Why I’m Skeptical About ChatGPT and AIs in General

About 20 years ago, I encountered a chatbot that blew my mind.

We used to communicate with my friends extensively through text-based chat systems such as IRC. On one of the university servers, there was a chatbot capable of fluently composing messages and answering questions.

While it was significantly less advanced than ChatGPT, it was still amazing, especially considering the technological landscape of the time.

However, what made it remarkable was its reliance on a very basic Markov Model.

Basically, it had knowledge of word sequences and their probabilities, recognizing which words typically followed others and at what likelihood.

This concept is remarkably straightforward, yet the final outcome was still phenomenal.

So this was my initial and original jaw-dropping moment; consequently, when I first encountered ChatGPT, my subsequent jaw-dropping moment wasn’t as profound.

Despite ChatGPT’s superiority in every aspect, being significantly more advanced and useful, I knew that I’ve already seen something similar 20 years ago.

No groundbreaking scientific discovery

ChatGPT represents an engineering achievement rather than a novel scientific breakthrough. Neural networks, the underlying technology, were conceptualized and developed in the 1940s and 1950s.

No one discovered anything new about human intelligence or its underlying mechanisms.

ChatGPT still operates on the same foundational principles established over 80 years ago.

Previously, we only had the concept; but now, we also possess the data and computing power. This is what enabled us to build an AI, rather than a groundbreaking scientific discovery.

AI has already happened before, multiple times

AI goes through cycles: first, there’s excitement, then comes a quieter period when optimism fades. This has happened several times before.

What follows the hype phase is known as an AI winter.

The first AI winter occurred in the 1970s and the second one in the 1990s.

At university, I recall everyone being super excited about the Artificial Intelligence class at first, but later, they became disappointed when they realised that all we were going to learn was how an SCV finds the Vespene gas without colliding with obstacles. No intelligent androids or robots with human-like cognition, unfortunately.

This was a Starcraft reference. In computer games, they utilize pathfinding algorithms (such as A star), which were among the biggest achievements of these early AI hypes.

Unrealistic expectations

I would be happy to be wrong about this, and I wouldn’t mind if a robot ended up taking my job, given that I’m (early) retiring soon.

However, I don’t believe this scenario is likely to unfold. We don’t yet possess a general AI capable of performing any task. We have one for example, that can generate sentences, comprehend context, and respond to questions; and a completely separate one capable of (semi) autonomous driving.

We’re dumb without our tools

We use certain tools to augment our physical strengths and capabilities.

Without a hammer, we wouldn’t be able to drive nails. A baseball bat extends our reach and amplifies our strength, allowing us to hit balls farther.

Similarly, there are tools to extend our minds and enhance our cognitive abilities.

The most basic one is called pen and paper. Without these tools, we wouldn’t be able to solve some of the most basic mathematical problems.

Perhaps Pythagoras would have never discovered his theory if he couldn’t draw and visualize geometrical objects in the sand.

Douglas Engelbart’s vision was that a computer is the next-generation of the pen and paper, enabling our mental power to be amplified even further.

We’re at the next stage, where AI can further enhance our intellectual capacity, much like how the pen and paper, and later computers did.

But just as neither the pen and paper nor computers took our jobs (in fact, they created new ones), I don’t expect AI to take our jobs either.

Following every major technological breakthrough, certain aspects of our work become notably simplified. This enables us to solve increasingly complex challenges, pushing the boundaries of innovation further.

But human innovation is still essential; without it, AI would remain stuck at its current level.