26. January 2021 By Jan Heuker
Which practical AI application fits your organisation?
Artificial Intelligence (AI) is applied in more and more areas, from virtual assistants to self-driving cars. At the same time, there is much discussion about its scientific, economic, social and political consequences. One thing is clear: AI is a many-headed creature with countless application possibilities. But which ones are best suited for your organisation?
It took a long time for artificial intelligence to make the leap from universities to business. In the meantime, however, the tide has turned, partly due to the drop in the price of data storage and the increase in computing power. We are now constantly reading stories about new breakthroughs and business applications, but also about the potential dangers.
The form of AI that is most used falls under the heading of 'weak AI'. This includes a comprehensive set of methods, procedures and technologies to develop or train accurate models based on data and apply them. This enables organisations to make better decisions and predictions. Exactly which methods are used can be most easily explained through concrete use cases, which I will cover later in this paper.
Symbolic and subsymbolic
AI can be divided into so-called symbolic and subsymbolic systems. In a symbolic system, rules and relationships are applied to concepts that can be understood by humans. For example, the properties of the term 'father' can be used to separate these individuals from a diverse group of people.
Subsymbolic systems, on the other hand, are largely black box systems for humans, both the method and content of which are virtually impossible to grasp. The fathers from the earlier example are broken down here into a multi-dimensional data space based on a plane division, which in itself is almost impossible to follow.
Which AI system is most suitable depends on the context of the application and the type of organisation. For example, if regulations require that decisions on loan commitments or approval of construction procedures must be understandable and transparent, then sub-symbolic procedures are out of the question.
Applications of AI systems
Artificial intelligence derives its intelligence from learning and modelling. Machine Learning (ML) offers the possibility to train a model automatically based on data. There are different forms, namely supervised, unsupervised and reinforcement learning:
- Supervised learning means that experts need to define a procedure in advance for processing a set of training data to arrive at the correct decision. Since a lot of training is often needed to achieve good results, the effort required is often high.
- In unsupervised learning, the system is left freer to find connections between data based on their similarity or distance, without the need for experts to introduce rules or procedures. In this form, the only input usually required is the number of insights to be found.
- The term 'reinforcement learning' covers procedures that learn on the basis of direct feedback and not by giving training examples. Here, the desired results are fixed, and the system looks for the most optimal route to the desired result. Well-known applications are learning to play strategic games such as chess and Go, or video games.
Regardless of the chosen variant, thanks to ML a model can recognise all kinds of connections and provide new insights. This results in a wide range of possible applications, like for example
- Alarm systems in mechanical engineering, where the system learns to interpret the mechanisms that lead to failure at an early stage.
- Integration and linking of data from different data sources in order to find connections. Think for example of customer interactions via social media and a webshop or helpdesk.
- Partly or fully automatic recognition of relevant text passages in unstructured documents, such as the process of reporting damage to insurance companies.
- Automatic procedures for detecting and preventing fraud in the financial or insurance sector.
- Camera registration and monitoring, where ML models act as autonomous cameramen, for example to detect burglars and thieves or to record football matches with the most relevant and exciting camera positions.
Organisations with complex systems often apply combinations of different AI methods. Take this real-world example: a bank where fund managers rely on internal bank advisors for their investment decisions. These experts read all kinds of analyses, for example on the development of industrial sectors in different regions of the world. It is almost impossible for individuals or teams to keep track of all relevant information sources and then extract the right knowledge and interrelationships from them. With AI procedures, you can automatically analyse all available reports and then make the insights available in natural language. Natural Language Processing analyses the texts to extract the information so that it can then be processed by machine. The information obtained is then stored in a knowledge system, which can suggest investment decisions based on a complex model. Financial experts can even ask their investment advice questions in the form of natural language, similar to the way we do with virtual assistants such as Alexa, Google Assistant, Siri and Cortana.
There is a growing demand for complex AI applications of this kind, where various techniques are used to provide an organisation with the right insights in an efficient manner.
It is clear that AI has moved beyond the conceptual phase and is now adding more and more value to organisations. However, concrete applications are not always obvious, especially when an organisation has to deal with complex systems, processes and regulations. It is therefore important to seek advice from an experienced consultant with experience in both the integration of AI applications into existing IT systems, and the complex algorithms and procedures required for the effective application of artificial intelligence. Only then will investments in AI actually lead to useful insights and results.