Artificial Generalized Intelligence (AGI) and Evolution
Prelude:
To design an AGI agent, we need a model which will process information gathered by it’s sensors in a hierarchical manner (can process structural and temporal data in different resolutions and frequencies). This model is then expected to predict the next state of the world model (environment around the AGI agent) and concatenate the prediction with next inputs data. The model should have capable structure to process these information and must be scalable to extension of newer data from additional sensors or appended worlds/environments. The model will probably use a reward (extrinsic, intrinsic, social and may be something else?) maximization approach and it’s performance is to be evaluated on some sort of task (Turing test, Park Test). This model is to be initialized with few manual instructions (like follow human face, imitate other human) and these instruction may be overwritten in future, to extend the exploration scope. Also, a bootstrap-like agent should be provided with (as an library of actions from mother-baby pairs) to simulate the motherese.
Why am I interested into these fields:
Next into my own thoughts and research interests. It has been my dream to combine learning algorithms from CV and NLP together to create a generalized knowledge acquisition system, just like a human brain. One of my favorite personalities is David Attenborough, and I watch/re-watch his documentaries quite frequently. Specially “Life on Earth” is my favorite as it goes through the evolution of life on earth from amino acid to human being. I think that the breakthrough in AGI will come from some sort of inspiration from the way nature created life-forms. Observing (and probably mimicking) nature on how she solves the problem of creating sustainable life forms in millions of different ways (different species) to adapt to the ever-changing environments, might be a very crucial step in developing AGI. To create something like a human brains, nature took billions of years of trial and error (through natural selection) and I get really fascinated by the thought that may be someday a new species will arrive to outperform human species. So, the same algorithm (natural selection) that created single-cell life from complex-chemicals, can give rise to more intelligent life-forms than human being.
Ultimately, is it not the quest of survival? I mean, if we set the goal of an agent (cell) to survive in its environment and as the agent find ways to do so, things like reproduction, cell collaboration (build multi-cell bodies and organs specialized for specific tasks), intelligence, forming colony (collection of multi-cell agents), inventing methods (sign language, body language, full-fledged modern languages) to communicate among the colony members, invent science to improve the survival rate and so on should be formed as a by-product.
Final remarks
Finally, I would like to end with some philosophical sort of remarks. After all, it is Doctor of Philosophy not Doctor of Computer Science!
Probably, the quest is to find what is life and why is life? May be, what we ought to ponder upon, is the purpose of life and it’s evolution on earth in the first place. Why go through this billions of years of optimization to create intelligent and sustainable life-form (all the species)? What is the objective of such experiment? If we can create such a mechanism can we give birth to an universe? Pardon my digress Professor, but I just like to think about these sometimes and search for some universal connections.