7 Paths Toward Superintelligence
Superintelligence is tentatively defined as “any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest” [1, p. 26]. In other words, an entity whose intelligence vastly exceeds a highly intelligent person of today would be called superintelligent.
Nick Bostrom, the author of “Superintelligence: Paths, Dangers, Strategies” book, considers several possible paths toward the creation of superintelligence, including advances in current AI technologies, the development of brain-computer interfaces and neuro-implants, brain emulation, and more; in this blog post, I briefly explore five major possibilities laid out in the Bostrom’s “superintelligence” book and then I discuss two more possibilities that are not directly discussed in that book, namely expanding human senses and the Conscious Empathic AI paradigm.
Is Superintelligence Viable?
Relying on the evolution of intelligence and the history of human technology advancements, Bostrom paints a picture that the emergence of superintelligence is not only viable but inevitable. He argues that the existence of humans as highly intelligent beings is a relatively new phenomenon compared to millions of years of natural evolution. Humans also demonstrate a considerable jump in intelligence compared to their closest animal relatives. According to what we understand from the history of life on earth, humans represent an explosion of intelligence on earth. Further, looking at the history of human invention and technological advancements, we can also identify similar explosive and exponential patterns. Humans have been hunter-gatherers for most of their 200 thousand years of existence. They learned how to farm and moved to live in bigger groups roughly around 15 to 10 thousand years ago [2]. This move demonstrates a jump in human intelligence and cognitive capabilities as learning to farm and domesticate animals are much more cognitively complex tasks compared to routines of hunting and gathering, as I have discussed here. After the invention of the wheel and basic tools supporting the farming lifestyle, there has not been any notable and lasting invention in human history until the dawn of the industrial age around 250 years ago. The industrial age brought many new scientific advances and technological inventions, including the steam engine and electricity that changed human life. We can identify another jump in technological advancement with the advent of computers around 75 years ago and then 50 years after that, the advent of the internet and the proliferation of mobile phones. Only around 25 years after the very first versions of dumb mobile phones, humans invented smartphones, and the explosion of AI-enabled software applications took place. If we plot these jumps and advances, we will arrive at an exponential growth pattern as opposed to a linear growth pattern. Assuming that this pattern shall continue, then the time between each novel human invention and the jump in intelligence should decrease over time. Following this exponential growth in intelligence and technological advancement, one can arrive at a similar conclusion that superintelligence is not only viable, but it is inevitable.
So if it is a certain thing, how will it happen? We may not know how it will happen for sure until it happens. However, we can have some educated speculations, which I am going to outline next.
Path 1: Artificial Intelligence & Machine Learning
In order to achieve general intelligence, a system must be able to learn. And the machine learning paradigm is one way of achieving such a goal. We know that an evolutionary process devoid of direction from a human being (blind evolution) has resulted in human-level intelligence. Hence an artificial intelligence designed and guided by humans should theoretically be able to not only achieve the same level of intelligence but exceed it to the degree that it can understand its own workings to engineer new algorithms and infrastructures to increase its cognitive performance. This ability would put AI well into a path of becoming superintelligent.
Path 2: Whole Brain Emulation
This path relies on closely imitating human brain structure and working where an intelligent system would be created by scanning and meticulously modeling the biological brain’s computational architecture and processes. The major challenge here is how to scan a living human brain’s workings in the level of detail that is needed for whole-brain emulation. Many technological advancements are required for us to achieve a complete whole-brain emulation and then improve emulated brain functioning and capabilities to the level of superintelligence.
Path 3: Biological Cognition
This path focuses on enhancing the functioning of the current human brain to achieve greater-than-current-human intelligence. This is basically a reconfiguration of the old idea of improving lineage and capabilities through selective breeding and feeding. Current technology helps us optimize infant feeding schedules and contents to enhance brain development and functioning in adulthood. Coupled with optimized education, this method should result in more intelligent humans in the future. Once such humans interbreed, then we are going to have smarter generations which in time should lead to superintelligent beings.
Path 4: Brain-computer Interfaces
Brain-computer interfaces, mostly through implants, could enable humans to integrate functionalities that are now available to digital computers, such as perfect recall, fast and accurate calculations, and broad-band and minimal error data transmissions. This also promises superhumans and superintelligence in the future. Elon Musk’s neuro link corporation, for example, is focusing on this path, but there are still many health and safety issues to be resolved before such technologies and interfaces can flourish.
Path 5: Networks and Organizations
The final path that Nick Bostrom suggests in his book is based on linking individual human minds with each other to create a much bigger intelligent organism and harness the power of collective computing that connected human brains provide. Humans have been enjoying wondrous technological advancement precisely through this mechanism, including the advent of language itself, which enabled humans to communicate and be connected to each other more efficiently than other species. The idea is further to enhance this connectivity between individual human minds through technology and achieve a form of supper intelligence called “collective superintelligence.”
Path 6: Expanding Human Umwelt and Creating New Senses
This path, in its essence, relies on human-computer interfaces, but it fundamentally differs from path 4 that it does not aim to integrate digital technology capabilities such as perfect recall with human cognition. It assumes that the human brain is a very capable computing organism that can learn to decipher any stream of data and create new realities without much direction from “humans.” The ultimate idea is not only to expand current human senses but also to create new senses. It is almost impossible to imagine what a new sense, in addition to the current five senses, might be, but we can imagine extensions of these senses. For example, this approach suggests that it is the possibility for humans to see infrared or ultraviolet or smell like a dog by introducing implants that can sense visual and smell data beyond what normal human sensors (eyes and ears) can do. Such implants then will send their data to the brain and over time brain learns to create a representation of such data and expand our senses. The idea that the brain can learn patterns from new streams of data and train itself is already tested and proven to be working. It is tested through sensory channel conversion and expansions where sound data are converted into a pattern of vibrations that are sensed through a deaf person’s skin. Over a few weeks of experiments and training, deaf subjects’ brains had learned to decipher the data and gave deaf people a means of hearing but through the sense of touch. You can learn more about these experiments by watching this Ted talk from the originator of this idea. He does not claim that his idea and work would lead to superintelligence, but I argue that it can lead to superintelligence employing a similar logic in favor of brain-computer implants suggested by Bostrom.
Path 7: Conscious Empathic Artificial Intelligence
This last path also can be seen as a variation of the first path, but it differs from the first path in its most fundamental goal. The first path tries to achieve superintelligence by creating a Thinking Machine as proposed by Allen Turing. The AI field is mostly influenced by the overarching goal of creating a machine that can comprehensively pass the Turing Test and becomes linguistically indistinguishable from humans [3]. There also have been calls to expand the Turning Test, hence expanding the AI development overarching goal to include functional indistinguishability from humans as well. As a result, most AI advancement efforts are focused on the relationship between humans and machines and creating machines that are indistinguishable from humans as opposed to the relationship between two machines and creating unique machines.
The Conscious Empathic AI paradigm, on the other hand, introduces an alternative path toward Turing’s Thinking Machine that focuses on machines co-creating a novel language among themselves [4]. Assuming that Turning’s Thinking Machine, similar to only confirmed thinking species that we know (humans), is going to be conscious, then achieving conscious AI would also enable AI agents to eventually pass the Turing Test if they wish. This paradigm argues machines that are capable of co-creating a novel language among themselves and then using such language to repeatedly achieve novel (not-programmed) shared goals would essentially be Thinking Machines without necessarily achieving language indistinguishability from humans. The conscious Empathic AI paradigm introduces an alternative goal for AI researchers and developers that calls for the modification of existing designs and goals to enable machines to co-create their own language under the watchful eyes of humans. Such machines would also be able to understand their own workings to engineer new algorithms and infrastructures to increase their cognitive performance!
References:
1- Bostrom, N. (2015), Superintelligence: Paths, Dangers, Strategies. New York, NY: Oxford University Press.
2- Ryan, C., & Jethá, C. (2010). Sex at dawn: The prehistoric origins of modern sexuality. New York, NY: Harper.
3-Bringsjord, S. and Govindarajulu, N. S. (2018), “Artificial Intelligence,” in The Stanford Encyclopedia of Philosophy, E. N. Zalta, ed. Stanford, CA: Metaphysics Research Lab, Stanford University
4- Esmaeilzadeh, H., & Vaezi, R. (2022). Conscious Empathic AI in Service. Journal of Service Research, 25(4), 549–564.
— The decorative picture is generated using Stability AI, Stable Diffusion Text to Image AI
