Innovation is one of those topics that often appears as much art as it is science. The very nature of experimentation underlines the unpredictability of the whole process, with the difficulties inherent in understanding which ideas will thrive in the marketplace and which won’t.
So you can imagine that when researchers claim they have developed an algorithm that can predict which innovations will succeed, my ears pricked.
In a recently published paper, Arizona State’s Nadya Bliss describes her work in creating an algorithm that she believes can detect the emergence of innovation in research.
Her model utilizes insights and concepts from electrical engineering, applied mathematics and graph theory to try and identify patterns within large networks.
“Analysis of networks is basically analysis of entities and relationships among them–for example, people and their friends and how they’re interconnected,” she says.
The research team focused their attention on the citations from over 300,000 academics in the developmental biology field.
This data was then cross referenced with extensive data on innovations emerging in this field over a period of time dating back to the 1960s. This included the dates of each breakthrough and the researchers involved in the breakthroughs.
This data was used to create a filter that can identify patterns within the citation network that the authors believe can be used to predict when innovative breakthroughs will occur.
When the algorithm was tested out on data from 1969 to 1980, the results were largely positive, with the algorithm identifying a number of key individuals that were involved in key breakthroughs during that time.
The algorithm was tested again, this time from data ranging from 1990 to 2000, again with positive results. The authors believe that their algorithm therefore provides a relatively accurate ‘formula for innovation’.
At the moment, the authors are limiting the scope of their findings to things such as directing funding to the right people at the right time.
“One application of this could be working with NSF to continuously track publications and apply the filters to these large networks and see where there are emerging patterns, and maybe detect them before they’ve emerged and identify those as areas of potential in the scientific community,” they say.
Broader application would obviously require much more detailed testing of the algorithm, which is something the researchers are planning to do over the coming year. They believe however, that their findings provide valuable evidence of the importance of working across disciplines.
“One of the things you actually see in publication networks is that a lot of times when there is a major change to the field, there is a set of fields that are touching each other, so authors from different areas end up working together,” they conclude.
So whilst the algorithm isn’t yet capable of predicting when the latest hot consumer product will emerge, it does nonetheless providing an interesting insight into the way breakthroughs occur in scientific research.