Telling real news from fake news has become a major issue, not only for individuals but for society as a whole. But how do we untangle true from false? And whom can we trust to tell us whether something is real information or a hoax?
Many fact-checking sites have existed for a long time – AFP and many media outlets have created their own. A new tool is now added to the list: Vera, an A.I. chatbot (or conversational agent) developed in 2024 by LaReponse.tech, a French NGO. What is the point of creating an additional tool, and what does it add? I tested it to find out…
Vera can be consulted in two ways: by message on WhatsApp (free of charge) or by phone on 09 74 99 12 95 (local call rate). It can communicate in more than 80 languages. It is an A.I., so it answers a question by consulting sources and compiling them. Like the others, it is therefore not “intelligent”, but does research for us…
However, Vera has several advantages.
First, to overcome the structural problem of artificial intelligence (which feeds indiscriminately on information and hoaxes), Vera has access to a limited number of sources. Sources that were selected for their reliability and several of which are signatories of the European charters IFCN (International Fact-Checking Network, created in 2018) and EFCSN (European Fact-Checking Standards Network). These are fact-checking sites (more than 150) and reliable, recognised media (more than 350) from around the world (mostly from Europe and North America, but all continents are represented). Their list can be consulted on the website www.askvera.org.
Vera always cites its sources. At the end of each answer on WhatsApp, there are links to the articles cited. On the phone too, Vera “reads out” the URLs of the links. This systematic citing of sources increases its reliability because it lets you check the information provided for yourself. However, on the phone, having hyperlinks read aloud is rather indigestible and hard to use: offering to send them by text message or via WhatsApp might be a more suitable option.
The other bias of conversational agents that LaReponse.tech sought to avoid is the problem of “hallucinations”. These consist of providing an answer to a question at all costs, even if it is invented or assumed. A.I.s are therefore generally programmed never to say “I don’t know”. Designed to avoid these hallucinations as far as possible, Vera can answer: “There is no public documentation on this subject.” or “I did not find any reliable information to answer your question”, or even “The sources available to me mention nothing that would allow me to conclude that…”. It must provide answers that are as factual as possible. Users may therefore be told: “There is nothing to indicate that…” or be given an answer stating that “the available data on the subject have significant methodological limitations” or that “he was charged but was acquitted”. It also indicates what is controversial, setting out the points of view of the different parties involved, for example, always providing the sources.
Vera therefore seems to be a fairly reliable conversational agent for checking textual information, with the added advantage of being very simple to use and transparent. It also complies with the GDPR regulations in force. However, no A.I. can be 100% sure, so it is wise to consult the sources it offers for yourself. Finally, we can keep in mind that using A.I. is infinitely more costly in natural resources than a conventional internet search, and so keep using recognised fact-checking sites (picked from Vera’s list of sources, for example!), and only use Vera when the information is harder to find… A good number to keep in “memory”, whatever happens!
H.B.
https://www.askvera.org/#comment-ca-marche
https://pia.ac-paris.fr/portail/jcms/p2_4630140/vera-le-numero-pour-verifier-les-faits
