According to Wikipedia, “in mathematics, a singularity is in general a point at which a given mathematical object is not defined, or a point of an exceptional set where it fails to be well-behaved in some particular way, such as differentiability”.

If we take that definition and apply it to robotics and more specifically to Artificial Intelligence, we could say that singularity will arrive when technology is able to create machines which are much more “intelligent” than a human being and able to find solutions for problems that are still unsolved, both at a mathematical (algorithms and theorems) and medical level. As that artificial intelligence is able to learn and grow exponentially so will society, getting to a point in which human beings will not be the only major figures writing the future of humanity.

An easy way to understand singularity would be to compare the evolution of the different civilizations that have inhabited the Earth. An inhabitant from Ancient Greece (3000 years b.C) would be surprised at the changes in the Roman Empire society (from 30 years b.C until 476 a.C) the same way one of our great-grandfathers would be surprised at the changes in our present society. In the first example there is a time gap of hundreds of years while in the second one the gap is only of 100 years or less, to a large extent thanks to technology’s progress. The same will happen with us, technology will shortly offer such a great change that at this moment we cannot imagine the society in 50 years time.

Therefore, we could say there is singularity when a society changes so much that previous generations would fail to understand the changes in that new society.

In March 1993, Vernor Vinge presented at a NASA’s Symposium a thesis called “The Coming Technological Singularity” where he predicted that within thirty years, we would have the technological means to create superhuman intelligence. Raymond Kurzweil also believes that in a short time we will have the medical advances needed to reverse the aging process which would mean, in my humble opinion, a radical change in the society we know nowadays.

But to have machines like the ones predicted by Vinge and Kurzweil some important changes must be done at a computing level. Regarding Artificial Intelligence a lot of progress has been made in neural network systems but we are still using computers based in systems from the 50s. No matter how fast our current computers are they still carry out “serial” operations instead of “parallel” ones like human beings do. To get an idea of what we are talking about a human brain has approximately 10 billion neurons and 60 trillion connections while a “typical” artificial neural network, operating in a CPU, does not have more than 1000 neurons.

Following these outlines, we are trying to develop a neural network based on GPU (Graphics Processing Unit) chips which have a high degree of parallelism since their calculation units are vertex. In some cases the computing power of GPUs is over one hundred times a CPU’s.

We are carrying out tests in a Multi-GPU machine formed by 4 Nvidia GTX295 graphics cards (8 GPUs) with a mathematical processing power of 7153,92 Gflops.

Some of Qbo prototypes have a Mini-ITX ASUS board, model AT3IONT-I, installed which graphic chip integrates a GPU. So, if our attempts are successful the neural network will be transferred to each one of these prototypes to be used with some API modules leaving the CPU free for other tasks.

Photo – Ray Kurzweil. – Dominik Gubi / Creative Commons

Comments are closed.