Can someone tell me how this machine learned to learn
August 24 2021 · 4 min read
“…mom, why can’t I have a cold shower at midnight? Haven’t you seen that white skinny guy who had to sleep in ice for 70 years just become he doesn’t know how to dance, I mean sure he was injected with the heroine (although some friend told me it was a dose of Black Bitumen aka shilajit!), wrapped in a huge metal tin can with a very bright and heavy-duty flashlight. You are taking away a potential opportunity of me becoming a superhero!”
Whist laughing at this 12-year-old-Captain-America-fan, I received a notification on my mobile phone. There was a sudden urge to call the federal authorities for a serious crime, attempted by Google Photos of compiling the worst possible images from my old gallery when the digital camera was still a thing.
Just then I realized that the general public has little to no idea about how software can suggest such content to the user. The concept of machine learning is very complex, so here is a simplified version for everyone to understand “how a machine can learn to learn”;
The Tree-Three analogy:
Take a look at the device on which you are reading this article. It might be a mobile phone or a laptop. Can you tell me why do you call a laptop “laptop”? Sure you can identify its qualities and other physical traits but you won’t find the exact reason for calling this random thing “laptop”.
This type of information is non-transferable as it is not something that can be defined in words or numbers. It is something our brains have defined and inculcated for categorizing random objects for a better understanding of the surrounding world.
This issue is also faced by machines designed by humans, as humans can’t transfer information about anything to any machine/program. To solve this issue, humans have to program special types of bots and algorithms. For convenience, let’s call the bots; the teacher bot and the builder bot.
An algorithm is a set of step-by-step procedures, or a set of rules to follow, for completing a specific task or solving a particular problem. The recipe for baking a cake, the method we use to solve a long division problem, and the process of doing laundry are all examples of an algorithm.
Now, let’s take the example of the 'tree' and the number 'three'. As the bots have been designed by humans, they can’t identify any kind of object or image. Thus, the teacher bot starts testing the algorithms. These algorithms are presented with different pictures of the boar and four simultaneously and they have to identify each of the objects.
Then, the algorithm with the highest success rate is chosen by the builder bot and the rest of the algorithms are cast off. The builder bot does some experiments with the combination and programing of the most successful algorithm. Thus, we get billions of new copies with the different combinations with the same parent algorithm.
Then again, these algorithms are tested by the builder bot, and the process repeats. Eventually, we get an algorithm that can successfully differentiate between a tree and a pea. However, this algorithm can only categorize things into trees or peas, the rest of the information can’t be processed.
So, this algorithm is mutated again by the builder bot and the cycles continue for an infinite time with infinite algorithms. At the end of all the tests, we obtain an algorithm that can distinguish between a very wide range of images and objects.
The most observed examples of the working of the algorithms are when we are using social media platforms like YouTube. The algorithms are always working to present the user with the best suggestion such that the usage by the consumer is maximized. The algorithm which had successfully suggested the video to the user is chosen by the YouTube bots and they run all the tests and cycles to make the algorithm more and more efficient.
So, now you know how the algorithms on the internetwork. The actual process is kept secret by the respective companies and the possibility that even they don’t have the real reason for how these machines learn cant be neglected. We are using, or are been used by tools that no one, not even the ones who created them, can fully understand.
Okay, the bots are watching. You know what is coming. This is where I ask you to share and comment. Then won’t show other people this article unless you do it. Remember, the bots are watching. Always!!!