Journey To The Last Dot
Why AI is at the Very Least Decades Away from Being Human
--
“By plucking her petals you do not gather the beauty of the flower” — Rabindranath Tagore
Prologue: A Journey Beyond Existential Questions
Is it possible to describe what experiencing life as a human feels like using any medium we have at our disposal i.e. words, images, paintings etc? The answer to this question alone should help us understand why the current state of Artificial Intelligence (AI) is rudimentary at best. Even with advances such as LaMDA, ChatGPT, Dall*E 2, we are at least decades away from achieving true ‘Artificial Intelligence (AI)’ let alone replicate our consciousness in a machine.
There are at least five dots, we must join, as the essay describes below for us to create a replica of ourselves. That too — knowing fully well that we have to understand our own selves parallely while replicating that understanding into machines. It is, perhaps, a challenge beyond being an existential question. We cannot simply stop at understanding our purpose in life. We have to go beyond and replicate that aspiration in a machine to be able to claim duplicity.
From limited AI to true AI
The state of AI today is that applications are built for specific purposes. For instance language models such as ChatGPT are intended to mimic human conversation but that’s it. It is not build lot to replicate other forms of human intelligence such as solve equations. Similarly, a lot of the hype surrounding Dall-E 2 (generates art based on words fed to it), GPT-3 which can write essays or poetry based on words are all language models (deep learning) that do a particular task. It is important to note that all these models have their flaws. For instance, ChatGPT has no memory. It cannot improve or improvise based on past experience ie it is not a constantly learning intelligence. However, tools such as ChatGPT or AlphaGo’s DeepMind that beat a human at the game of ‘Go’ or IBM Blue that beat Gary Kasparov at chess are all models that excel at a particular task but even these models have their flaws. The next step would be to start to fine tune existing AI to improve…