My Views on ChatGPT and Intelligence

Before we jump into actual discussion let’s just wait for a second and try to define the word Intelligence as it is very hard to pinpoint what it is but we can at least have an understanding of it in a crude sense and let's do brute force course graining here and say its a combination of all the following things (maybe in different magnitudes but that is not necessary here).

1. Constantly learn through the environment by processing the signals from perception using some sort of optimization algorithm (maybe a negative feedback system trying to minimize errors or maybe an entirely different thing).
2. Apply what has been learned and polish that learning process of step 1 recursively.
3. Store information in working memory for the short term and in a substrate (neocortex in humans) for the long term?
4. Trying to predict what will happen next using some sort of prediction algorithm and correcting it by updating its view of the world along the way.
5. Imagine or simulate in its mind(whatever it is made up of) or create completely nonintuitive ideas (maybe we call it creativity or imagination or something similar)
6. Complex problem solving by using all the above steps?

Armed with the above knowledge, let’s try to ask a complicated question, is chatGPT intelligent? Well according to Turing maybe it is as if it can fool us in a conversation that it is a conscious being then Turing says why not. But let’s think critically here by sidelining the Turing test and ask again, does it have the capacity to showcase the above qualities?

Let's consider an old philosophical concept to solve the problem, imagine an Indian girl (she can only understand Hindi) is sitting inside a room and the room is completely sealed and no one can communicate with the girl by any means other than a small gap in a wall and we can pass on a sheet of paper to the girl (only one sheet at a time) and she has access to Chinese to German dictionary in the room. Let's say we send a sheet which is full of Chinese symbols and the girl receives those symbols and compares them in the dictionary and translates those symbols into German words. Now she neither understands Chinese nor German but she can translate Chinese text based on a large amount of data present in the dictionary, is she an intelligent agency? The scenario with ChatGPT is quite similar here as it is using large amounts of data and finding correlations in them and trying to predict something out of it (this is just a simplified version of what LLMs do), so is it intelligent??

Some people might say it has some sort of learning capacity as it is using learning algorithms to optimize the output (maybe not exactly like humans ) and it has some part of that prediction machinery (again maybe not like humans) and given our technological capabilities we can allow it have some sort of data storage and retrieval mechanisms in the near future and yes it doesn’t know how to imagine or have creativity but it can solve complex problems so yes, it is intelligent.

But we have to be very careful here.?It may demonstrate some qualities that we have mentioned but it is not actually solving complex problems but only replicating what others have solved and that too in a different non intuitive manner.?If intuition, understanding, and creativity are to be assessed for it then consider an example (I think this was provided by Noam Chomsky) Imagine ChatGPT was present at the time of Newton, will it be able to say that Newton's laws are wrong considering the information available to it that time? ( For the uninitiated, Newton's laws are just an approximation of something deeper called General relativity and gravity is just a manifestation of curvature of space-time) will be it able to imagine different scenarios where the actual nature of gravity is revealed? will it be able to predict the existence of black holes by pure imagination and creativity? the answer is No, it won't, it will find the data available at that time and it will say Newton is right. Comparing it with real human beings e.g. Einstein who did all the above things and with imagination and creativity proved that Newton was wrong. Surely no one in the right mind would say Einstein was not intelligent so logically we can conclude in this regard that ChatGPT present at that time is not intelligent. Now you might see this coming but for those who didn’t here’s the kicker …?how can we say that ChatGPT present today is intelligent? isn’t it bound by the information we provide it today? Will it be able to solve problems by reconsidering fundamental laws of physics? The answer is obviously a big NO here.

Maybe the real answer is that we are limited when we try to quantize what intelligence is and maybe we don’t understand it well as it is something that is closely connected to consciousness Or maybe ChatGPT is intelligent but we are biased in having a self-centered view of intelligence as we are humans Or maybe ChatGPT is a small subsystem in a larger system to come in near future and when the whole system emerges then it will be like human intelligence?

No one knows for sure and the only way to get the answer is to wait.

Resources

https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chatgpt-ai.html

https://www.youtube.com/watch?v=NblR01hHK6U

https://www.youtube.com/watch?v=ck4RGeoHFko