Unlikeliness Of Anthropomorphic AGI

Feel free to add to or change any of the below. -- Anand

---

Some reasons why the goal of AI with human-equivalent General Intelligence may be anthropomorphic, while the goal of AI with Human-Similar General Intelligence, and thereafter human-surpassing General Intelligence, may be both self-consistent and realistic:

1. Because of the difference in “design” by natural and sexual selection, which eventually arrived at the subspecies Homo sapiens sapiens, once the numerous prerequisite adaptations on each level of organization were acquired, which were only selected for their survival or reproductive advantage, relative to deliberate design by human intelligence, which can handle quite a few things that evolution can't, e.g., simultaneous dependencies.
2. Because of the speed difference between thought occurring on 2GHz computing chips relative to thought occurring on ~200Hz neurons, which is an exponentially increasing speed difference, since computing chips quadruple in power, and reduce in cost by about one-half every year.
3. Because of the difference in new computing power being continually added to AI, relative to the fixed computing capacity of the human brain.
4. Because of the difference in a weakly self-improving process, e.g., knowledge-creating-better-knowledge or technology-creating-better-technology, neither of which directly change the individual human brains that allow for the occurrence of said improvement, relative to a strongly self-improving process, e.g., AI that favorably changes its own design, which in turn improves its overall intelligence, which in turn allows for more design improvements, which in turn achieves further improvements to its overall intelligence—a positive feedback loop of intelligence-enhancing-technology-that-further-enhances-intelligence.
5. Because evolutionary history offers little evidence that Homo sapiens sapiens represent a special bound on General Intelligence, and because we already have many examples of specialized autonomous intelligence achieving intelligent results that are similar to or better than our own. (See Artificial Specialized Intelligence Resources.)
6. Because if it’s possible for us to design an AI with Human-Similar General Intelligence, just as natural and sexual evolution has designed about 750 megabytes of DNA that resulted in human General Intelligence, then it also seems likely that that AI can in turn develop itself, by improving its own source code, which will have been something previously done—a problem previously solved—by the programmers, until that AI reaches human-surpassing General Intelligence. As explored elsewhere, an AI with Human-Similar General Intelligence, with human-surpassing competency in programming, and with numerous other cognitive advantages, will be able to use these interrelated qualities to quickly self-improve many aspects of its design, which may in turn improve its overall intelligence, which may in turn allow for more design improvements—a process that may open-endedly continue until human-surpassing General Intelligence is achieved. As Yudkowsky put it, "It would be the height of biological chauvinism to assert that, while it is possible for humans to build an AI and improve this AI to the point of roughly human-equivalent General Intelligence, this same human-equivalent AI can never master the (humanly solved) programming problem of making improvements to the AI's source code."

For more information, see "Levels of Organization in General Intelligence: 3: Seed AI: 3.3: Infrahumanity and transhumanity: "Human-Equivalence" as anthropocentrism"