An universal collaborative space between artificial intelligence and natural intelligence through the Web - Interview with Fabien Gandon

  • Research
Published on May 5, 2021 Updated on May 17, 2021


Fabien Gandon - Senior Researcher and Research Director at Inria - 3IA Chair “Combining artificial and augmented intelligence technics on and through the web” 

#knowledge graphs  #semantic web  #hybrid approaches  #multi-agent systems


The Web the ideal candidate today for a universal collaborative space between artificial intelligence and natural intelligence 


Fig 1. Shaping hybrid societies of intelligence types on the Web [6].


The evolution of the Web (social media, link data, semantic Web [12], APIs) makes it the ideal candidate today for a universal collaborative space between artificial intelligence and natural intelligence [6]. In fact, in 2012 already, out of the 50 most prolific Wikipedia contributors, only about ten were humans. The Web is also an essential space for exchanging data and knowledge in any community (e.g. medicine and biology) and for connecting more or less smart objects (e. g. cars connected to the Web, home automation via the Web, etc.). This evolution requires the development of methods to manage the interactions between different forms of artificial intelligence and different natural intelligences. This chair is interested in formalizing knowledge models and designing algorithms to support these interactions and we have made progress on the following three axes of the initial proposal: 
 

Axis 1:  Obtaining knowledge for and by AI from the Web -- We proposed a method for predicting the availability of data at a given Web address in order to optimize data crawling and this method was awarded best paper at ESWC 2019 [7]. Inversely, we proposed a micro-service approach for publishing linked data from legacy APIs [9]. Both contributions allow to access more data as knowledge graph using Web standards. We also showed that such knowledge graphs could be for instance injected in the vector representation of general practitioners’ reports to improve hospitalization prediction by coupling learning with pre-processing reasoning methods [1][11]. This was done in the context of a CIFRE thesis with the local company SynchroNext, partner of the 3IA and in collaboration with the Department of Teaching and Research in General Practice of Nice, who designs the PRIMEGE medical database we used for our experiments. 


Fig 2. Google Dataset Search result page showing an example of how a SPARQL micro-service turned an API into a dataset that can be queried [9].
 

Axis 2: methods contributing to knowledge graphs and AI results evaluation -- In collaboration with another 3IA chair we are supervising a PhD on web-based representation and negotiation of uncertainty in published data [2][5][10]. The idea is to contribute a standard ontology-based representation and access means of the different types of uncertainty theories used by online datasets. More recently, we also initiated a new line of research on the evaluation of explanations produced by link prediction algorithms in knowledge graph for which, until now, there were no ground truth. Finally, in relation to the topic of knowledge graphs management, we now have a 3IA Affiliate Chair with the local company Mnemotix, rooted in knowledge engineering, semantic web [12] and knowledge graphs querying and reasoning, targeting problems of scalability, temporality, transformation, validation and analysis in relation to the open-source CORESE software.  


Fig 3. The CORESE semantic Web factory for knowledge graphs on the Web. 
 

Axis 3: Combining different forms of intelligence on the Web -- In this longer-term direction we started to study and publish on the reconciliation of the distributed AI architecture of multi-agent systems and the Web architecture [4][8]. The 3IA funded a visit and collaboration with the Saint Gallen University and we launched the ANR-SNF project HyperAgents focused on the same topic. This work is complemented by a 3IA post-doc using agent-based modelling through evolutionary game theoretical mechanisms with social learning to study how populations of artificial agents with different languages perform under different collaboration scenarios. At a more micro-level, we have just finished the MonaLIA project where we combined knowledge graphs, semantic Web, semantic reasoning and machine learning on images and their metadata to create tools that allow curators of museum visual art collections to identify and correct the annotations of the artwork as well as to improve the relevance of the content-based search results in these collections. These results are to appear in a journal paper soon. 

Of course all this research activity also goes hand-in hand with several educational activities including a course on “Integrating Semantic Web technologies in Data Science” at DSTI a local partner of the education component of the 3IA, and also three MOOCs online on FUN and Coursera teaching semantic Web and linked Data to several thousands of registered participants. 


Fig 4. Hypermedia MAS: Agents are situated in a shared environment designed as a distributed hypermedia application [8].


