As I write this article, I am thinking that it might be simpler to write 'What Not to Expect from AI in 2025' than 'What to Expect'! How do you select 5 areas for a technology that is making progress in almost every field at the same time? To narrow it down and hopefully make it more interesting, I decided to focus on trends that are not directly related to the developments of ChatGPT or its competitors. It is safe to say that these trends will grow, and the applications and the companies behind them will strive to make these applications solutions to all possible problems.
Here are 5 AI trends for 2025 (in no particular order):
Trend One: Agents Everywhere
You may have heard the term Agentic AI. While AI has always been related to learning patterns, it has evolved through the following stages: (a) learning patterns from data (b) generating new content based on these patterns, and (c) taking action based on these patterns. When these three come together, you have an AI agent—a software that can learn, create actions, and execute those actions. More developments in this field are expected by 2025.
Trend Two: Transformation of the Education System
There has been much discussion about whether AI will encourage cheating, replace teachers, or fundamentally affect how students learn. While all of these are crucial, another equally or even more important force is emerging. This year, there is increasing evidence that due to AI-driven skills and economic conditions, recent graduates are struggling to find jobs. This not only raises questions about how students learn but also what they learn. The economic pressure from a downturn in the job market will force graduates, and ultimately the institutions that educate them, to confront new demands from businesses for employees. Students need to adapt first, improving their skills through various means, and institutions need to keep up. I expect we will begin to see these changes in 2025.
Trend Three: Applications of AI in Science
In the 2024 Nobel Prize in Science, two awards are related to AI. This should serve as a wake-up call: the application of AI in science will continue. It is also worth noting that while global attention and imagination are focused on generative AI, billions of dollars are pouring into the scientific applications of AI, with new announcements every day, ranging from space exploration to medical advancements. Similarly noteworthy is that despite so much investment and progress, data indicates that the success rate of AI-discovered drugs in phase two clinical trials is similar to that of other drugs, but it's important to note that some of these drugs were already 'known' in some form. As of the writing of this article, I have not seen any news of AI-generated drugs receiving FDA approval. What does this combination indicate? It tells us that the potential is huge, but it has not yet been fully realized.
Trend Four: Data Mining
Skeptics have long predicted that AI would exhaust data, while others have countered this. Among these predictions, what seems consistent is not the existence of data but the increasing difficulty of obtaining high-quality and ethically sourced data. I expect this will be a trend in 2025. Untapped data, especially about our physical environment, remains very large. However, large language models have already scoured most of the readily available data. By 2025, it is expected that more efforts will be made to acquire data, whether through commercial contracts, organizing untapped data via tagging systems, or deploying more sensors, among other means. Coupled with the aforementioned trends of AI in science, we can imagine that efforts to mine scientific data will accelerate.
Trend Five: Robotics
Artificial intelligence has made progress in all areas where problems can be solved by software (such as email, content creation, MRI analysis, etc.). In all these areas, AI has driven cost savings and job disruption. Robotics has brought AI into the physical realm—whether in manufacturing, surgery, agriculture, or space exploration. The applications of AI combined with physical automation are virtually endless. By 2025, we expect to see existing trends in this field continue to expand and gain broader public attention.
In summary
Over the past year, large language models and generative AI have developed rapidly and seem capable of solving any fundamental task. In 2015, we expect to see the next wave arrive, with a deeper impact on specific fields and institutions and a fusion with other technological trends becoming the focus.