By Jeon Seung-min WIRED Korea

It has been a risky business to make predictions. It is an expert opinion that a prediction made with the use of the conventional Delphi method has an accuracy rate of about 30 percent.

But the accuracy rate of a prediction about most promising technologies by the Korea Institute of Science and Technology Information (KISTI) is claimed to be raised to 86.7 percent with the use of artificial intelligence and big data.

At the top of the list of the selected technologies is the one for the storage of hydrogen and the conversion of hydrogen energy to electrical energy. The technology is key, among others, to the manufacture of hydrogen energy-powered vehicles.

It is followed by those for eco-friendly air conditioning; the use of carbon dioxide as a resource; vehicular control for self-driving; AI-based machine vision; the manufacture of ultra-high performance concrete; biodiversity research; high-voltage direct current(HVDC) electric power transmission; humanoid robots; and hyperspectral imaging.

Tech companies have good reason to pay keen attention to this and other similar predictions, as it is a matter of time that today’s dominant technologies are made obsolete. Newly emerging technologies are replacing them rapidly. Tech companies have to closely follow new trends in technology for this reason, and all the more so as technological change is exponential.

KISTI, which collaborated with the Data Science Lab of Myongji University in developing a prediction model, said on February 3 that great strides will be made in those technologies during the second half of the 2020s.
Even in the absence of KISTI’s prediction, few would dispute that the selected technologies are among the most promising in energy, bioengineering, robotics, imaging and other technological fields.

KISTI and the Data Science Lab used big data and deep learning, a branch of machine learning in AI that is capable of learning unsupervised from unstructured or unlabeled data, in selecting the 10 most promising from among so many technologies.

In the process, they analyzed 16 million pieces of data from research papers on science, technology and human science that have been published in the world during the past 12 years. Based on their findings, they developed 4,500 subject categories.
 

Then they quantified network data, research content and research fields by category with the use of AI, developed a prediction model by employing deep learning and produced a list of technologies, advances in which would be the most notable after seven years.

The use of deep learning in modeling is one of the most significant achievements made in carrying out the prediction project, say participating researchers.

Ko Byung-yul, who heads the Future Technology Analysis Center at KISTI, says, “We are able to raise the rate of accuracy for out prediction to 86.7 percent with the use of a deep-learning-based model. The rate of accuracy is about 30 percent when the conventional Delphi method is used.”

Data analyses are much more frequently used than before in developing research and development plans that include predictions about technologies, said Lee June-young, a KISTI senior researcher.

“Korea will have to increase investments in expanding the data infrastructure that can be readily put to use,” he said.