Does it exist a human-like artificial intelligence?
I define human-like as something that can act like a human in most aspects.
This depends on your definition of human-like.
If you mean a robot that looks and acts like a human, arguably, yes. Here's one of many examples: http://www.hansonrobotics.com/robot/sophia/
If you are looking for something that performs work and tasks, or works and thinks and talks like-or better than a human, the answer is mostly no, not yet.
I recommend you look at 'ANI, AGI, ASI"
ANI: artificial narrow intelligence. This is what you see around you right now.
AGI: artificial general intelligence. A theoretic AI that can "think" like a human. It does not yet exist. Estimates are between 20-60 years before we will successfully create AGI.
ASI: artificial super intelligence. In a nutshell, it is theorized to be everything we wish we could be or hope never to be. It does not exist yet. It is generally believed that, IF we create an AGI, ASI will evolve seconds or less than a decade after AGI is created.
I would say that we're not even close to a "real" human-like AI. For all the wonderful things that applications like Siri, Cortana and the like can do, they're actually really dumb compared to even a child. Of course part of that, IMO, is that most AI applications are not embodied and don't experience the world the way humans do. So if you show an AI a video of a dog walking behind a table and briefly disappearing from the frame, it has pretty much no ability to apply "common sense" and know that the dog will reappear in a few seconds, and that if it does't reappear, it's probably because it found its favorite toy on the floor behind the table.
For some examples of the kinds of "easy" questions that computers still don't do well at, check out the Winograd Schema Challenge. You might also find this page interesting: http://www-formal.stanford.edu/leora/commonsense/
It might be of note to comment/update that the SuperGLUE benchmark, which is a suite of common sense reasoning tasks, incorporates the aforementioned Winnograd Schema challenge, among other tests that are said to be reflective of natural language understanding (as opposed to simply its processing or the optimal statistical generation of language). The most recent result by Google (T5) has reached parity with the Human Baseline in terms of the average performance across all tests.
The leaderboard is available below, and the site includes subscores for each of the challenges: