AI Agents: A New Stage in Automation
Since the emergence of ChatGPT in 2022, generative artificial intelligence has revolutionized both professional and everyday life. Today, a second generation, embodied by AI agents, is opening up new perspectives while also raising strategic and ethical questions.
(Source: Le Figaro, November 30 – December 1, 2024)
AI agents represent a major evolution compared to conversational assistants. Capable of automating repetitive and complex tasks, these tools are being adopted by companies such as Microsoft and Salesforce. Satya Nadella, CEO of Microsoft, speaks of an “agentic world,” where these technologies would operate autonomously 24/7 to maximize productivity.
A striking example is the tool developed by McKinsey to accelerate customer onboarding, reducing administrative tasks by 30%. Salesforce, for its part, aims to deploy one billion AI agents by 2025, targeting sectors such as marketing and customer service.
However, challenges persist. Yannick Léo, consultant at Emerton Data, emphasizes the need to create accessible models with reduced computing costs. These agents must also integrate seamlessly with existing business processes.
Cautious Adoption of AI in French Companies
(Source: Le Figaro, November 30 – December 1, 2024)
In France, while generative AI shows promise, its adoption remains limited. Crédit Mutuel, for example, saved 2.4 million hours thanks to a relational AI platform. L’Oréal uses AI assistants for 24/7 product recommendations. However, Jacques Pommeraud, CEO of Inetum, observes that only 26% of French companies are fully utilizing this technology.
The main barriers include high costs, insufficient data quality, and fragmented technological infrastructures. As Jean-Baptiste Bouzige, president of Ekimetrics, points out: “Adoption requires a profound transformation of business processes.”
French Start-ups: Betting on Innovation
(Source: Le Figaro, November 30 – December 1, 2024)
The start-up LightOn, a pioneer of generative AI in Europe, recently went public on Euronext. Having raised 13.5 million euros, it aims to finance its development while gaining visibility. Laurent Daudet, co-founder, states that this strategic choice will also secure the next three years of growth.
However, this approach remains risky. Pierre-Yves Meershman, from the Daphni fund, emphasizes that LightOn will need to maintain a stable share price and meet its financial promises to convince investors.
AI and Creativity: From Music to Education
(Source: Sciences & Avenir, December 2024; 01Net, November 20 – December 3, 2024)
Artificial intelligence is not limited to industry and commerce. In Dresden, a symphony orchestra was conducted by robot conductors, a world first. This project, the result of two years of collaboration between the University of Dresden and engineers, shows that technology can enrich art while paving the way for innovative collaborations.
In education, the Gingo software, developed by Compilatio, reduces teachers’ grading time tenfold. Thanks to AI, it provides grades accompanied by detailed comments while allowing teachers to validate the results. This tool, still in the testing phase, illustrates how AI can lighten professionals’ workload.
The Limits of Generative Models: Hallucinations
(Source: Sciences & Avenir, December 2024)
Generative AI models, while impressive, are still prone to errors, particularly hallucinations. A recent evaluation revealed a 27.2% hallucination rate in responses from Palm-Chat, Google’s model. Andy Zou, from Carnegie Mellon University, proposes an innovative method, “artificial linear tomography,” to analyze and improve these models.
These errors highlight a fundamental challenge: ensuring the transparency and reliability of AI. As Amy Lockey from ServiceNow reminds us: “AI is only as good as the data that underpins it.”
The Race for AI Search Engines
(Source: 01Net, November 20 – December 3, 2024)
OpenAI is challenging Google on its own turf with SearchGPT, a search engine based on generative AI. Unlike traditional search engines, it provides direct answers in natural language. However, this technology is still in its early stages and will need to prove its ability to compete with Google’s precision and scale.
Google is fighting back with “AI Overviews,” an enhanced version of its search engine enriched with AI. The competition between these two giants could transform how we search for information online.
Regulation and Ethics: The Challenge of Explainability
(Source: Le Figaro, November 30 – December 1, 2024)
Faced with the acceleration of use cases, companies must comply with regulations, particularly the European AI Act. BNP Paribas rigorously tests its AI tools before deployment to ensure their reliability. Matteo Dora, CTO of Giskard, compares adopting an uncontrolled generative model to “driving a car at full speed without a seatbelt.”
Meanwhile, initiatives like the Earth Species Project showcase another, more ethical face of AI. This project aims to decode interspecies communications using machine learning techniques.
