That is, if AGI were an existing technology, how much would it be valued to?
Obviously it would depend on its efficiency, if it requires more than all the existing hardware to run it, it would be impossible to market.
This question is more about getting a general picture of the economy surrounding this technology.
Assuming a specific definition of AGI and that we implemented that AGI, what is its potential economical value?
Current investments in this research field are also useful data.
I will try to give some sense to this question.
Artificial general intelligence (AGI) is the hypothetical[1] intelligence of a machine that has the capacity to understand or learn any intellectual task that a human being can. It is a primary goal of some artificial intelligence research and a common topic in science fiction and futures studies. AGI can also be referred to as strong AI,[2][3][4] full AI,[5] or general intelligent action.[6] Some academic sources reserve the term "strong AI" for machines that can experience consciousness.
These are the first sentences on AGI on wikipedia (link), and the softest limit there is
[learn] any intellectual task that a human being can.
Even taking only this, it would mean that any AGI has infinite economic value. As soon as there is something that can learn any human task and has the speed of current GPUs/CPUs it could potentially immediatly replace every human in every task. There are certainly enough computers with CPUs and GPUs out there.
This question is still a little flawed because you not only have to constrain the definition of AGI but also how it would actually be implemented.
The economic value would be high indeed, as, combined with robotics, AGI would be able to replace all human workers. So:
Of course, there would also be the question of the cost of computation, the cost of the hardware & software required for AGI, and whether that cost is higher or lower than the cost of human labor. (My guess is biological machines such as humans would be cheaper, both in production and processing, until AGI leverages molecular computing via an inexpensive, ubiquitous substrate. Also worth noting that biological systems such as humans and canines may be more fault-tolerant, and more resilient in that they can persist even where the technological base collapses.)
Currently, cost of training even narrowly superintelligent Neural Networks which exceed humans at a single function is extremely high.
