The other day I showed Meow a picture of Angela Merkel; precisely this one:
and taught her to say “Merkel”along with me. I made her repeat it thrice and let her resume watching Tinkerbell.
Last week, Angela Merkel was shown many a times in the news after the Brexit results and I was on cloud nine when Meow identified her every time she was shown on TV. Like all proud mommas inhibiting earth, I was bragging about it to my mother when she saw Evelyn Harper from Two and a half men on TV and yelled “Mekkel!!!”
For those of you who did not pay attention to that insanely hilarious show your sibling downloaded from torrents, this is Evelyn Harper.
As an adult you wouldn’t think of this a as a resemblance but apparently the only things that caught her attention were the red suit and blonde hair. To confirm , I showed her this:
But this was just “Akka” (young lady) to her. I was baffled.
And then we googled for Angela Merkel in a different colored suit:
She got this one right. “Mekkel Mekkel Mekkel”.
I took the experiment a bit further and showed her a picture of Meryl Streep but wearing a dress instead of a suit.
She marked this as a half hearted “Mekkel” followed by “Mekkel illa” (Not Merkel).
This experiment has seriously got me thinking about the cognitive learning process in human beings. There is no hard and fast rule in identifying or classifying things; there is something much more than just the defined features that makes the process of identification precise. Had it been only the haircut and the suit then Meow should have labelled Diana as “Mekkel” and if it had been just the haircut and the aged look, she should have called Meryl Streep as “Mekkel” without any hesitation.
The results mimic my many experiments with classification problems using machine learning. Some results are true positives and true negatives but at the same time some others are false positives and false negatives.
When I showed her more and more distinct examples of Angela Merkel, Princess Diana, Evelyn Harper and Meryl Streep and gave her a surprise test later, she was able to accurately identify all four of them irrespective of their hair and attire.
Our identification/ classification accuracy increases as we see/ experience more and more over time, but what is that hidden ingredient that makes our classifications 100% accurate after sufficient amount of data but not any machine’s ?
Is it the absence of a pushy momma?