The field of artificial intelligence (AI) is a vast one.
AI has been making enormous strides in a myriad of different technological industries such as self-driving cars, automatic translation systems, speech and textual analysis programs, image processing software, and not to mention, all kinds of diagnosis and pattern recognition systems in a varied number of professional fields.
In many cases, AI may well surpass even the best human performance levels at executing specific tasks.
Indeed, AI is the heart of a new lucrative commercial industry with massive activity, huge financial investment, and a tremendous potential to boot. It would certainly seem like there are really not too many areas that are beyond improvement by AI — no tasks that can’t be automated, no problems that can’t at least be helped or improved upon by an AI application.
But is this the full extent of AI?
The Potential of AI is Nearly Limitless, Nearly
One such problem that any AI system may face is that not everything in real-life is computable.
Alan Turing — father of theoretical computer science and artificial intelligence, brilliant mathematician, and code-breaker extraordinaire — has proven that some computations may never be completed within our short, meagre lifespans. Whereas, others may take years, decades, or even centuries to complete.
For instance, using AI, you may easily compute a couple of moves ahead in a simple albeit straightforward game of chess. Though, if you were to examine all the moves to the end of a typical 80-move chess game, you’d find it highly impractical to compute. Even by using one of the world’s fastest supercomputers (running at over one hundred thousand trillion operations per second), it would take well over a year just to get the information processed on a tiny portion of the explored chess space.
This problem is known as scaling-up.
Early AI research shows a bright future for the tech mainly due to the fact that initial experimentations produced good results on a small number of combinations of problems such as noughts and crosses, or otherwise known as toy problems. This result would not necessarily scale up to larger computations like the aforementioned chess or other real-life problems.
Fortunately, modern AI has developed a workaround when dealing with such scaling-up problems. The best AI systems do not look at all possible scenarios (or moves ahead), but instead, it looks a lot further than what the human mind can manage or perceive.
It does so by using methods involving approximations, probability estimates, larger neural networks, as well as a variety of machine-learning techniques.