ChatGPT is not that smart… but it can still ruin us.

A general-purpose, all encompassing, intelligent and self-aware Artificial Intelligence has been the stuff of dreams and nightmares of humanity for generations. We have been anticipating its emergence so much so that we have given it a name way before its birth: Artificial General Intelligence.

But what is it that makes us so excited and scared of something that does not even exist. How come we are so afraid of something that we will create ourselves? And maybe most importantly… Is GPT-4 an AGI (Artificial General Intelligence) or does it even matter whether it is an AGI or not?

AGI is hard to define and even harder to measure. It has been traditionally defined by being able to perform any task that humans can. Seems like a simple enough definition. But today, even the CEO of OpenAI — which is the company that makes ChatGPT — is unable to say whether ChatGPT is AGI or not. By the Wikipedia definition, it almost certainly is. Strictly speaking, no one knows.

Should we be so afraid of AGI?

We as humans put ourselves on such a high pedestal that we measure everything against ourselves. Therefore, the Terminators, the HAL9000s, the Moons all told us to be scared to death of any machine that can talk, act or feel like us. We were scared of the machines controlling every aspect of our lives because they can finally think and decide like us.

The novelty is not imagining an intelligence that can be smart like we are or can steal our jobs. The novelty is that we have found a way to exploit the cumulative total knowledge of humanity. Which means that these technologies will be developed at a pace that humanity has never before seen. At no time in the history of mankind, have we witnessed so many breakthroughs in such a short time in any other field. We knew for a while that the rhythm of technological milestones ticked at a higher BPM each passing year thanks to Moore’s Law but we were not prepared for a computer software that could pass the Bar Exam.

With unimaginable foresight, Alan Turing has foreseen the coming of the age of AI in 1950s and devised a blind test that can define an AI as AGI when it can fool a group of humans that it was one of them. We have already passed that point by any measure. ChatGPT is so advanced that a Turing Test is irrelevant as a bar to pass. (Pun intended) According to some critics, ChatGPT can never pass the Turing Test since it will readily admit that it is a computer program even before it is asked. So GPT is a technology that rendered the most visionary AI test irrelevant. Critics say that it is just a dumb machine that spews out best guesses based on probabilistic algorithms.

So is it really smart?

Probably not.

Should we be afraid of it?

Yes.

But not for the reasons that HAL9000 or Terminator conditioned us to do.

Whether ChatGPT is AGI or not does not matter one bit for humanity. What matters is that it can pass as a human in longer interactions, it is infinitely scalable and it learns by retaining information.

A smart killer robot armed with lasers seems obviously alarming to everyone at first glance. But the combination of rogue human-like chatbots that can run amok on Social Media is nightmarish only to a few. Having so many of these agents swarming whole platforms in order to sway political views or to divide nations is beyond the contents of any nightmare. Ever got stuck on a so-called “Captcha” that asks you to prove that you are a human to a machine? Now imagine a machine that can fake being a human easily.

When we finally realize that ChatGPT-like systems can pass as a human in written and sometimes even voice-based media, it will probably be too late to do anything about it. This is the frightful but probable truth that so many experts skirt around or downplay.

Today’s world is too fragile because it is too polarized in almost every imaginable way. Sam Altman said in a podcast that no one wants to see the world burn. Hopefully so. Unfortunately, there are too many actors in this spherical spaceship of ours who cannot follow feedback loops far enough to prevent the world from burning.

Especially in this very risky and sensitive topic, I call everyone to “Innovate Carefully”. Let’s hope that the ChatGPT and all its clones have the decency and foresight enough to apply protective layers to their solutions so that using them for mass-manipulation becomes too difficult if not impossible.