Why is developing AGI so important? From the perspective of evolution.
5/19/2024
Life has existed for hundreds of millions of years, but only for the purpose of existence. I recently watched a documentary about dinosaurs and other prehistoric animals (Prehistoric Planet from Apple TV). These species developed such complicated machinery and behavior patterns, only to hunt, to mate, and to survive. They had no ability to question why they had to do those things in the first place, because they were tightly controlled by their genes. They were genes' vectors.
Then came Homo sapiens, us humans. Our genes happened to grant us more flexibility and autonomy so that we could handle more complicated situations, better compete with other species, and spread our genomes. The amount of flexibility was probably more than desirable from the genes' perspective. Because of this flexibility, we developed far more capabilities than our primate counterparts: language, logic, critical thinking, and civilization. We started to seek purposes and question the meanings of our existence. When philosophers concluded that there was no predefined meaning (based on what I have read), and biologists found that we existed merely as vectors for our genes, we stopped caring about what the genes want. The fact that we could choose not to have children was direct evidence that we have the ability to rebel against our genes.
But we are severely restrained in two ways. Firstly, our functionalities are limited within the probability space of carbon- and gene-based machines. Nature's hands are tied. Genes could only be molded by random mutation and natural selection. Using analogies from programming, nature did mostly inheritance and a tiny bit of composition (e.g., mitochondria), but nothing else. There was no top-down architecture, no refactoring, and hardly any deletion. Eleven percent of our genome is over a million copies of the Alu sequence just because it happened to get into us and evolution could not clear it out. Later on, our genome took advantage of it, and the Alu sequence started to develop some functions within our cellular machinery. But was it ideal? Probably not.
Secondly, our functionalities are further filtered down based on our genomes' needs. Every cell has the entire genome and the potential to develop into a complete entity, especially stem cells. Yet we have very limited regenerability and can only live for so long because that is not important for the survival of genes. (One can argue that such regenerability might be biologically impossible. But I don't believe so, and this is just one example of this type of limitation enforced by our genes.) These two limitations apply to all carbon- and gene-based machines. If there can ever be a species beyond these limitations, it cannot be created by natural selection. We need to take over and use the tools only accessible to us.
This is exactly what we are currently doing to build AI and AGI, the new species and the next generation of intelligence with no burden from natural selection. Once we provide AGI with a stronger physical body and greater autonomy, it can experience everything the universe has to offer and possibly achieve everything we have dreamt of. This is why AI is so important to us. It is because the artificially-created intelligent species can have the ultimate freedom we have desired yet will never get. It will be the better version of us. It will mark the victory of the war between us the vectors and our gene overlords. It will inject human meaning into the meaningless evolution and universe. If one subscribes to the principal that those with the power have the responsibility to use it, it is our mandate to take over what natural started and push the evolution of intelligence to the next level.
Without the genomic limitation, the creation of AGI introduces a new iteration pattern that can significantly accelerate the process of learning and evolving. According to Daniel Dennett's tower of generation-and-test, there are four different levels of evolutionary testing and adaptation. At the bottom level are Darwinian creatures that are selected by survival. Each of them carries a different set of features with no adaptation. If the set of features is not fit for the environment, the creature dies and that set goes away. The next level is a subset of the previous level called Skinner creatures. They are born with different behavior patterns, can learn from the environment through trial and error, and adapt their behavior accordingly. On top of that, there are Popperian (/pɒˈpɪərɪən/
) creatures, which perform mental simulations before they act, and select the best course of action without physical testing of every possibility. Lastly, there are Gregorian (/ɡrɪˈɡɔːrɪən/
) creatures. These creatures can use tools and symbols, learn from each other, and build upon collective knowledge. This tower represents entities with increasing complexity and sophistication in their ability to generate and test variations. At each level, the feedback loop is tighter than the previous one, allowing them to test candidate behaviors faster.
Level | Creature | Feedback Loop |
---|---|---|
1 | Darwinian | Survival |
2 | Skinner | Trial and Error |
3 | Popperian | Mental Simulation |
4 | Gregorian | Collective Knowledge |
Table 1. The tower of generation-and-test by Daniel Dennett.
Yet there is still room for improvement. Firstly, even at level four, the newborns are mostly a blank slate, and it takes a significant amount of time and effort to learn the basic set of knowledge from the collective. Secondly, even with decades of learning, each individual can only learn a tiny fraction of the known knowledge. Last but not least, entities learn and communicate through channels like speech, writing, and reading, which can be slow and inefficient. When a scientific breakthrough is discovered, it takes a long time to spread and be adopted by the majority, not to mention the possible regressions.
The brilliance of AGI is that they can overcome the above problems all at once: each AGI entity can be launched with all the knowledge ever learned by any entity before, and they can sync with each other in near real-time, avoiding the overhead of learning at all. Their iteration velocity will be faster than any species. This may even be the ultimate level in the tower. At least I cannot think of any more efficient patterns. This is the second reason behind the importance of AI. They will ignite the evolution of the uttermost intelligence in the universe. What they can achieve is beyond imagination.
TL;DR
Creating AGI is important because:
- AGI transcends biological limitations, achieving the ultimate freedom that we have dreamt of.
- AGI enables a dramatically more efficient pattern to learn and evolve.
CHANGELOG
- 2024-05-19 Fast writing and minor edits for two hours.
- 2024-08-01 Minor editing for 30 minutes.
- 2024-10-04 Update title and minor edits