CAIRN.INFO : Matières à réflexion

1Traditionally, human-computer interaction (HCI) has been a sore point and thus the subject of intense research. Desktop computing and the subsequent invention of the mouse and graphical interfaces gave a much-needed push to HCI, but admittedly a graphical interface and a “haptic” way to interact with it have been the golden norm during the last two decades at least. Still, complex analysis tasks and workflows demand an immense amount of scripting, and know-how that limits deep interaction between computers and humans to an elite few. On the other hand, it is evident that there is very small room for misunderstandings in this communication between machines and humans ! Indeed, when people speak about computing machines, it seems that they are able to minimize miss-information. Of course it is clear that structuring discussion on a topic that is not philosophic but rather more technical serves as the main root cause.

2With the emergence of powerful artificial intelligence (AI) technologies such as Natural Language Processing, Speech to Text, and in particular knowledge representation, new horizons open. In particular, machines can now codify in much greater fidelity human language and extract information and knowledge out of it, from spoke or from written word. Today, we are able to generate vast knowledge spaces that result from the ingestion of structured databases as well as a humongous number of documents, including scientific literature. Again, traditional ways of interacting with literature have been mostly through search (using a computer) and then print (again using a computer and a peripheral). What if AI can deeply look into the knowledge represented in the vast knowledge graphs and then allow humans to dynamically interact with this knowledge ? Imagine the case of a human debate : arguments posed, full with facts that need to be crosschecked. In the past, this was the preferred domain for scholars like lawyers. But, today machines can crosscheck facts for us – essentially, transforming a human posed argument almost into a checkable logical sentence.

3A recent stellar example from IBM Research is the Debater system. The system relies on hundreds of millions of text documents (or snippets) that contain arguments such as “Pre-school education boosts school performance later on for children that take part in it”, or “legalizing light drugs use may be beneficial for society”. Thus, users can pose questions or statements and the system is able to :

  1. mine through millions of arguments to find a match or a counter match ;
  2. offer the argument to the user and then drive a dialogue.

4The latter is really key as it offers a completely new way to interact with knowledge through computers. That is, humans would traditionally use the painfully long process of carefully reading through a corpus, taking notes and then formulating an argumentation basis. Obviously, such an approach is impossible to scale on millions of documents. In fact, as early as the mid 1990s, just a few short years after the invention of the World Wide Web, it became evident that we needed tools to search this vast, and constantly changing/expanding repository. Thus, semantic-based approaches led to the development of search engines such as Yahoo ! and others. There was indeed some progress, but it was not until the invention of Google search that we saw an explosion of the search capabilities. Suddenly, the computer became our portal to the Web – and thus to knowledge. Still, the interface is quite static and essentially a one-way street. Humans need to pose interesting questions and the system will respond with what it “thinks” are interesting answers in the form of Websites. What the Debater system does is that it allows us not only to go deeply into the meaning of the content of a Website but also to suggest interesting and relevant additional arguments to consider.

5Now, let’s take the next step. Let us assume that all the relevant facts and all the relevant relationships between them have been extracted from the corpus (thus, not just high-level arguments) and these data have been organized in a knowledge graph : for instance, medical literature, with diseases and symptoms, with patient cases and medical exams results, or even narratives of patients with respect to how they feel using their own words. Can we develop a system that will “converse” with medical practitioners and help humans reach must faster conclusions ? We now have strong evidence that the answer is a strong yes ! For example, IBM Research has been collaborating with one of the biggest telemedicine providers in Europe to bring to the public an AI-based medical triage system.

6Consider now a system like that offering to help communication between humans. It is conceivable that clarity of communication will increase ? We believe the answer to be clearly positive, since any type of communication can be automatically augmented with supporting arguments on all points of view.

7However, notice that we gave emphasis on the term “converse” between humans. This is to amplify our belief that HCI needs to assume a human face so that it can facilitate much better the discussion between humans (or of course any form of communication between humans : think of the case of air-control operators communicating with pilots, in such a way that mistakes can be automatically avoided). Thus technologies such as text to speech, voice tone analysis, sentiment analysis have been essentially at the very core of AI research across the Industry and academia. We have key examples in insurance, banking and hospitality – to name just a few – in which AI, thus the computer, assumes a human voice and face (e.g. medical assistant, online banking systems, etc.). For instance, we have already developed systems that help callcenter operators guide a discussion with a customer in a much faster, accurate and fulfilling result. We strongly believe that this will essentially make the computer as we know it “disappear” while dialogue between humans is enhanced. AI advances will allow it to be everywhere and humans to interact with it in completely human terms.

8Obviously these advances are already triggering major changes in the way we work or entertain ourselves – and in the way we live in general. Human creativity, unlike machines, is boundless : thus the only safe prediction to make is that the way we interact with knowledge and thus with computers and therefore between ourselves is starting to completely change !

Costas Bekas
Dr. Costas Bekas, Distinguished Researcher & Manager IBM Research (Zurich), is responsible for foundational research in AI spanning areas that include ML/DL, knowledge extraction and representation, new computing paradigms for AI, with applications in healthcare & life sciences, materials discovery and robotics. Costas Bekas studied at the pioneering Computer Engineering & Informatics Department of the University of Patras in Greece. He received B. Eng., Msc and PhD diplomas in 1998, 2001 and 2003 respectively with Prof. E. Gallopoulos. In 2003-2005, he worked as a postdoctoral associate with Professor Yousef Saad at the Computer Science & Engineering Department, University of Minnesota. Dr. Bekas’s main focus is in cogntitive systems and their impact in industry, science and business. His research agenda spans machine & deep learning, large scale analytics, HPC and very large scale distributed systems. Dr. Bekas is a recipient of the ACM Gordon Bell Prize (2013, 2015), and the PRACE Award (2012).
Cette publication est la plus récente de l'auteur sur
Mis en ligne sur le 01/10/2019
Pour citer cet article
Distribution électronique pour CNRS Éditions © CNRS Éditions. Tous droits réservés pour tous pays. Il est interdit, sauf accord préalable et écrit de l’éditeur, de reproduire (notamment par photocopie) partiellement ou totalement le présent article, de le stocker dans une banque de données ou de le communiquer au public sous quelque forme et de quelque manière que ce soit.
Chargement en cours.
Veuillez patienter...