Artificial intelligence: significant advances in 2025

A true technological revolution, artificial intelligence should enable notable progress in many sectors, notably health.

What progress will be made in AI in 2025? Consulting experts and chatbots agree that this year will be the year of the advancement of a very particular type of AI: one capable of making decisions and carrying out tasks without human intervention.

Workflows radically changed by AI

According to the AI Predictions report published by consulting firm Deloitte in 2025, 25% of companies already using AI will be ready to deploy intelligent agents by the end of the year. The firm specifies that this figure should reach 50% by 2027.

This year will see “workflows will fundamentally change”with AI taking over administrative tasks, but with human supervision, the report continues.

According to predictions from auditing firm PricewaterhouseCoopers, these tasks could involve customer inquiries, first drafts of software code or even turning design ideas into prototype projects.

“In many cases, AI will take over routine or repetitive tasks, freeing human workers to focus on strategic and creative activities”also confirms ChatGPT.

The commercial expansion of artificial intelligence agents is expected to continue into 2025. The International Data Center estimates that global spending on AI will reach approximately $632 billion (€605.1 billion) by 2028.

Narrow AI, soon to be used in the medical sector

In 2025, progress is expected in the field of“Narrow AI”a type of AI particularly suited to medicine, according to Catherine Breslin, founder of AI consulting firm Kingfisher Labs.

“It’s not necessarily difficult to make AI work in a specific domain. It just takes a little work to figure out what’s useful in a particular domain,” she specifies.

According to IA Perplexity and ChatGPT, narrow AI could be particularly useful in the development of drugs and products.

According to Kate Devlin, professor of AI and society at King’s College London, one of the advantages of narrow AI is that companies can use small or medium-sized language models to train it, making it possible to ‘use fewer resources over time.

30% AI in all our devices

Many companies have already started marketing smartphones using AI. According to the report from the consulting firm Deloitte, generative AI should equip around 30% of devices by now and even 50% of computers by the end of next year.

According to Catherine Breslin, in the long term, AI uses great computing power today, this could be reduced in the future even without an internet connection.

“ChatGPT or Meta’s Llama are very large models which require very significant computing power to work, but not everyone has this computing power. They also need to be connected to the internet, which does not “is not ideal in many cases.”

According to Catherine Breslin, this is a phenomenon that has already started. With its Phi-4 model which excels in complex reasoning, Microsoft has found an optimal compromise between performance, speed and resource savings.

The rise of multiple media

By 2025, AI models will be increasingly able to generate different types of content, such as text, images and speech at the same time.

This type of AI system, called a multimodal system, processes information from text, images, sounds and videos to give users a more complete answer to their questions or to produce media.

For example, multimodal AI could be used to analyze market commentary videos, taking into account tone of voice and facial expressions to give people a “more nuanced understanding” of investor sentiment in the economic market, explains Google.**

Multimodal AI could also analyze data such as noise and vibration in a manufacturing plant to proactively respond to needs on the ground.

The latest version of Google’s Gemini 2.0 chatbot, which processes text, images, audio and video, provides a first glimpse of multimodal AI.