Interview with Urs Widmer, CEO Digital Learning, CREALOGIX

“Video is the new text” was Koemei’s credo. We have taken over the Artificial Intelligence (AI) technology of the Swiss start-up and spin-off of the research institute IDIAP (partner of the Swiss Federal Institute of Technology Lausanne EPFL). I asked Urs Widmer, CEO of the Digital Learning division, what Artificial Intelligence can do for learning today and where future potential lies.

Urs Widmer, CEO Digital Learning CREALOGIX

 

Jasmin Epp (JE): Artificial Intelligence, Machine Learning, Deep Learning – we encounter these terms everywhere today. What is it about the buzzwords and how do they actually differ?

Urs Widmer (UW): It is true that artificial intelligence in all its forms is not only popular among “nerds” (which is a positive term for me, by the way), but also in the business world. With good reason: Developers of applications based on artificial intelligence take advantage of human learning behavior to automate and optimize processes. Every computer program that behaves intelligently from a human point of view has artificial intelligence.

Roughly speaking, there are two ways to teach software how to think: Either the system follows rules and learns to apply them to the data material better and better, or it is fed with an immense amount of “good decisions” and recognizes according to the principle of trial and error how a certain task can be solved best. We call this “trial and error” principle Machine Learning. So it’s part of Artificial Intelligence.

A subset of machine learning is deep learning.

A very complex, data-intensive procedure with which neural networks simulate human decisions. The neuronal networks are computing units with artificial nerve cells that are connected to each other to form circuits. These “nerve cells” only send a signal to the computer when the sum of the input exceeds a certain threshold value. An example: It takes an incredibly large amount of data for artificial intelligence to learn to reliably distinguish the image of a dog from that of a wolf or fox. So this is only worthwhile if the benefits on the business side are correspondingly high.

JE: We now have a technology based on automated speech recognition. Machine learning with neural networks and natural language processing (NLP) was developed with enormous effort. Eight years of research and investment of around 20 million Swiss francs. What can the technology „do“?

UW: The solution includes several AI elements: firstly ASR (Automated Speech Recognition): It is used to transfer video and audio recordings into text. Siri, the Android version, Google Now, and other programs can do the same, but our software outperforms speech recognition quality. Secondly, NLP (Natural Language Processing): This analyses the content of the texts and recognizes and indexes key terms. And thirdly, deep learning by means of neural networks: This allows the texts to be assigned to so-called concepts on the basis of content analysis. Unlike Google and the other big names, we bring this solution to businesses. For confidential applications there is the on premise solution for companies.

JE: When we move throughout the German-speaking world, we hear a multitude of dialects. Especially in Switzerland.

UW: This is the biggest challenge in any language area, but in this respect, our machine learning solution is also very far-reaching. The speaker is not restricted to the use of a particular vocabulary. The standard solution contains approx. 80’000 words, the Advanced solution is extended by company-specific lexicons. The vocabulary and taxonomy is constantly being expanded by means of deep learning.

JE: „Video is the new text.“ However, we are talking about audio files. Why?

UW: This difference does not matter. The software treats video files as audio files. Cineasts may forgive me: We assume that the meaning of a video is conveyed via the soundtrack. Especially when it comes to learning content.

The idea of treating recording sequences like text has an enormous explosive power. Automated speech recognition produces texts that are then analyzed and tagged using artificial intelligence tools and natural language processing. This makes it possible to transfer the principle of Google search to videos and audio.

With this procedure, the University of Geneva has prepared more than 5,000 hours of lectures in a user-friendly way. For example, when students search the university’s content for exam-related content, texts as well as audio and video files are displayed in equal measure. The clou: You learn exactly to the second where in the lecture the topic you are looking for is dealt with and can jump directly to it.

JE: If only this technology had existed in my university days… This would undoubtedly make learning much more comfortable and research can be carried out in a targeted manner, but does it also pay off in the area of digital learning?

Educational institutions can use their own content sustainably, for example by researching the content already available for the preparation of a new teaching unit. User behavior can also be analyzed. The offer can thus be much better adapted to the learners. For example, if a company is often looking for ways to create a particular function in Excel, it is a good idea to offer a training course.

JE: This will help you to bridge the gap between educational institutions and digital learning in general. Can you think of other applications?

UW: We see a further main application in in-company training and further education. For our corporate clients, we will integrate these functionalities into the existing solution of the Swiss Learning Hub.

I also see great potential for digital banking (article will follow shortly here in the blog). For example, recorded calls can be automatically checked by customer advisors for compliance. Or if a client searches online for details of a particular investment product, the appropriate sequence would be displayed in the explanation video.

JE: Speaking of the market: What will happen to the technology now?

UW: By integrating artificial intelligence into both digital learning and digital banking, we have a key technology for the next generation of solutions. We are currently setting up our own team with our specialists to further develop the solution. We will offer them on premise and cloud-based according to the Software as a Service (SaaS) model. As for our entire product range, we also guarantee absolute data security for the AI-based product.

JE: We have now talked around the topic a little bit. Under what name do we offer the Artificial Intelligence tool?

UW: We presented the product SPEXIAN at this year’s Education Forum on the 19th of September, 2017 at Trafo Baden. By the way, interested parties can register for a demo account anmeldento get an idea of how it works.

JE: So it remains exciting – and it was also exciting to talk to you about the potential of Artificial Intelligence. Thank you for the interview.