The McKinsey Global Institute writes that ‘artificial intelligence (AI) is poised to unleash the next wave of digital disruption.’1 Faster computers, increasingly sophisticated algorithms and vast amounts of available data – the fuel of AI – have opened the gates to a true AI revolution.
Many industries are already feeling its disruptive effects, from financial services to carmakers and telecommunications companies. The technology also holds great economic potential for firms that produce and adopt it, making AI an attractive theme for investors.
AI taps into unconventional data sources
The recently launched STOXX AI Global Artificial Intelligence Index tracks companies that invest heavily in AI research and development, using AI techniques in order to identify them. Proprietary AI algorithms developed by our partner Yewno, an award-winning AI technology company, help us determine which companies hold substantial intellectual property (IP) in AI-related technologies.
The technique’s main component is called a ‘knowledge graph.’ On a basic level, it is a framework that turns unstructured data into quantifiable metrics. In the case of the STOXX AI Global Artificial Intelligence Index, the knowledge graph is fed all granted patents from a number of US and international databases; it ultimately returns a list of International Patent Classification (IPC) categories that the algorithm has determined are AI-related. The output serves as a basis for calculating AI IP scores, which determine whether or not a company is included in the index.
Without AI techniques at our disposal, analyzing such a large amount of patent data would simply not be feasible. The world’s patent databases include documentation on a very high number of granted patents. Humans would not be able to complete the task due to the sheer volume of information. Traditional algorithms using keyword searches would not be up to it, either, as they require keywords to be manually assigned, another impossible task given the amount of data involved.
Turning unstructured data into knowledge
So how does a knowledge graph work? Essentially, it scans documents and relates them to one another, similarly to what a search engine does with individual pages. In order to do so, however, it needs to extract meaning from the text instead of simply identifying which pages are connected via hyperlinks.
To do so, the algorithm scans all patents and extracts a set of potentially associated concepts for each one. The concepts are then disambiguated, i.e. the algorithm checks whether a concept such as ‘apple’ refers to the fruit or the computer company. It does so from context; if a paragraph mentions the taste of an apple and then compares it to that of a pear, it is probably referring to the foodstuff whereas a mention of Apple’s IPO likely refers to the company.
Connecting the dots
Crucially, the extracted concepts are not simply stored as isolated pieces of information. Instead, the scanned patents are mapped in a graph relative to each other, allowing the algorithm to discover relationships among the concepts associated with them. The knowledge graph thus creates an inferential ‘semantic space’ with clusters of concepts and their potential inferences – a web of knowledge.
For the STOXX AI Global Artificial Intelligence Index, the algorithm then picks concepts it has determined to be related to AI, finds all the patents that contain those concepts and selects their IPC categories. Companies which hold patents belonging to these categories are then considered for inclusion in the index.
Unleashing the power of big data
Using human-like strategies but at exponentially greater scale and speed, Yewno’s AI-based knowledge graph thus helps in the selection of constituents for the STOXX AI Global Artificial Intelligence Index. This makes the index a real-life example of how AI technologies can unlock the potential of big data for investing.
1 ‘Artificial Intelligence – The Next Digital Frontier?’ McKinsey Global Institute, June 2017.