AI - From Insight To Action

When we look at the field of Artificial Intelligence (AI), oftentimes the focus has been on different algorithms and techniques for AI, which in turned opened up new capabilities to provide insights in different fields. With vastly improved automation and telecommunication infrastructure, today the focus is shifting from providing insights to taking actions.

The core elements of AI including search, machine learning and natural language processing achieved great progress in the last two decades in terms of new algorithms and capabilities. Many of the techniques, which were applicable in academic environment only, have been rolled out to different use cases in a number of industries. In recent years, the introduction of deep learning techniques combined with great improvements in data processing capabilities opened up the way of new approaches previously not possible. Today, we are able to talk about using AI in areas including human resources, healthcare and customer service to name a few.

In what we observe today, there is one important trend. The application of AI in today's solutions is orienting more and more towards action in addition to providing insight. Compared to techniques and applications in the past, the ones used today are not just providing recommendations, but they start to have capability to take action at different levels. AI applications based on reinforcement learning algorithms are able to play games, other AI applications are able to drive cars. Based on the application and the degree of freedom provided, they are able to converse with customers to solve customer service related issues.

What enabled these developments? Some important elements are listed below:

• Recent developments in AI techniques is one driver. New techniques (and in some cases new applications of old techniques) including deep learning and reinforcement learning enabled AI applications to learn from vast amounts of structured and unstructured data. Developments in natural language understanding techniques also helped applications to work on unstructured data to derive insights previously out of scope.
• Developments in data processing and storage capabilities changed the game. For example, using GPUs was not the norm a decade ago, but it is among the table stakes in executing many machine learning algorithms. Even more, new structures like Tensor Processing Units (TPU) are pushing the envelop even further.
• Natural Language Generation (NLG) capabilities are showing rapid advances. These capabilities enable AI applications to converse with the user in a more realistic way. This action capability opens up the way for AI applications to be used in many "last-mile" areas including customer service applications.
• The ascent of Robotic Process Automation (RPA) concept and its applications provided many AI techniques with improved action capabilities. With RPA capabilities, the AI applications are able to turn their insights into actions by accessing and using different systems. As algorithms in these two areas develop in parallel, it is possible to expect more novel combinations of applications integrating AI and RPA capabilities.
• Increase in data volume, variety and velocity is another important component. Huge amounts of structured and unstructured data is enabling AI applications to elicit and learn many different scenarios and preparing them to handle a diverse set of novel scenarios not explicitly present in data. Having a wider set of knowledge about the subject matter and the context, AI applications are getting better and better in deciding and executing actions.
• When you take an action, you should be able to read the impact from the environment, so you can decide on the next set of actions. Improvements in sensor capabilities and communication infrastructure are increasing the capabilities of AI applications as they are better able to check the changes on the environment, in which they operate, based on their actions. Coupled with the developments above, AI applications are gaining a better sense of their environments compared to the static information sets in the past.

What are the potential developments?

• Increase in the number and type of use-cases. As AI applications are covering more ground with action capabilities, it is possible to consider many different use cases in a number of industries.
• Different versions of human-machine collaboration. Integrating insight and action can lead to more productive ways of human and machine collaboration as the level of interaction increases in doing different tasks.
• Development of new AI techniques. Ability to act and to improve insights based on the results of those actions can lead to new and different AI techniques helping to achieve more use-cases in different industries.

In terms of use in business, AI's development is still ongoing. Improving action capabilities together with insight generation can lead to novel business applications in the future.

The views expressed in this article are the views of the author only. This article provides general information and point of view, and should not be considered as professional advice.