Machine Vision

One approach to AI may be of intelligent design, whereby the AI "final product" gets crafted by our ability of high level abstraction.

An alternative would be survival of the fittest, whereby simple and basic AI objects are experimented with and "genetically" propagated to future generations based on their survival rate. A continuous mutation would ensure the process goes on to form evolution.

A framework may be proposed, for the integration of many such modules, and higher level association and lingual abstraction of their respective functionality.

To create evolution, we need movement and interaction. As a starter, the objects are provided with a visual view field and basic rules for obstacle recognition and avoidance.

The objects are represented by dots, view field is highlighted in yellow, and obstacles are randomly generated on each run.

Download the application here.

Artificial Empathy

In a quick experiment, a system was trained to associate language objects with empathic emotions; the emotions were then associated with colors. As a result, words composing more than 65% of the written English literature were associated with color representation of emotions.

Point made is that it does not require much to artificially embed simple representation of human empathy in machines.

A derivative of this experiment can include probabilistic empathy recognition in complete sentences, poems, or stories.

The application, named INSPIRO, was adopted for inclusion in an ambient light system peripheral, and is available for free download at http://www.arfx.nl

Reality is merely an illusion

Albeit a very persistent one - Albert Einstein.

Consider this - from months before we are even born, our senses feed us with input of what's around. Then as reality stays persistent and repetitive, we get hard wired recognition of the environment, even without yet having names or abstract association of it.

I have therefore set to attempt the same in a case study of heat, fire, and smoke recognition. A visual input of a dynamic scene from Quake 4 game was fed through an artificial neural network and translated into probability of association with heat, fire and smoke.

Abstract association of two of the objects was also put in place - "there is no smoke without fire".

Here is the result

Supercomputing

In 2000 the first supercomputer passed 4 teraflops, or 4 trillion calculations per second. By June 2007, the fastest supercomputer - IBM's BlueGene, with 131,072 processors - achieved speeds of 280.6 teraflops.

Growth in total performance of all top 500 supercomputer systems in past few years is as follows:

06-2007 4.92 petaflops (or 1,000 teraflops, or 1,000 trillion calculations per second)
11-2006 3.54 petaflops
06-2006 2.79 petaflops

It is apparent that computing power is not what we currently miss to create intelligence...