Beyond the wildest dreams of our ancestors

Two months passed since I read the book “Life 3.0” by Max Tegmark, yet I’m still obsessed by its perspectives upon our past, current, and future world. The positions of certain molecules that give the letters and colors of this book, according to Tegmark’s words, had become my memory – long live enough to encode the information until its needed. Have I learned? I have certainly acquired complex representations to process new information, just as how a neural network fit. What have I learned, or in other words, what are the new information that “Life 3.0” brought? “Life 3.0" brought nothing new relative to specific information nodes, but instead built many intricate yet balancing bridges between existing ones. This blog is my personal reflection upon the amazement of the book, including my personal interpretations and new bridges built between existing nodes.

Our wings take us higher than the skies of our ancestors. These wings act like the quote on the main page of Google Scholars: they grow by “standing on the shoulders of giants”. It is our ancestors who nurtured those wings, eventually taking us higher than their initial assumptions. The Transformer paper came out as an alternative method to next token predictions, yet the authors would never come to imagine how this paper boosted artificial intelligence development by scales in years. Life tends to be fascinating, not by how things follow mathematical relations, but by the increasing entropy that brings unimaginable outcomes which are mostly positive in the eyes of humans.

Imagination is a scary thing. It gives men hope, incentivizes them, but at the same time destroys them. We human beings were gifted with regulating body systems, senses interacting with the environment, and all those factors which were investigated and derived into most of our science courses. Imagination, however, remains unexplained. We could replicate a circulating blood system or connect a mechanical arm to a nerve system, but even the most recent large language models cannot “imagine”. Our imagination, in most cases, tends to grow across generations. Ancestors living in ancient Mesopotamia wish the weather tomorrow to be rainy to wet the crops. Later generations learned to grow crops in water-efficient ways, and we now can know the upcoming weather with a few clicks on our phones. Once a tribe starts to imagine, the question turns from “can it evolve” to “how it evolves”, because imagination is the fire root in a vast grass plain. After thousands of years of dreaming, imagining, and pursuing, humans now stand at the top of the hierarchy.

So far, imagination seems to only incentivize men and give them hope, which is not something worth being aware of. How can imagination come to destroy humanity? It is humans who are trying to validate this paradoxical question themselves.

From horses to steam engines, Nokia to iPhone, we tried infinite ways to improve our living standards. One critical component indicating this standard is efficiency. One such example is the replacement of mailman with email, from physical to digital. The path we laid in the past decades to improve efficiency is the internet, but in most recent years, we turned our steering wheels to acquiring artificial intelligence into our daily lives. ChatGPT came out in December 2022, and it became the fastest platform to gain 100 million users, achieving this within a month. The company behind this, OpenAI, made significant investments in it, eventually releasing newer and newer versions of this seemingly improving our life tool. Isn’t everything sounding good? We are looking forward to GPT-5.0! There comes the scary part of imagination: we imagine the future, yet we cannot build the same thing, just like one can never remember the beginning of a dream. The authors of the Transformer paper were expecting a method for the next token prediction, but the effect turned out to be far bigger. In past centuries, a slight misalignment between the product and our criterion is negligible, because the product won’t achieve much damage with its tiny capability. Artificial intelligence is different. It is only a matter of time before AI products surpass human-level intelligence if we continue to pour in the same time and investments, not to mention that we are investing more and more. And once AI achieves human-level intelligence, it is not a toy anymore, because one slight misalignment could result in devastation. Just like how we wander back to dreams, we often magnify the positive effect of our imaginations yet reduce the possibility of a black swan event.

If imagination is a gifted miracle to humanity, how can we possibly replicate it in Silicon-based life? If these life forms cannot be imagined, how will they achieve human-level intelligence? This relationship between intelligence and imagination might seem paradoxical, but that’s only because our modern technology hasn't reached that stage yet. Imagination, or in a more general term: consciousness, can be developed in artificial intelligence. A crucial underlying justification for this assumption is physics. In the end, we all are made of physical atoms. The difference between an ice cube and water lies in the way the atoms are put into. In the same manner, what differentiates our thinking minds from a Koala’s is the patterns for the arrangements of atoms. Therefore, “optimistically” speaking, we will discover the secret of consciousness someday. That day might come very soon, as our exponential technical development mirrors Moore’s Law.

After all these assumptions, we must come to a steady fact – AI is significantly more efficient than human beings. On one hand, this is good news, or else we won’t implement it into our daily lives. However, if AI does achieve human-level intelligence, then the entire hierarchy will be shifted, and Silicon-based life forms will arrive at the top. Imagine you are competing with a robot that is as intelligent as you but can compute factors a trillion times faster than you. Sounds scary? This is not the scary part, however, because the robots will likely follow our instructions as they are programmed to do. What brings controversy is an intelligent explosion.

An intelligent explosion is when an artificial intelligence starts improving itself repeatedly in many iterations. It works like how human researchers trained AlphaGo to become the world’s best Go player. They did it by letting it play against itself thousands of times, where it improves every time. Once a human intelligence level AI starts improving in iterations, it becomes smarter and eventually will reach superintelligence. A superintelligence AI will still follow human instructions, as it was programmed to, but it will have its interpretations which may result in catastrophe. Imagine you gave your superintelligence AI assistance a goal: make the world more sustainable. It will still strictly follow this direction, but it may start to wonder about human’s role in the situation. Eventually, it will come to realize that it is we humans who are the greatest obstacles in this goal-oriented behavior. Humans easily dominated ants because of the intelligence gap, and it is likely the same for the superintelligence AI to eliminate us. If we want to pursue our dreams on the correct path, to utilize artificial intelligence in an aligned fashion, we must evolve a new logical mechanism, one that highlights the future.

At the end of the day, we are all sparkling crystals of an amazingly beautiful universe. We are the generation shaping the course of civilization, and even if superintelligence AI broke out and built a cosmos empire one day, we have at least pursued our imaginations. We are living beyond the wildest dreams of our ancestors, and it is up to us to shape ours.