AI Predictions for 2025 from the Associate Director of our AI Solutions Hub
What will the new year bring for AI?
Omer F. Alis, the associate director of the AI Solutions Hub at the Institute for Experiential AI, has many years of experience delivering AI solutions in banking, finance, telecommunications, healthcare, education, and biotech. Omer frequently talks with business leaders about their work developing and implementing AI strategies, giving him a unique look at the state of the technology in the market.
Below, Omer gives 12 predictions for where AI will thrive in 2025 — and where it might hit a wall. Explore what Northeastern AI leaders are keeping an eye on here.
- LLMs will approach saturation.
Large Language Models (LLMs) can be expected to plateau as they run dry of digitally available human-generated data. This will probably spur a shift in research toward innovative model architectures, synthetic data generation, and improved inductive biases to address limitations. As a result, Omer believes we’ll see lively debates about whether LLMs have reached their peak and whether future progress might require an entirely new paradigm.
- Autonomous agents will arrive...
Omer thinks AI agents, powered by LLMs, will dominate 2025. Building on theoretical advancements from 2024, these agents will integrate conversational AI with functionality-specific tools to automate human workflows. For example, call center applications will see LLMs engaging with customers to interpret caller intent, followed by automated actions like completing transactions with text-to-SQL functionalities
- …As will collaborative AI agents.
Omer thinks we’ll also see the proliferation of collaborative AI agents capable of working alongside humans. These agents can provide real-time assistance, project management, and complex decision-making for enterprise operations.
- Speech processing will leap ahead.
Omer expects a transformative leap in speech processing, much as we did for image processing in 2012 and natural language processing in 2017. A new kind of inductive bias may emerge to redefine what speech processing is. Even in its absence, multimodal AI—which combines speech, text, and image processing—will lead to groundbreaking applications for industry. Real-time transcription, sentiment analysis, and contextual understanding will make 2025 a pivotal year for speech processing.
- AI “reasoning” will take center stage.
With LLMs nearing their performance ceiling, researchers will emphasize reasoning and planning abilities to solve practical challenges. Reinforcement learning is one paradigm that will see a resurgence, fueling debates over whether current deep learning approaches are sufficient or if hybrid symbolic-deep learning models are needed.
- Small language models will go mainstream.
We’ll see companies move toward smaller, more domain-specific language models to address specialized tasks with proprietary datasets. These models will outperform generic LLMs for niche applications. Investments in data acquisition and curation will underpin this shift, marking 2025 as the "year of small language models."
- Work will transform as practical AI takes hold.
Recent theoretical advances in LLMs will trigger a race for practical applications. Businesses will develop LLM-powered chatbots, document processors, and question-answering systems tailored to specific industries, ranging from customer support to regulatory compliance.
- AI will be a shot in the arm for drug discovery.
AI in drug discovery will achieve significant breakthroughs, accelerating research and development timelines. The sector will see a surge in AI-driven startups, with many achieving unicorn status due to their ability to streamline drug design and validation processes.
- AI-generated video will carve out a highly competitive sector.
The race to develop advanced video generation technologies will intensify. Innovations will enable the creation of highly realistic, AI-generated video content, with applications spanning entertainment, education, and marketing.
- LLMs will need to deliver practical solutions for everyday users.
Public enthusiasm for LLMs will transition from unbridled excitement to measured expectations. Users will demand robust, practical applications that genuinely enhance daily life. The gap between current AI capabilities and science fiction portrayals, like the virtual assistant from the movie “Her,” will be a focal point of criticism.
- AI will continue to augment human work.
AI will augment human productivity by integrating into workflows. Applications will include real-time analysis of emails, documents, and meetings to provide actionable insights, assist decision-making, and improve career outcomes. Omer expects AI tools capable of annotating meetings, managing tasks proactively and augmenting humans in real-time.
- Meta programming languages will redefine software development.
Finally, Omer thinks we’ll see a paradigm shift in software development as "meta programming languages" emerge for specific domains. These languages will enable developers to interact with specialized LLMs, revolutionizing how large-scale software is conceptualized, written, and maintained.
Are you ready to turn your data challenges into opportunities? Partner with the AI Solutions Hub, where Northeastern University’s top AI experts like Omer deliver cutting-edge, responsible solutions tailored to your business needs. Schedule a strategy session today.
1200w.png)
