Development retrospective — The steps taken towards AI-powered investor platform for Business Finland
Several years of R&D bear fruit
This article doesn’t describe a single customer case but a glimpse of Headai’s R&D history. Headai has come a long way to reach the state where it’s possible to create comparable language-based digital twins of different entities like textual descriptions of labor, industries, companies or persons. Over several years of development work, we have improved our ontologies and the core algorithms that create and visualize digital twins. Many projects with our customers have created an understanding of the possibilities and the challenges of data and language.
Headai is now bringing digital twin approach into the investment market by building AI features for Business Finland Connect investor platform. The aim is to open interesting and detailed views for global investors to scout potential Finnish companies to invest in. The data, collected from multiple sources, is refined by AI algorithms and brought to investors to give valuable insights. This will help their data-driven decision making.
The aim is to open interesting and detailed views for global investors to scout potential Finnish companies to invest in. –Harri Ketamo, Chairman of Headai
Where it all started
Let’s take a moment to look back at where we came from. Headai was founded in 2015 focusing on intelligent content discovery with a strong scientific background (15 years of research on learning & cognitive sciences). Right from the beginning, we started reading a lot of open data. It worked as a fuel for our AI which would use it to build digital twins on different knowledge domains. Our ontology, which uses self-organizing networks, improved along the way.
During 2016, we further studied the possibilities of textual Open Data and Natural Language Processing (NLP) methods (excluding deep learning) which would serve Headai development best. Our AI was already able to autonomously build training courses of desired topics and teach them to learners. This required not only expertise in adaptive learning, but also gaining contextual understanding of different topic areas through reading and modeling relevant open data.
Harri Ketamo, who has done some 20 years of research on learning sciences, gave a Ted Talk in late 2016 on how AI could help the global challenge in education. Understanding learning as a concept is crucial for those who want to build effective self-learning AI. You can find more on the subject of learning as a phenomenon in Ketamo’s article in Unesco MGIEP 2019.
The implementation that started paving the way for the Business Finland Connect AI was the Slush Shanghai 2017 matchmaking component done by Headai. It used our algorithms to link together attendees based on their textual description. By the time, Harri Ketamo and Peter Vesterbacka, the founder of Slush, had already discussed the idea of creating an AI matchmaking tool for investors and companies. The experiences from Slush Shanghai were good, but also pointed us two things: 1) We need more and better data for accurate matching 2) We should continue our ontology development.
Towards Talent Economy — AI to scout individual, organizational and national human capital
Headai invested a lot of effort to attend Sitra’s Ratkaisu 100 Challenge in 2017. Our mission was to create solution for national employment mismatch by using AI. We started modeling regional skills’ demand from open job data. We read the curriculums from education providers and modeled the skills they provide to understand the areal and national skills offering. Our algorithms calculated skills maps (digital twins) from both, demand and offering, which could be then compared with each other. This all would have taken thousands of man-hours of human labor.
As the world and labor market undergoes radical change, it is difficult to know what types of expertise will be in demand in the future. Headai harnesses artificial intelligence to find out.
— Sitra article Nov 2017
Even before the grand finale of Ratkaisu100, our solution was chosen as one of the three winners of an international innovation competition at the Future of Work conference, September 17, 2017, Spain. Innovations from around 50 countries took part in the competition.
In November 2017, Headai won the Ratkaisu100 competition gaining the prize of 500,000 euros. The award was presented to our team by Sauli Niinistö, the president of Finland. The victory opened a great opportunity to start further development of the solution for not only national needs but also for international markets.
Headai awarded by Sauli Niinistö, the President of Finland.
Headai victory was widely noticed in Finnish media.
The media interest after the victory was remarkable. All of the main news channels noticed the awarding. YLE, Finnish public service media company, did also an in-depth interview with Harri Ketamo (in Finnish) in early 2018.
In 2018 Headai gave lots of presentations. One among many was presented by Eero Hammais at the ISKU Tallinn seminar, which was part of the Estonian 100-year-anniversary seminar series.
Technically, Headai’s Talent Economy is about creating multidimensional semantic decision models, from open and public data written in natural language. Artificial Intelligence is able to process this knowledge as an analyst or researcher would and visualize the outcomes as comparable skill maps.
In spring 2018, Headai published Microcompetencies, the reference UI for everyone to try out our solution’s freemium version. The core development was still in the algorithms and the UI was like a window to see the possible organizational and individual use cases of our AI-features.
By getting the Headai license, all of the AI operations could easily be implemented into organizational customers’ systems through our API. For a single freemium user, the interesting part was the option to insert CV or copy-paste LinkedIn description for AI to find out the skills that are related to the textual description. Another useful function presented, was the skill suggestions based on the current skill profile. It reminded the user of possible missing skills that could be found in his / her repertoire.
“You know more than you think you do. You’ll get whole new perspectives on it.”
— A test use comment from a CV workshop in Ohjaamo Pori, Finland 2019.
Our organizational customers have used the solution to simulate and run predictive analysis on their data, as well as the areal labor market data. One of the good use case examples is presented in Metropolitan Universities of Applied Sciences customer case article. They use Headai tech to optimize future supply of education to match working life needs. In numbers, it means optimizing 181 study modules and 10749 course contents for 10,000 skills sought in the Helsinki area.
For the rest of the year 2019, we have our calendars booked with interesting customer projects. We are already waiting that we can tell more about the process with AI features in Business Finland Connect. Until then, let’s do some algorithms!
Lessons learned on data and language
- Language evolves all the time. It changes differently in each industry, culture or other domain. We have to keep up with the change by continuously reading through new open data. In other words, always refreshing our contextual understanding.
- The data quality has to be good and detailed enough for AI to reach good results.
- AI can help in decision making — do the background work routines which are not possible to do with manpower and make the actual decision making more transparent and explainable.
- Headai’s semantic AI is capable of opening up the reasons behind its decisions unlike deep learning solutions — most of the Natural Language Processing being done with them. That makes such applications black boxes, narrow, dependent on training sets, and so disables secure and transparent decision making.
General Semantic AI — Headai tech stack in a nutshell
Headai’s cognitive artificial intelligence, AI, is 100% owned by the company and in commercial use internationally since 2015. It is based on self-organizing semantic networks which enables more complex reasoning (with Natural Language Processing) than traditional ontologies (eg. O*net) and methods. It structures complex data (e.g. job market data) into a (visually) comprehensible form with its one-of-a-kind AI system. The AI service can also be utilized via the REST API (Application Program Interface). APIs enable a standard mechanism to share data and functionalities.