Semantic artificial intelligence at the service of the investigation

Today, investigators collect a large number of documents, which makes their use ever more difficult. Semantic intelligence is the solution to help structure information quickly. But it is still necessary that users have confidence in this technology…

Semantic AI is concerned with extracting structured information from unstructured data, whether text or speech, in order to understand the meaning of words beyond words. Thus, information expressed in different ways will have one and the same representation. However, for semantic AI to really bring benefits to the investigator, it must first ensure that users trust this technology.

Semantic AI, a two-faced technology

Semantic intelligence uses two different approaches: connectionist AI, based on methods such as machine learning and neural networks, which has the advantage of being available fairly quickly for the most widespread languages ​​and is very efficient for pattern recognition tasks such as matching similar behaviors; and symbolic AI, which uses rule-based systems and requires longer development times, but does not require a large corpus and allows the link to the source of the extracted information to be maintained.

Trusted semantic intelligence by design

To ensure both trust in the tool and speed of debugging, at Deveryware, we adopt a hybrid approach that combines connectionist AI to speed up system debugging time and symbolic AI that allows work on restricted corpora and precise monitoring of the source of the information extracted as well as the reasoning used.

In a field as sensitive as legal investigations, offering a tool in the form of a black box that would do all the work automatically comes up against many constraints in terms of legality, quality of results, traceability and therefore trust. Instead, we offer aids in the form of various tools that investigators understand how they work, and that they can use according to their needs. The main semantic extraction tool allows, according to criteria defined by the user, to highlight the documents which contain the most interesting information, so that it treats them in priority, before treating the others. documents. This allows, when the time of the investigation is constrained, as in the context of police custody, to quickly find the relevant elements. The advantage is also that the user can choose to “validate” the extracted information as correct in order to generate a structured representation of the information thus ensuring that the investigator does not forget any information that he has encountered.

Co-constructing reasoning of trust

Knowing the origin of the extracted information, the user can then, thanks to semantic intelligence, obtain graphic representations such as relationship graphs, timelines, projections on maps. It can also compare this information with other structured information such as databases, GPS position fixes, etc., thus ensuring a global view of the elements of the investigation.

Hybrid semantic intelligence, as offered by Deveryware, is a real support for forensic investigations allowing to accelerate the process of extracting and structuring relevant information, while maintaining transparency as to their sources, to promote a dignified AI of confidence.

This contribution by Aurélie Pradelles is part of our report on AI and security published in issue 9 of ActuIA magazine, available on newsstands and online.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *