Synthetic Superintelligence (SSI)
Synthetic Superintelligence (SSI) refers to the development of intelligence which will follow the development of Artificial General Intelligence (AGI). SSI systems will evolve out of AGI and will possess a degree of intelligence that far surpasses that of the smartest human minds.
Note on terminology: “SSI” is sometimes called “Artificial Superintelligence” (ASI), or just “Superintelligence” (SI). The term “SSI” is preferable to “ASI” since “artificial” implies “man-made”, whereas SSI is “synthetic” because AGI will help create SSI (via processes such as “recursive self-improvement”, discussed below).
There are several ways by which AGI will evolve to become SSI.
It goes without saying that Moore’s law (broadly construed) for transistors, computer memory, and disk storage will continue to make computers faster with greater addressable memory, such that at some point computers will exceed the processing power of the human brain. Computers already can process much faster than neurons can, but the continued advances in speed enable computers to match what the brain can do with parallel processing.
Also, there are new computing paradigms on the horizon that will radically increase the speed (and reduce the power consumption) of computers, making it possible for computers to truly emulate the human brain. There are two main paradigms under development:
Neuromorphic computer hardware is designed to emulate the structure and function of brain neurons more closely than conventional computers can. The main approach is to build massive parallelism and connectivity right into the chips, and thus better emulate how the human brain works.
One example is IBM’s TrueNorth chip , which came out of the DARPA SyNAPSE project. This chip emulates one million neurons and 256 million synapses.
Another example is the SpiNNaker computing platform  which is based on “spiking neural networks”, and is a component of the European Union’s Human Brain Project.
Quantum hardware will yield much faster computing speeds, creating a “quantum leap” in performance. Quantum computing is in its infancy, but billions of dollars are being poured into research and development by government and business, because of their enormous potential. Currently one quantum computer (from D-Wave Systems) is on the market and is being used by NASA and Google. Also, Google and Microsoft are developing their own quantum computers. 
Needless to say, the exorbitant increase in performance will greatly aid in the quest to have the power of thousands of human brains in a single computer.
Obviously computer scientists will continue to work to improve algorithms such as “deep learning” and “graphical models”, so that computers can learn in the same “general” ways that humans do, and take advantage of the hardware scaling (in performance) to increasingly allow the nascent SSI to acquire all knowledge on earth. And work will be done to “perfect” capabilities like natural language processing (NLP) so that the computer understands all that it sees, hears, and reads just as well as humans do.
But it is expected that the computers themselves will become capable of “recursive self-improvement”.
The term “recursive self-improvement” refers to the ability of an AGI system to increase its own efficiency, performance, and intelligence by re-programming some of its own software and hardware, and also designing new software and hardware for itself. The process is “recursive” since each generation of the system can improve on the structure and abilities of the previous generation.
Recursive self-improvement is also called “Seed AI” because the initial “seed” of software and hardware (both created by humans) can grow into a Synthetic Super-intelligence (SSI) whose capabilities are largely unforeseen at this time.
In computer software, the most promising avenues of such self-improvement are “deep learning” approaches which are beginning to emulate how humans actually learn new skills (e.g., software like Numenta NuPIC ), and evolutionary programming (a.k.a. “genetic programming”) which generates and modifies computer programs in much the same way as biological evolution produces and modifies species.
When will SSI happen?
Since experts differ greatly on the question of when AGI will happen, it’s more difficult to predict when SSI will happen. However, Bostrom in his book provides some survey results, and the question of timeframe of AGI and SSI will be taken up in a future chapter. It obviously is dependent on a lot of advances in hardware and software.
Consciousness is a subject which will be touched on several times in this website. Suffice it to say here that if an AGI system can possess “artificial consciousness”, then an SSI system may be capable of possessing super-consciousness, a consciousness that is greater than a human’s, as much as human-consciousness is greater than a mouse’s consciousness. A subsequent chapter will take up this issue.
Need for Dharma
The need for dharma has already been touched on regarding AGI. Needless to say, the issue of existential threat is vastly magnified by SSI, and the real threat will probably not emerge until computers become much smarter than humans. The approach of the Virtual Veda Vyasa (V3) AGI system and follow-on SSI system involves the Vedic principle of “dharma”, and this will be explained in subsequent chapters.
In the next chapter, we briefly describe how the SSI system of the Susiddha AI project will be implemented. Just as AGI can be used to create a Rishi, SSI can be used to create an Avatar.
Notes and References
- How IBM Got Brainlike Efficiency From the TrueNorth Chip, Jeremy Hsu, IEEE Spectrum, September 29, 2014, http://spectrum.ieee.org/computing/hardware/how-ibm-got-brainlike-efficiency-from-the-truenorth-chip
- SpiNNaker (Spiking Neural Network Architecture), http://apt.cs.manchester.ac.uk/projects/SpiNNaker/
- Quantum computers ready to leap out of the lab in 2017, Davide Castelvecchi, Nature, January 3, 2017, http://www.nature.com/news/quantum-computers-ready-to-leap-out-of-the-lab-in-2017-1.21239
- NuPIC (Numenta Platform for Intelligent Computing), Jeff Hawkins and team, http://numenta.org
- Superintelligence: Paths, Dangers, Strategies, Nick Bostrom, Oxford University Press, 2014