My View on A.I.

We cannot perfect Artificial Intelligence at present because we still do not fully understand Human Intelligence.

Characteristics of Human intelligence including epiphany and eureka moments are still a mystery. The main barrier though in developing AI is that we cannot teach programs to believe something that is false. This is an integral part of Intelligence, Humanity has been able to evolve because it can pursue beliefs that have no foundations whatsoever. In other words Humans can have faith. Machines can't they are restricted to logic and can only learn what that logic can process.

The problem here isn't really the issue of programming computers but ironically that Science leads the way in AI and holds no place for Religion. Human Intelligence is defined by the ability to believe something even when all evidence points against it. This is integral in our ability to form a hypothesis and experiment. It brings forth new knowledge that is later connected to what we already know. This, in philosophy and in psychology is referred to as a priori knowledge  i.e. knowledge independent of experience. Machines are limited to using posteriori knowledge - knowledge dependent on experience and empirical knowledge - in other words Machines can only learn through experience.

The other main barrier is that computers and machines are incapable of true random actions. A computer cannot act in true random fashion, their actions are always predeterminable. Computers cannot generate true random numbers - this can only be achieved through incredibly complex processes. Processes which for which it is not feasible to include within any AI setting. As a result AI programs cannot "evolve" instead they are "trained" in which an initial set-up is defined and they progress through millions of variations until they reach a variation that solves the task they were given. This still does not constitute machine evolution as it cannot encompass randomness in the way real evolution does.

This video below depicts Machine Evolution.  This approach is a novel approach to Artificial Intelligence design that basically accepts that we do not understand Intelligence so rather than try to mimic it we should create very basic structures and give the A.I. the ability to evolve.  Evolution is by far a simpler concept to model than Intelligence in itself.  The difference here as opposed to Machine Learning is that the way in which the A.I. operates changes rather than staying static and building experience.  Watch the video to understand more:

Edit: Unfortunately the original video has been removed and those that have replaced it show algorithms beyond the scope of the time when this post was first written.

Now before you start worrying this still does not fully constitute Machine Evolution on a level that mimics Human evolution.  Likewise these structures cannot evolve sentience.  These evolutionary programs only work when the goal-set is tightly controlled.  It is however interesting as a friend pointed out that these programs can develop 'brains' that work but we have no idea why they work.  Being able to produce complex structures that mirror natural selection which results in the same complexity as natural evolution is potentially worrying as the potential for future generations to have extended fields of study such as 'Machine Psychology' arises - i.e. the rationalisation and study of Machines in an attempt to understand how they work, in the same way we study Humans; ultimately this is caused by the reality that future AI will reach a level at some point where we will have no understanding of its structure and coding at all.

In brevity Machine Intelligence needs a purpose, Human Intelligence has no purpose, or purpose unknown, which we have sought to define for thousands of years.  We seek to define it because we believe there must be a purpose even though evidence for all intents and purposes would conclude that the possibility that we have no purpose is plausible.  In other words, our continued existence is one pursued out of faith.