We live in interesting times. For some, this is a curse (as in Chinese:”may you live in interesting times”), for others a source of emotions and creative excitement. Before our very eyes ideas known from science fiction literature are becoming reality, and solutions we have known from everyday life – obsolete.
Speed, amount, accessibility – these feature of information are not evenly distributed, and so our knowledge of the technical possibilities of today’s world is never complete. In addition, we sometimes mix expectations with dreams and actual possibilities.
And so the question: „Does artificial intelligence(AI) exist?” is answered „yes of course” by some, and „no, and it won’t for a long time” by others.
But it is hard to deny the facts – machines can learn.
Of course, it is not their innate ability, they require not just a teacher but also a creator which designs its cognitive apparatus, sets goals for it and ways to achieve them in such a manner, that we will be able to call it a learning process.
Only why teach a machine, if the process it performs is assumed to be optimal?
This assumption, unfortunately, does not hold, if we take into account a number of processes that affect the realisation of the process – that is when only ML allows the machine to reach optimum.
There are different ways a machine can learn – from learning rules, through recognising and learning from examples through imprinting a decision tree.
All methods lead to optimisation, although depending on the functions performed by the machine and the environment it operates in, teaching methods must vary.
In case of robots working in identified, uniform environment the learning method is relatively simple(haha :)) and is aimed to lead to the most economical accomplishment of a given task – parameters such as time, number of performed operations, cost of a single operation etc. are all taken into account. Naturally – instead of self-learning ML – the system can display the aforementioned data and let an operator set the machine up to manually achieve optimisation. The problem is these parameters can change – for example temporarily varying energy cost, the number of operations dependant on the delivered components etc. In this case, the trainer must stand next to the machine ever ready to push the lever or turn the potentiometer(in case of RPA he or she would naturally sit at a keyboard ).
If a machine can learn the relations between these setting itself and change them dynamically, not only steering becomes easier but only then it can be assumed to strive for optimum.
Other learning functions that find application in RPA are all kinds of cognitive techniques -speech recognition, handwriting reading, the ability to follow and recognise objects.
It is anticipated that Machine Learning will keep developing at a fast pace, reaching close to 40% progress in the course of the coming five years. That is hardly surprising, given the quantifiable benefits coming from the ability of machines to intelligently react to changes in their environment. A good example is an advanced logistics action, such as designing a delivery chain – it is anticipated that within the next five years more than half of the errors occurring at the moment.
Intelligent machines – form the simplest robots replicating the moves of the mouse cursor on the screen, through advanced solutions controlling the warehouse stock, logistics or even company’s assets are the next step in process automation.