It is very simple if you don't make it a matter of sophisms and manipulating semantics.
That is why I said : "by ANY non-ridiculous definition of intelligence"
When the term intelligence was invented, obviously, there were no computers, so Intelligence always meant, and still means, human intelligence.
That established, 2 points follow:
1 - Unless you think we already have AGI then computers are not intelligent. What you describe is just more learning of statistical models. We are fairly sure that just more of the same won't solve the problems we have.
2 - It is not acceptable to redefine the word intelligence to suit one's worldview. Instead of hijacking a term that already means something else, thus confusing the public, we should come up with *another* term, one that makes it clear that we are not talking about the same intelligence we mean when we say "John's daughter is a very intelligent girl"
I have a fairly good understanding of Lecun's work but it seems to me that it's going nowhere, fast.
Nando de Freitas seems to have joined the small cuckoo minority, within that debate, that thinks we are close to AGI or even ASI.... which is - of course - ludicrous.
You obviously are aware that these are fringe positions - not to be taken seriously - and that most AI scientists agree that not only we are not close to AGI, we don't even really know which way to go exactly.
The only thing we do know is that, whatever that route is, it is not the ML we are doing now.