Discriminator

Luís Ángel Fernández Hermana - @luisangelfh
5 December, 2017
Editorial: 156
Fecha de publicación original: 23 febrero, 1999

To free oneself of bonds, a good head is better than good arms

Forget artificial intelligence. This season something much more sophisticated and suited to the age is going to be in fashion: “artilects” or artificial intellects which, as usual when describing these devices, will apparently be much more intelligent than us. To start off with, things do not look very promising: they will inhabit the body of a cat, not just any cat, but a robo-cat or Robokoneko. If they show that they are able to transcend the “oil age” and get an electronic miaow out of this heap of hardware, “artilects” will have a brilliant future ahead of them, especially in computer networks. Over the years to come, the demand for encouraging cooperative work, enhancing collective intelligence and increasing the capacity of individuals and organisations to discriminate when it comes to looking for and selecting information, is going to grow in the Net. And, this is a job that will suit “artilects” to a T. If we add to this automatic learning and evolutionary reproductive functions, the stage is set for great things. As well as human beings, a substantial part of the next generation in the Internet will come from robotics. And, more specifically from artificial intellects.

“Artilects” are the latest product of a cross between various disciplines which, while they have spent quite a while frolicking in laboratories, in the last two years have advanced by leaps and bounds. The combination of interconnected technology and communication between electronic processors, the neurosciences, molecular biology, artificial devices based on brain functions (vision, speech and sensorial and spatial recognition sensors) and the specific demands of the Internet are the basic driving forces behind its development.

At the centre of these new “elements” is Hugo de Garis, a Briton living in beautiful Kyoto and coordinating the work of a select group of Japanese, British and American laboratories. And the mother of all the brains behind it all is CAM (Cellular Automata Machine), developed by Genobyte of Boulder, Colorado, US, which will shortly start to function with 37,7 million artificial neurons. In short, a helluva lot of neurons, above all when one bears in the mind that until now only a few experiments have been done with a hundred of these types of neurons. Although we are admittedly far from the 100,000 million neurons that our brains contain, we have left the 10 million that insects have way behind. Researchers working with “artilects” are only asking for a little patience. If we give them a little time they might get closer to the nervous systems of higher animals by the end of the century (20 months away). Then it would be just a step from them to us.

If the fact that an artificial brain with almost 40 million neurons is able to function is interesting enough, this is not, perhaps, the most remarkable thing about the whole venture. The spectacular thing is the possibility of applying evolutionary theories to it in order to adjust and diversify the tasks it is able to do. For this to occur, artificial neurons are not simulations obtained through software as has been the case in most of neuronal networks constructed up until now. This is the reason why up to the present we have only managed to interconnect a few hundred artificial neurons. The key lies in using real electronic connections. CAM works with special circuits that are capable of altering their own connections. In this way, the best ones for each function are sorted out and, at the same time, new functions based on acquired capacities which were not in the original design are discovered.

In order to progress from the most simple connected states to complex architectural structures with “experiences” accumulated during the growth process, CAM will be divided into modules, made up of a thousand neurons each. These bundles will be randomly interconnected and will give rise to new modules called “chromosomes” (interesting this tendency engineers have for biological language). The modules which are most efficient at processing information will transmit their chromosomes to the next generation of connections. During this process, it is anticipated — and this is in fact what has occurred in the laboratory — that mutations will take place when they exchange parts with other surviving chromosomes. The other curious thing is that CAM will regulate the rate of this evolutionary process and will speed up as CAM acquires greater efficiency with each new generation. There will be more and more information in the system but, simultaneously, there will also be more modules with a notably improved capacity for discriminating and using that information.

Translation: Bridget King.

print