I wanted to do this second article about the summit of the action for the AI to which I attended last week, not to put a layer on the political aspect of the thing, but rather to talk to you “Initiatives” (companies and NGOs) that I was able to meet quickly there.
This Paris summit was above all a Global conversation on AIwhere all voices, not just those of Washington or Beijing could be heard. And on this occasion, France has decided to invite players from the sector around the world & MLDR; At 100% Tech for Good, that is to say who use AI for the common good. I found that it was good, rather than having a mega stand openai, google, meta or I do not know what, there were lots of projects already presented for the forum for peace, that I I found super interesting and promising.
I start with situationhub of Gannet, an innovative initiative that puts AI at the service ofhumanitarian. This project shows how artificial intelligence can help analyze and understand crises in real time. For example, Situationhub Use AI to collect and analyze data from multiple official sources during humanitarian crises. The platform transforms this data into usable insights which then help organizations to better coordinate their interventions in the field. For example, in Sudan, SituationHub allows humanitarian actors to make faster decisions while being better informed.
I also crossed pluralism, which in another style, developed an engine ofpolitical analysis advance. Their tool scans in real time the speeches of more than 3000 public figures (TV, radio, certified social networks) to offer a clear vision of public debate. Already active in France, Germany and Uruguay, Pluralism has also associated with media like BFM, M6 and the Süddeutsche Zeitung. Its ambition is to become the global reference in terms of political transparency. Magic Lemp, the company behind this project, relies on a “made in France” AI to strengthen our democracy. Via automated newsletters and Fact-Checking B2B tools, they thus help citizens, journalists and decision-makers to access diversified and reliable information. This project thus shows how AI can concretely contribute to a more transparent and healthier democratic life.
Diversifair is also an interesting project. It’s European and it attacks a crucial challenge: intersectional biases in AI. Basically, the idea is to form a new generation of experts capable of detecting and preventing hidden discrimination of AI systems. Because yes, biases are often more complex than you think! For example, some facial recognition tools that work well separately for black people or women can have up to 34% error when it comes to black women. To fight these dead angles, Diversifair therefore deployed 2 programs: Equality4Tech To learn how to bias, and Tech4Equality To prevent them upstream. This project shows how to build a fairer AI for everyone, whatever our gender, our origin or our social background.
In the Democratic Republic of Congo, the Hirondelle Foundation Also uses AI to combat hatred speeches east of the country. Their approach is clever since they use automatic natural language treatment to monitor local community radios and identify violent speeches. But they do not stop there, because in response to these hate speech, they produce and broadcast adapted radio content to neutralize these toxic messages. The project affects the vulnerable populations of Ituri and Kivus via an impressive network of 200 media partners, including 50 in the east of the country. The programs are thus disseminated in 5 national languages, with particular attention to include women and young people. It is, I find, a fine example of how AI can concretely strengthen social cohesion and local stability by giving access to reliable information!
Side health, an interesting project is Cervicaid. It uses AI to literally save lives. Developed by a Tunisian startup with Medntechtheir mobile application helps caregivers to detect cervical cancer early, in poorly served geographic areas at the medical level. The principle is ingenious: during the collar exam, the app analyzes the images in real time to identify possible anomalies. The project is in place in Rwanda, with the ambition to extend to all of Africa. What is great is that they directly involve African health professionals in the development of the tool to ensure that it really meets the needs of the field. Here again, AI can make preventive care more accessible to women in order to prevent this kind of cancer!
Mozilla was also there. I did not expect to see them there, but in fact it is logical, because they presented Common Voice which I already told you about here. As a reminder, it is a project that offers a different approach to creating vocal data for AI. Instead of plundering data on the net without consent or subcontracting at low cost, they bet on citizen participation to create a public and inclusive database. And it works, because with more than 750,000 contributors (including me !!) who have already given 30,000 hours of recordings in more than 140 languages, Common Voice has become the largest participatory vocal database in the world.
I also discovered Dalila media analysis tool boosted at AI which helps to see clearly in the current chaos of info and infox as they say. Launched in 2023 by SIREN Analyticshe quickly established himself as a precious ally of Fact-Checkers, journalists and investigators, particularly in France and in the Balkans. Basically, it monitors thousands of media sources (press, TV, social networks) to detect weak signals in real time such as the sudden propagation of false information, or the emergence of a sensitive subject, in order to behind Alert journalists or competent authorities. They also developed a tailor -made legal AI model which is capable of analyzing complex situations in international humanitarian law! It’s great!
There are also a little more ecological projects such as this partnership between Meta and the Wri (World Resources Institute) which maps the world canopy with a precision of a meter (yes yes, they can spot each tree!), Which is perfect for following reforestation projects. Or Neomiadeveloped in Côte d’Ivoire, which uses AI to analyze satellite images and fight against illegal fishing while monitoring the impact of climate change on the coasts. And then there is Gaiaa great project that wants to democratize AI in agriculture by offering open source tools. Gaia’s idea is to create a kind of ready -to -use kit with AI models specializing in agriculture, accessible to everyone for free.
I also crossed out, Prism Evala box that makes cybersecurity. Obviously it interested me, because it is one of my favorite subjects with AI. Basically, they develop tools to test the safety of AI models.
Their flagship project is called Bet For Behavior Elicitation Tool and I found it really interesting because instead of testing AI with fixed prompts, they dynamically explore their vulnerabilities, a bit like a super intelligent ethical hacker!
They have just published a classification which I think will interest you. Their approach is cool: first, they map the potential vulnerabilities with a library of prompt & other tips allowing to bypass the safeguards of the AI. Then they use an algorithm inspired by the ants colonies which simulates the behavior of a hacker to estimate how many attempts it would take to “crack” the model (yes, I put quotes). Crackering the model means getting around the security of the model and thus obtaining information as explanations on the creation of weapons, from incentive to violence, disinformation, hate speeches & mldr; etc.
And even if some models like Claude-3.5 resist better (it takes more than 400 attempts to derail them), we see that there is still work to make the LLM really safe. In short, I found it really cool, I invite you to go see their leaderboard. You will learn lots of things.
Here, obviously, I couldn’t see everything, there were too many things, but that’s for my little return to projects using AI for the good of the community. There were others of course And you can consult the list here.
Have a great day!
Source link
Subscribe to our email newsletter to get the latest posts delivered right to your email.
Comments