Artificial intelligence: predictability, the fifth dimension

An area that has attracted a lot of interest and investments has recently been Artificial Intelligence, often shortened to AI. The goals of AI are to imitate human behaviours.

This definition has caused in the past a lot of confusion as it was interpreted as if AI aimed at imitating by means of electronic machines as much of human mental activity as possible and, eventually, at improving upon human abilities.

Today, AI is correctly interpreted as machines that imitate human behaviours (and not human mental activity!) for instance in driving cars, in answering telephone calls or like in our case at Cogenpower in running profitably and efficiently small scale Combined Heat and Power plant with annexed District Heating (in short CHPDH systems).

There has been in the past interest in AI from at least 4 directions.

The first direction is definitely the study of robotics, which is concerned with the practical requirements of industry for mechanical devices which can perform intelligently several tasks (usually whose danger, versatility and complication are better handled by machines than by humans). Often in this field we can add speed and reliability beyond any human capabilities.

The second direction is related to psychology; in this case AI is used to imitate the behaviour of a human brain (or of any other animal for what matters) by means of an electronic device and by observing it to deduct theories and learn how the brain may work.

The third direction that was relevant in the past but that today has seen a reduction in interest is involved with philosophy, assuming that AI might help finding deep insights on how our mind works (note the clear difference we are making between the brain and the mind).

Finally, the fourth direction is the expert systems, according to which the essential knowledge of an entire profession (i.e. legal profession) can be coded into a computer or more sophisticated architecture, until a point when you can even think about substituting it.

We believe that this traditional separation (see for example Roger Penrose in The Emperor's New Mind) is missing a fifth dimension that has to do with the ability of the coded electronic device to set up a problem and find the proper solution before the environment in which it may (as in the case of robotics) or may not act (as in the case of a laboratory simulation of an experiment) changes. We define this direction as predictability.

The challenge of predictability is extremely interesting and difficult as it involves the ability to model the environment as a stochastic system and manipulate it as if our system were simultaneously embedded in it and perturbing it.

What we have discovered is that these different dimensions are not clear cut but they often overlaps creating a more realistic (at least it is perceived like this) and a more complex definition of what is AI today.

In our CHPDH systems for example our technology, named Anaconda, lives in a new dimension defined as the overlap of robotics and predictability that we believe will be where more and more everyday applications will show up as in the self-driving vehicles or office assistants.