Data Driven Foresight

Being able to predict the future is one of mankind’s oldest dreams. There are systems, for which this can be done quite satisfactory. For example, astrophysical simulations suggest that our sun will turn into a red giant in about five billion years. However, it is impossible to predict with certainty what the weather will be like over the next two weeks.

And what about “predicting” future developments in technology, society, politics or business? Future studies have developed an extensive portfolio of different methods for this purpose. Fraunhofer INT has been working intensively on these for many years. Broadly speaking, the methods can be divided into qualitative and quantitative approaches. While the former includes traditional approaches such as the Delphi method, various workshop formats and the scenario technique, quantitative approaches involve methods such as bibliometrics, forecasting and patent analysis. In recent years, the scientific discourse has increasingly centered on whether methods from the field of artificial intelligence (AI) would be suitable for expanding the future studies methodological portfolio and, if so, whether an AI could be capable of providing insights into the future on its own. 

In this context, the concept of data-driven foresight was developed. The central question of this research field is how data-based methods can be utilized to gain insights into (technological) futures and support foresight processes. The aim of this approach is to support and improve future-related decisions.

The concept is based on three pillars. In practice, the focus is usually on specific questions and problems that arise throughout a typical foresight process. In the case of a technology analysis, for example, this may involve structuring a topic and identifying new fields of application. Or it might be important to know which research priorities are set by different countries in order to use this information to assess a topic. Furthermore, future scenarios for a specific topic might be needed as part of a scenario process. In these and many other possible use cases, the following questions can be asked:

1.     Can I answer this question using data?

2.     Which data sources do I need to do so?

3.     Which methods do I need to use?

This is at the core of data-driven foresight.

On the one hand, this core is formed by the necessary data. A whole range of — in some cases, very different — data sources are available. These include publication data, patent data, economic data or even social media data. From a technical point of view, the difference between structured and unstructured data plays a key role, as this determines the level of effort required to unlock the potential of the data.

Suitable analytical methods are needed to process the raw data and extract valuable insights. To this end different approaches exist, that can be broadly divided into three categories: traditional data mining, text mining and methods from the field of machine learning.

However, it all comes down to the specific issue, the so-called “use case”, that is being addressed. This essentially determines which data sources can be considered and which methods must be used. You can find a short overview of some of the use cases that were implemented in KATI here.

Over the past few years, Fraunhofer INT has developed KATI (Knowledge Analytics for Technology & Innovation), a system that realizes the ideas/concept behind data-driven foresight, harnessing its full potential.