To prove that a human is human, the following question must be answered first: what is human?
Defining the essence of the human condition should be the first step to establish the proper criteria to detect one.
The only skill that singles out our species from others is the capacity of our human cognition to interpret semantics… and this ability seems to be targeted by deep learning.
I actually expect machines to eventually take over the turing test and score even better than humans some day.
But until that fateful day, I think that semantic tests would replace efficiently captchas to detect humans.
For instance, identify metaphors:
"The sea flowed inconsolably from her face"
Options 1: It is about sea tides
Option 2: It is about someone crying
Option 3: It is about water currents in the sea
Option 4: It is about someone drowining.
I gotta be honest, there were kids failing metaphors in SATs (Standardized exams for college admission in the US), so it may filter illiterate or undereducated people as a side effect. Is that good or bad? I leave it to you guys to judge…
Another way of doing semantic tests would be visual, where images are shown and it has to be matched with its context.
Example: "Jeremy is very tired, he needs a ____"
Option 1: (show a picture of someone running)
Option 2: (show a picture of someone driving)
Option 3: (show a picture of a chair)
Option 4: (show picture of a comb)
Now, if you need proof of UNIQUE human, that would mean that a means to identify the individual in a way that any behavior or characteristic will be detected as singular. That is a huge challenge and the only reliable way of doing that is by forcing a deanonymization of the user, which we know it is not acceptable.
I think we have to ask ourselves if “proof” is the right objective here, I think that is misguided.
What we need is not “proof of unique human”, but a high cost for bot makers and mass account creation, and that is a much more achievable goal.