intelligence as a five-year-old would have almost no economic value at all - you'd be better off hiring a bright teenager at minimum wage.
