Algoritmes in AI
The role of algorithms in AI, or artificial intelligence
Algorithms and AI or KI, artificial intelligence, are inseparable. In fact, algorithms form the basis of artificial intelligence. But what exactly is artificial intelligence and what role do algorithms play in this? As the name suggests, artificial intelligence creates an artifact that shows some form of intelligence. This brings great benefits for humanity but also has a downside. Because how much thinking do we want to give machines?
Algorithms and AI are inseparable
Before we can clarify the role of algorithms in artificial intelligence, the definition of an algorithm is needed first. An algorithm is a set of instructions that leads to an intended end goal from an established initial situation. In principle, an algorithm is therefore separate from a computer program, although computers are generally used for the execution of an algorithm. The intended end goal of an algorithm can be anything. The finite series of instructions are prepared in such a way that they can generally deal with eventualities that may occur during implementation. Often algorithms have repetitive steps, which is called iteration. They also generally require decisions, comparisons or logic to complete the intended task.
Comparison with cooking recipes
One and the same task can usually be solved with various instruction series. The difference lies in the amount of effort, space or time that the algorithm needs. This is called complexity. For example, you can compare an algorithm with a cooking recipe. Suppose you want to make a potato salad. With one recipe you must first cook the potato according to the instructions and according to the other recipe you must first peel the potato. In both recipes, however, these two steps are required for the correct execution of the intended end result, the intended potato salad.
Formal systems and the 'internal state'
This works in a similar way with an algorithm. It must correctly implement the intended end result and the algorithm itself must therefore be properly executed by the computer program. In formal systems, algorithms are essential for, among other things, the way in which a computer processes the information. A computer program is therefore a formal algorithm that gives the computer instructions which steps must be carried out in which order to reach the intended end goal. Information is therefore processed with algorithms. The data is read from the input device and then written to an output device. It can be saved here. This is also called the 'internal state'.
Machine Learning (ML)
Back to algorithms and AI. With artificial intelligence, algorithms can be developed for computers, among other things, to enable them to learn. Machine or automatic learning is a comprehensive field of research within AI and is referred to as Machine Learning (ML). Patterns are extracted from data. Today, these developments are gaining momentum because the cloud brings computing power in abundance. For example, a computer can translate texts, understand spoken word and also understand images. You will find Machine Learning in the self-driving car, among other things.
Analyzing data and data mining
Machine Learning is focused on analyzing data or statistical analysis. It is focused on implementation in programs or algorithmic complexity. In addition, it is related to so-called 'data mining'. Hereby relationships and patterns in large amounts of data are searched for in an automated way. In both research areas, the use of algorithms in AI is the fundamental basis. They are used to analyze data, to gain insight and to subsequently make a prediction or create a determination with it. Instead of manually coding software with a specific instruction set, the machine is trained at Machine Learning. This by using algorithms and large amounts of data that should offer the possibility to learn how to complete a task.
Neural networks and Deep Learning (DL)
A Neural network is used to discover patterns in data. In addition, the software simulates a structure that is similar to the neurons of our human brain. This structure can be made up of dozens or even hundreds of different layers. This is therefore called Deep Learning (DL). This neural network can be trained with examples that are classified by humans, but the network can also find patterns in data that are not controlled by humans. In addition, DL can be semi-controlled. DL is part of a wider family of methods for Machine Learning.
Deep Learning: controlled, uncontrolled and semi-controlled
In the case of controlled Deep Learning, AI algorithms provide examples of the input and related output. The learning aspect here is that based on these examples, the input properties determine the final output. Once the learning phase has been completed, algorithms can independently realize the correct output for new input. With uncontrolled Deep Learning, no examples are given of the desired output. The algorithms themselves then discover a structure in the input that is given. With semi-controlled Deep Learning this is in between both learning methods.
Correlations and no necessary causal relationship
With algorithms in AI, the relationship between different variables is referred to as correlation. These correlations need not have a causal relationship. This is the reason why they often provide a simplistic picture of reality. For example, there is a correlation between the number of winners of the Nobel Prize and the frequent consumption of chocolate, but this has no causal link. The reason for this is that Switzerland has relatively many universities.
The chance of stereotypes
Another phenomenon with algorithms in AI are stereotypes. These can easily sneak in there. Suppose, for example, that the computer is fed with incorrect data. An example of this is that speech recognition did not function well with women's voices for a long time. Simply because the software in question was only trained with men's voices. Another example is software that can recognize emotions. This software was wrong in recognizing emotions in children. The reason was that their photos were not processed in the dataset.
Algorithms and AI can also cause unwanted results. For example, when algorithms pay attention to incorrect factors based on data from the past. The risk of this is indicated by the term 'predictive policing'. Suppose extra checks are carried out by agents in districts with a lot of crime. Then the chance is likely that more arrests will take place there, as a result of which the crime rates will irrevocably rise even further.
The artificial neural networks, the architectures of Deep Learning, have of course also been successfully applied in practice. For example, think of areas such as speech recognition, image recognition and language processing that occur naturally. An example of this is customer service in the form of an online assistant that is automated. In addition, machine translation, bioinformatics, audio recognition and social network filtering are excellent examples of artificial neural networks. Sometimes the results of algorithms at AI are so good that they are not only comparable to those of human experts but are even superior to this.
Benefits for humanity
That algorithms and AI offer great benefits to humanity is something that we can experience every day. Self-steering systems, recognizing patterns, speech, image and sound, running robots and so on. Question-answer systems and translation machines also help us in areas such as efficiency, convenience and optimum user experience. Healthcare is an industry in which artificial intelligence plays an important role. Great things are also expected of artificial intelligence here. For example, an intelligent machine could detect and recognize small deviations and changes much more accurately than a physician of meat and blood. In addition, AI can also perform complex and intensive operations tirelessly.
The down side ...
It is important that we, as humanity, remain critical about the thinking capacity that we give to machines. Hundreds of scientists worldwide, for example, are concerned about weapons that make autonomous decisions. That could potentially make them even more dangerous than nuclear weapons. For this reason, they argue for more research to be done on the 'self-learning aspects' of algorithms at AI. Only by staying alert to this can developed systems continue to do what humans want them to do… Another example of what humanity did not envisage with artificial intelligence is the phenomenon of 'deep fake news'. This is very topical today and is better known by the term fake news.
That some people fear this downside of artificial intelligence is expressed among other things in the amount of films that have been made about this. For example, films such as The Matrix, Autómata and I Robot are based on the idea that AI will be responsible for the demise of humanity. In these films, machines become autonomous and can fully develop themselves. This takes place without human intervention. In such cases, super intelligence is spoken: the moment when AI is smarter than humanity.
For a personal conversation and a demo of our platform!