Human-AI Expert Teams Set to Transform R&D with Next-Generation Machine Learning at ELLIS Institute Finland
AI in R&D often fails outside its training data, but combining human expertise with machine learning boosts adaptability and innovation. This hybrid approach enhances the design-build-test-learn cycle.

A New Approach for R&D: Integrating Human Expertise with Machine Learning
Artificial intelligence (AI) has advanced significantly, but its practical use in research and development (R&D) still faces key challenges. Current machine learning (ML) methods often struggle when applied outside their original training data, limiting their reliability in real-world scenarios. Addressing this gap is the focus of a European Research Council (ERC) funded project led by Professor Samuel Kaski, founding director of the ELLIS Institute Finland.
Why Current Machine Learning Falls Short
Machine learning models typically rely on data sets that represent the environment where they will be applied. However, in many cases, especially in scientific research and product development, conditions change unexpectedly. This leads to “out-of-distribution” failures where AI tools can no longer make accurate predictions or decisions.
Professor Kaski explains that this limitation arises because new challenges often require stepping beyond existing data. Collecting large amounts of new data to retrain models can be expensive or impractical, particularly in innovative research areas.
The Missing Link: Human Expertise in the Loop
One promising solution is to actively involve domain experts as part of the learning process. By combining human knowledge with AI models, the system can adapt to new situations more effectively without extensive new data collection. This approach integrates experts into the design-build-test-learn (DBTL) cycle, a fundamental process in R&D.
- Design: Creating new hypotheses or products
- Build: Developing prototypes or experiments
- Test: Running experiments or trials
- Learn: Analyzing results to inform the next cycle
Embedding AI tools within this cycle, with expert guidance, can improve decision-making and accelerate innovation. This “active re-learning” allows AI to handle unexpected variables while keeping human experts in control.
AI-Assisted Virtual Laboratories
The project envisions virtual labs where scientific processes are partly automated, yet remain collaborative. AI assistance would help scientists design experiments, interpret data, and explore new ideas faster and more efficiently.
This does not aim to replace scientists but to enhance their capabilities. Automating routine or computational tasks frees experts to focus on creativity and problem-solving, making R&D more productive.
Building AI That Understands Human Goals
For AI to be a genuine partner in research, it must grasp the often tacit objectives of human experts. This requires developing machine learning methods capable of understanding human reasoning and adapting to shifting goals within a team.
Such AI would function as a cooperative and adaptive teammate, capable of reasoning about the behavior and feedback of multiple experts. This would foster interdisciplinary collaboration, essential for tackling complex societal challenges.
ELLIS Institute Finland: A Hub for Collaborative AI Research
The ERC project is part of a broader effort at the ELLIS Institute Finland, a network uniting 13 Finnish universities. The institute focuses on advancing machine learning research and its application in various domains, including R&D.
ELLIS Institute Finland is actively recruiting principal investigators and fostering partnerships between academia and industry. This environment aims to accelerate the translation of fundamental AI research into practical tools that support expert teams.
Looking Ahead
The integration of human expertise with adaptive machine learning models promises to reshape how research and development processes unfold. By enhancing the design-build-test-learn loop with AI assistance, teams can tackle complex problems with greater efficiency and insight.
This approach offers a path to more dependable AI tools that perform well even when faced with new, unforeseen challenges—bringing us closer to truly collaborative human-AI expert teams.
For those interested in expanding their AI skills relevant to R&D and product development, exploring Complete AI Training’s latest courses can provide valuable knowledge and practical tools.