University of Jyväskylä receives over 800,000 in funding to develop more reliable interactive AI
University of Jyväskylä Faculty of Information Technology has received 822,000 euros in Research to Business funding from Business Finland. Funding was granted to the Large Cognitive Models (LaCoMo) project to develop more reliable interactive artificial intelligence. The project’s total budget is 1,028,727 euros.
Research to Business funding helps public research organizations commercialize research results and accelerate the development of new business opportunities.
The project leader, Associate Professor Jussi Jokinen, explains that the project is developing large-scale interactive cognitive AI systems that are more reliable, understandable, and useful in real-world applications in business environments.
“For many organizations, the key question is not whether an AI system can perform a task once, but whether it can perform reliably over time. It is also important that the models are able to handle error situations and improve after their mistakes”, Jokinen explains.
In the project, AI systems are expected to not only operate over the long term but also to reliably follow instructions, collaborate with humans, and adapt to changing situations.
The researchers use computational cognitive science in the AI models to combine memory, planning, action, and learning into a single system. The technology’s effectiveness is being evaluated in realistic conditions together with industry partners.
The project aims to produce a pilot-ready software platform for organizations that are developing or deploying interactive AI systems.
Asutosh Hota, a doctoral researcher involved in the project’s commercialization and business development, says that potential application areas include, for example, simulation and digital twins, robotics, defense and security training, critical infrastructure, operational coordination, and support for human–AI decision-making.
“For example, in a cybersecurity setting, an AI agent could not only help investigate unusual network activity, but it should also identify relevant threat patterns, recommend response actions, and keep the case updated as new evidence emerges”, Hota explains.
This is where LaCoMo differs from many current AI agents that are built mainly around large language models. A typical language-model-based agent can produce useful summaries, explanations, and recommendations, but it often relies on the current prompt, available context, and external tools. LaCoMo’s approach is to give the agent an explicit cognitive structure around the language model. The system keeps track of what it has observed, what actions it has taken, what constraints it must follow, and how previous decisions turned out.
In the cybersecurity example, this means that the system would not only answer a question such as “what is happening in this network log?” Instead, it would maintain an evolving case history, compare new evidence with earlier observations, select possible next steps under organizational rules, record why a recommendation was made, and support an after-action review if the response failed or succeeded. This makes the AI’s operation more traceable: the organization can inspect the final answer as well as the the reasoning path, constraints, memory, and updates that led to it.
According to Hota, this would make investigations faster and more cost-effective. The key is that AI does not only produce answers, but can also remember, explain, keep track, and improve over time.
Hota reminds that this is just a single example and that various use cases will be explored together with interested partner companies and organizations.
“The plan is to work together with partners to identify concrete use cases that can be used to define measurable success criteria for AI and, at the same time, to prepare pilot packages for evaluating the technology under realistic conditions”, Hota says and emphasizes that industry collaboration is a central part of the project.
It is essential for organizations to understand why AI behaves in a certain way and whether it is able to follow relevant constraints. Researchers stress that the technology’s behavior must be understandable and analyzable for it to be used reliably as part of business operations.
Companies and organizations interested in collaboration are invited to contact the LaCoMo project team at the Faculty of Information Technology, University of Jyväskylä.
