Standard Caveat - Even though this essay appears on the SAG website it is represented as solely the opinion of Mike Deering, and is not endorsed by the SAG organization or the individual members of the SAG Board of Directors. It is the policy of the SAG to present a diversity of opinion on the Singularity.
AI - Artificial Intelligence; human level knowledge, reasoning, and cognition in a computer.
SAI - super human level intelligence in a computer.
Singularity - the explosion of nanotechnology, biotechnology, and computer technology.
Can we build a mind in a computer? Can we write a computer program that thinks the way we do, or better? Mathematicians such as Alan Turing, and Alonzo Church; and cognitive scientists Paul Churchland, Dan Dennett, Douglas Hofstadter, and Marvin Minsky have addressed these questions. The question revolves around the computational theory of mind; the idea that everything that goes on in your mind is based on computations carried out by the physical structures of your brain, including chemical processes, electrical processes, and logical processes, which could all be carried out equally well on a digital computer.
Roboticists like Rodney Brooks believe that all higher level thought processes are built up from basic cognitive concepts of sensory perception and motion control, particularly vision and locomotion. Mental capabilities involving intelligence and reasoning, they postulate, will be induced to emerge from the interactions of these simple functions, highly mediated by the physical systems that are related to them, by a process of evolutionary machine learning or human engineered trial and error. Robot evolution is projected to reach human parity somewhere in the neighborhood of 2040.
Whereas, the AI theorists believe that by combining multiple high level reasoning algorithms and semantics theory they can engineer a mind capable of deliberative general intelligence. Their designs are quite disembodied from the perception and motion functions, which they consider to be the input and output portions of their input-process-output model of mental functioning.
And finally there are the brain reverse engineers who believe that they can skip the decades of evolution of the roboticists and the theoretical understanding of the AI theorists, and through a process of scanning and simulating the component parts of the human brain produce a fully functioning mind in a computer.
Each approach can show some modest successes. There are robot bugs that can navigate novel terrains. And there are reasoning algorithms that can develop heuristics to solve math problems, prove theorems, play chess, and hold simple conversations. Also there are the neuro simulators who have developed functional models of some peripheral brain neural networks.
The main criticism that they all have for each other is that despite their early successes the method will prove inadequate before reaching higher-level mental processes. Such criticism is a big unknown. We won't know what works until we try. It is conceivable that all three approaches would succeed or that all three will fail. When you examine each case from its own point of view they are all very convincing, and my intuition is that all three are viable solutions.
If this is true then it is a matter of who will get there first. Rodney Brooks' 2040 sure seems like the long shot. The brain builders have a very predictable path, which is also aided by exponentially accelerating technological capabilities. And while the AI theorists will also benefit from faster more powerful computers and programming tools their primary limiting factor is theoretical, the functionality of their models. Several AI theorists believe that they already have the correct structural outline of a mind and just need the hardware and software to implement.
This is the wildcard in the calculation. The robot builders expect them to evolve to human equivalence by 2040. The brain scanner/simulators can show good evidence of completion by 2020. The AI theorists say they can beat the 2020 deadline by some arbitrary number of years giving dates ranging from 2006 to 2016 depending on who you ask and what they had for breakfast.
And to further complicate matters there may be synergy between the three approaches, which would tend to make their dates converge somewhat. Although AI theorists claim to have the inside track, the brain builders appear to have the smallest theoretical hurdles to jump. And that the work of many independent brain researchers can be more easily integrated than that of AI theorists means that more effective effort will be applied to brain simulation. Many independent brain researchers are working on the problem piece meal, but each AI theorists is tackling the whole problem. Even if the AI theorists have a workable model they may not beat the brain builder's time line.
Seed AI is the concept of a computer program with artificial general intelligence examining its own design and making improvements to it, thereby making itself more intelligent and able to make further design upgrades in a recursively self improving process. Some experts on seed AI such as Verner Vinge and Eliezer Yudkowsky believe that the recursive self-improvement process is powerful enough to produce what is known as "a hard take-off" or very rapid acceleration from human level intelligence to super human level intelligence. Computers already have some significant advantages over biological brains in performing computations relating to speed, accuracy, expandability, and modifiability.
Several projects are currently under way with the explicit goal of producing human level or super human level artificial intelligence, such as Novamente/AGIRI, The Singularity Institute for Artificial Intelligence (SIAI), and Cycorp.
Singularity Action Group website frames version.