Some people may still think of artificial intelligence AI and machine learning as future science fiction things. They may relate it to movies that show some robots developed to be independent and starting to deviate from their intended use. The reality is that artificial intelligence and machine learning are not science fiction anymore.
It is increasingly being used for various advanced applications. A well-known example is the virtual assistant available on mobile phones like Apple Siri and Google Assistant. Autonomous cars which are currently available for commercial use are also largely based on artificial intelligence AI.
Artificial intelligence is part of computer science that aims at developing computers or machines to be able to think in similar manner to human brains. It can be defined as the ability of the machine to learn from the external environment by acquiring data, analyzing data by reasoning to reach conclusions and then correct the machine behavior.
The aim of such field is to make the machines more independent in doing more complex tasks. Machines provided with efficient artificial intelligence AI will have the ability to autonomously develop through gaining more experience. This is very similar to a child learning from the surroundings while growing older or to an employee becoming more efficient by gaining more experience.
The definition is however not exact and it is more related to how the performed task is viewed as complex and magical to perform. This is called the “AI Effect” which claims that when things become more familiar and simple to perform by machines, they are considered as traditional computing and not artificial intelligence AI.
There is no definitive approach to make a machine intelligent. Some scientists may try to simulate the human brain logic and psychology to make the machine intelligent. Others say that this adds too much unneeded complication. Approaches used for AI include symbolic and sub-symbolic, cognitive based, logic based or knowledge based AI.
AI can be roughly simplified into few principles. The ultimate goal of these approaches is developing machines that are able to “decide” the best way to perform a task and to evaluate results instead of providing the exact way to perform a task. This is particularly important for performing complex tasks with large number of variables.
Machine learning can currently be considered as the most common application of artificial intelligence AI. Machines are allowed to use trial and error in simulation experiments to learn the best way of interaction.
For example, an autonomous car is allowed to move to a certain target scoring the path that the car followed based on certain criteria such as the distance, time and safe arrival. The simulation experiment is repeated many times till the score cannot be improved further indicating that an optimum path was reached.
Another method of machine learning is to provide machines with a large database of conditions and results that can be used to determine patterns and rules using statistical analytical methods. This is currently possible since most companies currently store data regarding their operations, consumers and any other details. These databases provide invaluable insights about their consumers and their trends.
Amazon was among the first to use this technology which helped the company to reach its world-leading position. The company collects all possible details about the users browsing including clicks, time for browsing each page, products viewed and purchased. The data is then used to offer suggestions to users with products most suitable to their needs and wishes which effectively increased the purchases on the website.
Amazon developers say that they still have much to achieve using this technology particularly as their database grows massively with millions of purchases completed yearly. Amazon claims that they may be able in the future to know, for example, that a woman is pregnant based on changes in her shopping trends. The company can then send special gifts to the user or offer suitable suggestions.
This approach is currently used for many applications including face, sound and handwriting recognition. This also explains how this technology advanced greatly with the huge increase of available data until it became commercially available on all new smartphones. There are also currently some open-source databases that can be used by developers and researchers as MLDB which has been acquired by ElementAI.
The most advanced and state-of-art technology used for artificial intelligence AI is the deep neural networks. This design of machine intelligence is based on using several layers of data analysis to reach patterns and conclusions.
For example, the machine identifies the features of the face on one layer. The second layer will focus on defining and recognizing the facial expressions while the third layer will relate these facial expressions to emotional patterns and then the program will relate these emotions to acceptance or rejection of certain actions.
The use of layers is the reason behind calling it deep learning. Artificial neural networks are similar to biological neural networks which are composed of layers of neurons connected through synapses.
Additionally, biological neural networks are characterized by their plasticity which refers to reinforcing of certain neural pathways that result in desired results. Artificial neural pathways are characterized by having weights which are expressed as numbers.
Activation of a pathway is triggered by adding up the weights of the receptors to a certain threshold. The weights of the neural pathways can be increased or decreased based on certain logic or conditions that represent favorable or unfavorable results. The terminal aim of deep learning technology is to allow machines to learn by analyzing examples without being programmed by rules for specific tasks.
Although artificial intelligence AI holds promise to unlimited potentials, the results are sometimes much less than expected due to several barriers.