Why does some AI professionals think deep learning is "intelligent"?

Question

Human is intelligent, this is the ultimate definition (although not rigorous) of intelligence.

Ordinary programs are not considered "intelligent", even they can do so many smart things with high efficiency, only because they can't do human "intelligence" jobs (e.g. object recognition) well.

Deep learning is called "intelligence", only because they can do some form of human "intelligence" jobs much better than ordinary programs.

But deep learning does not have the intelligence of human at all, they are essentially just some new form of algorithm that can solve some type of problems much better than ordinary programs.

Big language models are "amazing" only because they give human the feeling of "amazing" and human are so easy to be amazed. For example, a random number generator can get the right answer of 25% of multiple-choice questions, isn't that even more "amazing" and "intelligent"? It is just the big language model algorithm happen to have this level of result, it doesn't matter the result is 25% or 75%, anyway it is not intelligent even in the right direction towards to human intelligence at all. So there will be definitely some upper limit of such algorithm that it can never "fake above", like the 25% for the random number generator.

Even if human intelligence is just an algorithm too, obviously it will be architecturally different than today's deep learning algorithms. The chance that you can happen to find the algorithm that is equivant to human intelligence is basically zero. Even more, if human intelligence can indeed be described by an algorithm, the algorithm will be too complex to run on ordinary computers for a high chance.

So why not just admit it is nothing but a new form of algorithm and use it to make life better and stop talking about intelligence? Is is understandable that the general public think these are intelligent and are on the road to AGI, but why does some AI professionals think deep learning is "intelligent"? (some even says big language models begin to have consciousness...)

Answer 1

Partly due to the fact that deep learning can solve tasks considered to be challenging, and it does so by learning. This is the main differentiator w.r.t. old fashioned hardcoded algorithms. Indeed you will also encounter the wording artificial intelligence used for machine learning algorithms such as random forests or any other algorithm using learning. I'm not saying I agree with this, I'm just stating that it is what happens in practice.

But deep learning does not have the intelligence of human at all, they are essentially just some new form of algorithm that can solve some type of problems much better than ordinary programs.

As of today, you cannot back this statement. While we can understand how simple deep learning actually is, we don't fully understand human intelligence, so it may turn out that human intelligence it's not that hard either.