Pollikobi, A Mango, and LLMs

In our excitement with AI (artificial intelligence), we are forgetting how knowledge is formed within the human experience, and a limited view of this process might make us underestimate the individual worth.

In a short story by the poet Joshimuddin (more commonly known as "Pollikobi"), we get introduced with a Emperor from Iran who comes to the Indian subcontinent in winter, and demands he is served mangoes immediately.
excerpt from the short story "Taste of a Mango"
 
Since it wasn't the season for mangoes yet, and the Emperor had no time for patience, the right hand man comes up with this unorthodox way of giving him a feel for it, combining molasses, tamarind, on his beard for a fiber replacement. The emperor was happy with the results true — but anyone who has had mangoes can tell, this is far from the experience of actually eating a mango, whatever the quality might be.

illustration from the short story "Taste of a Mango"

Latest research and thinking suggest that humans have a lot more than just 5 senses, some even stretching it to 20+. Pressure and pain beyond touch, balance, direction, feeling full, blood pressure etc. are some of the often overlooked senses.

Of course, we see the AI labs trying to make up for the gaps left by Large Language Models (LLMs), and combining information from image/video models, depth information etc. You may have seen new captchas that require you to point a 3D hand towards the direction of a dog in an image. This might help the machines to understand directions better. I am sure we will keep seeing newer inputs being harnessed for better models in coming days.

When human perception and internal knowledge creation is driven by inputs from so many different sources, it is difficult to capture it all in highest fidelity in just text, even if combined with images, audio, and videos. Even though the rise of civilizations owe everything to the invention of writing, we cannot forget that every encoding of communication — is inevitably — lossy. When we unpack the encoded information in our heads, we are bound to lose and misinterpret a lot of the meaning. No amount of reading about flights can make up for the lack of the experience of actually flying for the first time. 

I was listening to David Epstein (American journalist, author of "Range") yesterday, and how AI today helps him save hours by quickly summarizing latest research articles in a new topic. We have to acknowledge the benefits of quick summary, but we cannot deny that it also robs us of a more nuanced understanding, or interesting sidenotes that could have resulted from a thorough reading. 

Add to that, no two people ever read the same book, they all interpret it at least a bit differently from what the author intended. No two opposing fans see the same football match, one feels they won it fairly and another feels like being robbed, even after seeing the exact same footage. We all attach our own identity to form subjective perception of objective reality.

Maradona's 1986 World Cup goal: 'Hand of God' to some, 'cheating' to others

With the zealous pursuit of AI into all realms, we shall continue to see massive changes in all of knowledge jobs. Trying to regulate changes now while the rest of the world goes ahead, and catch up with changes 40-years later will not be beneficial to us. So I am not proposing an overregulated demolition of new technology. My hope is that at the decision making levels, it is important that we acknowledge the limits of data we can capture, and continue appreciating the richness of individual experience and human knowledge.

P.S. Thanks Farhin for reminding me of this story by Joshimuddin while I was rambling about epistemology.