The Promise of AGIs that Grow Themselves

It is indeed perplexing how a multi-cellular organism (like us) is able to create itself from a single cell. It’s curiously interesting to think about how cells are able to join together to create multicellular organisms where every cell (i.e. every cell lineage) is needed to create the final organism.

A cell divides 50 times, each division leading to two specializations of itself. Cell divisions eventually lead to an organism in which there are 2⁵⁰ single-celled organisms composed of over 1 quadrillion cells.

Constructing itself not through a single centralized coordinator, but rather a multitude of cells reading their own copies of DNA. The different cells within an organism read their own copies of the DNA and proceed to create a multitude of cells with their own agendas and tasks to perform independently from the others. This is how distributed autonomous organizations work. The interesting implication behind this fact is that there is no single point of control or planning. Instead, each cell can proceed without coordinating with the rest, as long as the DNA to perform a certain task is copied and understood by the cell itself. Independently of one another, thousands of cells of the organism can perform actions that will result in the desired goal.

Each cell reads its epigenome and senses it’s immediate surroundings to determine the appropriate goal in the appropriate context. It’s a massively parallel process. But a cell’s immediate surroundings are extremely dynamic. The cell is socially woven into a complex web. This extracellular matrix is constantly moving and changing, the cell must, therefore, be transforming in response. A cell must be effector and executor at the same time. It must respond to the changing world, and by changing itself it changes its environment.

In order to attain the goal of a complete organism, its embryonic cells must completely crawl through a constantly shifting maze. But to get to the end of the maze, the members of this collective must be able to work together. Every single cell is responsible for performing its own error-correction to ensure the integrity of the whole. Chaos can easily break-up this collaborative process. The integrity of the whole only depends on the integrity of the cells. What happens in one cell will have an impact on the whole — like a big drop of water that makes all the other drops overflow. Small changes in the environment can snowball if they are not regulated in time.

How is it that each self-transformation preserves the structure of the whole? We see that each transformation maintains a symmetry to the whole system, thus preserving the structure as a whole. This means that the structure is stable, that it is stable with respect to certain interactions occurring at every scale. To say that it is stable means that it pursues a coherent orbit under such interactions. Given a complex adaptive system, the main emergences of the system are determined by the fact that the interactions of the system promote certain interactions that amplify and maintain the relevant symmetries of the whole. Whatever is significant to the system is the way it adapts and stabilizes its structure in the interactions with its environment.

How does it preserve structure in a decentralized manner? Indeed there must be some sophisticated cognition taking place in the morphogenesis of any animal. What maintains conformation in the absence of centralized consciousness? Perhaps the similarity of each epigenome in a cell’s vicinity provides the context for coordinated behavior. It is conceivable that the similarity of epigenomes creates an energetic tendency, and each cell behaves in a manner that is compatible with its adjacent cells. The local geometry is in an attractor landscape that determines the structure and progression of the local region.

It, therefore, is not a leap to imagine that the brain itself also uses an analogous self-preserving transformation engine to also construct itself. In this way, we may understand not only how the brain forms but also how understanding itself the most important component of thought processes — may be made to develop organically, from the inside out. Only in this way can we account for the way the brain forms ourselves, stores our understanding, make it available to us and, when that understanding has been used, then makes a part of our memory again, so that it can be called upon again and reused.

Not only does it do so to construct the neurons and its neuroglia, it probably employs the same mechanism in the virtual space of cognition. That is, our mind creates itself from within. Indeed minds are themselves all creations of their own.

This notion of self-referentiality can easily fall into the trap of infinite regress and just blow up. But it simply doesn’t, it just does what it needs to get done. If it was not for this strange loop, it would be impossible to conceive of the mind as something real or concrete, in the same way, I can’t conceive of a universe without relation to my mind. This relationship and imagining of self-referentiality is evidence of the realness and concreteness of the mind.

Therefore any model of general intelligence must take into account the same principles as what makes life possible. At a minimum, this includes self-referentiality, autopoiesis, homeostasis, and learning. These principles are common to all-natural systems, so any non-biological model of general intelligence would have to reproduce these principles. Neglecting these principles leads to the artificial intelligence community making slow progress toward developing a model of general intelligence or self-improving autonomous systems.

At a fundamental level, morphogenesis involves a learning process. An organism can only create itself using an adaptive error-correction process that shares a commonality with inference and learning.

The very thesis that cells are simple mechanistic automation is absurd. There is simply not enough intrinsic competence to make self-creation a possibility.

This implies an impossibility in creating an alternative general intelligence without a biological substrate. This is the easy problem. The hard problem is to use existing silicon and perhaps discover a slice of functionality that is sufficient enough to lead to useful high-level intelligence. In short, remove all the extra-baggage and begin with a minimal set required for cognition.

There are several observations that may imply that this may be possible. (1) The intelligence to construct oneself is already contained in a single tiny cell. (2) high-level cognition may be sufficiently decoupled from physical structure such that this is not a physical problem but rather a virtual one.

To achieve AGI requires simply the virtualization of the same process that allows life and brains to recreate itself.

Disclaimer: Parts of this text were generated by GPT-3.

Related posts:

Shopping UK will automatically categorise your list into aisles. This makes it easy to find things in the supermarket. The app has a built-in list of over 20,000 items, and it does a pretty good job…

The cost of living is high, and given the current economic climate, it’s more competitive than ever to find a stable career. I work hard, learn quickly, am well-rounded in my industry, and can adjust…

For the majority of teenagers in India, especially when you are from a science background in class 12, their real-life journey starts from JEE coaching centres. The most common advice that an average…