Some Notes on Artificial General Intelligence

Instrumentalism in Quantum Mechanics is a perspective where we consider the various equations and mathematical formulations strictly as a means to predicting physical outcomes.  If the formulae work well, great!  We don't really care about the philosophical meaning of the equations, or the ontological implications (uncertainty, duality, the observer effect, etc.).  We have a series of physical measurements, we apply certain equations, we get predicted measurements in the future, and we are happy.

In a similar manner, debates about consciousness, self-awareness, emergent properties, Chinese room, etc. in computational intelligence are fine as conversational pieces.  But that does not diminish one bit the extraordinary capabilities shown by LLMs in responding to complex inputs.  Those capabilities in LLMs are usable whether or not we endow them with "real" intelligence or "understanding".

For example, see this sample problem from Humanity's Last Exam (a benchmark for LLMs):

Hummingbirds within Apodiformes uniquely have a bilaterally paired oval bone, a sesamoid embedded in the caudolateral portion of the expanded, cruciate aponeurosis of insertion of m. depressor caudae. How many paired tendons are supported by this sesamoid bone? Answer with a number.

 ...

We will see a rapid evolution in AI capabilities.  We are still in the very early stages of this new technology and for many years and decades, the capabilities will increase non-linearly.  Capabilities will evolve in various dimensions:

• Computational advances: It is very likely that the current foundations of AI in terms of context and token-input-output neural networks will morph into something very different that requires a very different quality or quantity of computational horsepower.
  
  
• Multi-modal expansion: For computation to excel in the real world, input processing from high-fidelity environments (such as visual or tactile or auditory) will become more tractable.  Multi-modal output will also become increasingly realistic.  Augmented reality will become as good as lucid dreaming.
  
  
• Agency: After a few initial hiccups, AI agents will be increasingly left alone, with very little oversight except from AI supervisors, to inspect, decide and take action.
  
  
• Price and Availability: AI will become cheaper and cheaper, eventually uncensored: almost omnipotent AND almost free.  What that will mean for human consumption patterns is scary.
  

...

AI will lead to economic devastation for those who offer repetition and basic pattern-matching as their value-add to human society.  Many regions of the world will become empty of humans and will be managed by AI.  Flourishing as a human being, having power, commanding AI, and reproducing will be a luxury that only a few will experience.  Vastly fewer in percentage terms.