The launch of ChatGPT in 2022 was the catalyst for a series of changes driven by Generative AI (GenAI), enabling a wide range of technological innovations worldwide. Naturally, there are many speculations about what still lies ahead in the coming years regarding the topic, but in 2025, what can be expected in terms of AI trends?
According to the specialist in the area, Paulo Henrique de Souza Bermejo, who has a post-doctorate in Innovation from Bentley University, in Massachusetts/USA, and Executive Certification in Strategy and Innovation, through MiT, some issues in this sense are already being discussed, both in the market and in the field of research.
He highlighted that regarding 2025, one can expect, for example, a more capable and personalized Generative AI. According to him, the major creators of large-scale language models (the so-called LLMs, from English "large language models"), which are a kind of core for generative AIs, continue to evolve rapidly. "Investments are being made to make these models more robust and specialized, capable of generating even more accurate and creative content, as well as functioning in highly specific contexts. Additionally, new features are being added to allow users, including non-programmers, to create their own AI agents by training them with their personal data. This type of functionality, initiated with ChatGPT, is expected to expand to other platforms and LLMs, including those developed by Anthropic, Meta, and Google," he noted.
When it comes to health, Paulo stated that especially diagnoses could become more accurate. "Investments in devices and systems have expanded the capacity and accuracy of AI-based medical exams, directly contributing to improving healthcare services, especially in less privileged regions. This ranges from medical diagnoses to the interpretation of tests and treatment recommendations," he noted. From another perspective, AI-based tools should predict diseases before they manifest, using data from wearable devices (such as smartwatches and other sensors) and integrated medical histories.
Another point raised by the specialist concerns greater integration of advanced personal assistants into daily life. "AI will be integrated into devices such as augmented reality glasses, autonomous cars, and smart appliances, optimizing everyday tasks. This is already becoming a reality in some contexts, through AI agents," explained Bermejo.
And, when it comes to education, it will be more personalized and engaging. "Educational platforms should expand their resources and personalize learning paths based on students' cognitive profiles and interests. For example, students with higher aptitude in mathematics will find it easier to engage in training related to exact sciences, while those interested in arts may excel in courses focused on creativity," he emphasized. In general, according to the specialist, the tools will become increasingly accessible. "Smaller companies and individuals will have access to powerful AIs through low-cost platforms powered by open-source LLMs (i.e., open-source, freely available), such as LLaMA (Meta AI), Falcon (TII), and Mistral (Mistral AI). In this context, online courses and platforms will expand AI education, democratizing essential skills for the future," he listed.
Regarding the labor market, Paulo pointed out that AI could automate many administrative, legal, and financial functions, requiring greater adaptation of the workforce. "It's not about replacing professionals with AI, but about replacing those who do not use AI with those who make good use of it. If AI is not yet helping you with something at your job, stop and reflect: there is very likely something it can make a significant difference in. New demands will arise for professionals who know how to integrate, manage, and regulate AI. For example, until recently, it was hard to imagine job postings for prompt engineers. In 2025, this demand is expected to grow, along with the emergence of new roles," he endorsed.
For Paulo, in 2025, it's not just about speculating what will come, but about preparing ourselves for a world where AI will be increasingly present and indispensable. "The real question is not 'if' AI will be integrated, but 'how' and 'by whom'. And I believe it would be good to ask ourselves if we are ready to harness the transformative potential of this technology," he emphasized.
Responsible AI: Regulation, Ethics and Sustainability
According to the researcher, in 2015, AI regulation will be more prevalent, including in Brazil. "Proper regulations will provide greater transparency, safety, and ethical use of technology. On another front of the issue, responsible AI initiatives have focused efforts to minimize algorithmic discrimination. This involves practices to address distortions caused by training data, as well as improving data diversity," he said.
Paulo also informed that regarding sustainability, AI will be used to optimize supply chains, reduce waste, and improve energy efficiency, especially with the popularization of AI agents in corporate systems. "Furthermore, predictive tools will help monitor climate changes and plan mitigation actions," he revealed.
In the entertainment industry, according to him, AI could generate scripts, music, visual arts, and even complete movies. This should be intensified with the emergence of startups that offer services, including free ones, to create these types of content. "Another point relates to interactive games and simulations, which will be conducted in real time based on user choices. Although the hype around the metaverse has diminished, these technologies continue to evolve, especially in the gaming sector," he detailed.
Paulo also highlighted that AI will be essential for managing homes, cities, and even infrastructure autonomously. Today, as he mentioned, it is inconceivable to think of smart devices without generative AI, so what is expected is a greater connection between AI and the Interconnected Smart Devices (IoT).
Advances and concerns
“Artificial General Intelligence (AGI) differs from AI, including generative AI, in that it is not limited to a specific domain. It is a form of intelligence that can adapt to new situations, solve complex problems, and learn continuously. While LLMs like ChatGPT demonstrate impressive abilities, they are still limited to specific tasks based on the training they received and lack the ability to understand broad contexts or transfer knowledge between different areas in a genuine way. Sam Altman, CEO of OpenAI, recently reported that the challenges for General AI are of an engineering nature, that is, a lot of work, and do not require basically new scientific advances. He added that his company is at stage 2 of 5, and even reported that this can be overcome by 2025”, explained Paulo.
According to the researcher, despite numerous benefits to people, companies, and society in general that can be provided by the conveniences this technology can generate, there is much discussion about the dangers involved, especially regarding ethical, control, and security aspects. "How to ensure that AGI remains aligned with human interests? Who will control this technology? If not properly addressed, it could further widen the gap between nations and social classes, and systems that utilize it could even cause catastrophic impacts if misused. Whether with general AI or even generative AI, I believe that semi-autonomous systems, where the final decision necessarily passes through a human being, are the safest option until these dilemmas are definitively resolved," he concluded.