 

References: 

1. Raphaël Gazzotti, Catherine Faron Zucker, Fabien Gandon, Virginie Lacroix-Hugues, David Darmon. Injection of Automatically Selected DBpedia Subjects in Electronic Medical Records to boost Hospitalization Prediction. SAC2020 - The 35th ACM/SIGAPP Symposium On Applied Computing, Mar 2020, Brno, Czech Republic. ⟨10.1145/3341105.3373932⟩. https://hal.inria.fr/hal-02389918  

2. Ahmed El Amine Djebri, Andrea G. B. Tettamanzi, Fabien Gandon. Task-Oriented Uncertainty Evaluation for Linked Data Based on Graph Interlinks. EKAW 2020 - 22nd International Conference on Knowledge Engineering and Knowledge Management, Sep 2020, Bozen-Bolzano, Italy. https://hal.inria.fr/hal-02933190  

3. Hai Huang, Fabien Gandon. Learning URI Selection Criteria to Improve the Crawling of Linked Open Data (Extended Abstract). The 29th International Joint Conference on Artificial Intelligence (IJCAI 2020), Jan 2021, Yokohama, Japan. https://hal.inria.fr/hal-03064912  

4. Simon Bienz, Andrei Ciortea, Simon Mayer, Fabien Gandon, Olivier Corby. Escaping the Streetlight Effect: Semantic Hypermedia Search Enhances Autonomous Behavior in the Web of Things. IoT 2019 - 9th International Conference on the Internet of Things, Oct 2019, Bilbao, Spain. ⟨10.1145/1122445.1122456⟩. https://hal.inria.fr/hal-02289497  

5. Ahmed El Amine Djebri, Andrea Tettamanzi, Fabien Gandon. Publishing Uncertainty on the Semantic Web: Blurring the LOD Bubbles. ICCS 2019 - 24th International Conference on Conceptual Structures, Jul 2019, Marburg, Germany. pp.42-56, ⟨10.1007/978-3-030-23182-8_4⟩. https://hal.inria.fr/hal-02167174  

6. Bettina Berendt, Fabien Gandon, Susan Halford, Wendy Hall, Jim Hendler, et al.. Web Futures: Inclusive, Intelligent, Sustainable The 2020 Manifesto for Web Science. Dagstuhl Manifestos, Schloss Dagstuhl - LZI GmbH, 2021, ⟨10.4230/DagMan.9.1.1⟩. https://hal.inria.fr/hal-03189474  

7. Hai Huang, Fabien Gandon. Learning URI Selection Criteria to Improve the Crawling of Linked Open Data. ESWC2019 - 16th Extended Semantic Web Conference, Jun 2019, Portoroz, Slovenia. ⟨hal-02073854⟩ 

8. Andrei Ciortea, Simon Mayer, Fabien Gandon, Olivier Boissier, Alessandro Ricci, et al.. A Decade in Hindsight: The Missing Bridge Between Multi-Agent Systems and the World Wide Web. AAMAS 2019 - 18th International Conference on Autonomous Agents and Multiagent Systems, May 2019, Montréal, Canada. pp.5. https://hal.inria.fr/emse-02070625  

9. Franck Michel, Catherine Faron Zucker, Olivier Corby, Fabien Gandon. Enabling Automatic Discovery and Querying of Web APIs at Web Scale using Linked Data Standards. WWW 2019 - LDOW/LDDL Workshop of the World Wide Web Conference, May 2019, San Francisco, United States. ⟨10.1145/3308560.3317073⟩. https://hal.inria.fr/hal-02060966  

10. Ahmed El Amine Djebri, Andrea Tettamanzi, Fabien Gandon. Linking and Negotiating Uncertainty Theories Over Linked Data. WWW 2019 - LDOW/LDDL Workshop of the World Wide Web Conference, May 2019, San Francisco, United States. ⟨10.1145/3308560.3317074⟩. https://hal.inria.fr/hal-02064075  

11. Raphaël Gazzotti, Catherine Faron Zucker, Fabien Gandon, Virginie Lacroix-Hugues, David Darmon. Injecting Domain Knowledge in Electronic Medical Records to Improve Hospitalization Prediction. ESWC 2019 - 16th Extended Semantic Web Conference, Jun 2019, Portorož, Slovenia. pp.116--130, ⟨10.1007/978-3-030-21348-0_8⟩. https://hal.inria.fr/hal-02064421v2  

13. Dean Allemang, Jim Hendler, Fabien Gandon. Semantic Web for the Working Ontologist. ACM, 2020, ⟨10.1145/3382097⟩. https://hal.inria.fr/hal-02939606