To create Real AI for Real People in the real world, FCAI's research HAS THREE OBJECTIVES:
DATA EFFICIENCY – TRUST & ETHICS –
DATA EFFICIENCY – TRUST & ETHICS – UNDERSTANDABILITY
The world needs new kinds of real artificial intelligence.
The issues artificial intelligence still faces across fields in science, technology and society—Data Efficiency, Trust & Ethics, and Understandability—are also the keys for creating the new wave of the coming AI revolution.
To overcome these three bottleneck problems, FCAI’s research brings together a wide spectrum of scientific disciplines, business domains and societal initiatives. This cross-discplinary work—AI Across Fields— together with our impact actions will yield both our concrete solutions for applying AI and scientific advances in fundamental AI development.
I DATA efficiency
Current AI solutions can be very successful in domains where tasks are relatively simple and well-defined and an abundance of high-quality, properly annotated data are available. Existing AI methods do not, however, easily extend to domains where such data are not available or are difficult or expensive to acquire. Real AIs will be able to work with real-world scarce data – ill-defined, hard to acquire or unavailable
II Trust and ethics
We will create AIs that are secure, give trustworthy results, preserve privacy, are fair, and whose use is ethically sustainable. We will develop the required privacy-preserving and secure methods to address challenges related to susceptibility to manipulation, information stealing and unethical approaches. We will provide new resilient deep learning approaches for the currently popular and successful deep neural networks.
AI does not yet understand users. We need to open the “black box” of many AI methods: to understand how methods such as neural networks operate and what are the uncertainties inherent to their outputs. Modeling the user and the interaction will help the AI understand the user and vice versa. The outcome is AIs that are able to augment human capabilities in a multitude of tasks.
FCAI runs currently five research programs with multiple research groups contributing to each program. The programs do fundamental AI research and reach across a variety of scientific disciplines; groups from multiple fields work together. New openings are constantly sought after and new programs launched on a yearly basis.
Under construction: all dark purple cells have on-going reseach, which will be linked to soon.
AI research programs (columns) and the disciplines linked to them (AI Across Fields rows), with expansion plan in colors: currently in operation (dark purple), starting in 1–2 years (middle purple), and in planning (light purple). Examples of on-going work and initiatives are being linked to from the matrix (work in progress).
In order to maximize the positive impact of AI, we work together with top experts in an increasing number of scientific disciplines, business domains, and societal initiatives. The resulting new insights from each we leverage to inspire new AI methods.
Up to now, FCAI has initiatives in a variety of fields, such as materials physics, social sciences, and economics. The list will be updated constantly, and new ideas and disciplines are welcome to join our common endeavor.
Collaborations between the research programs and accompanying disciplines across fields will yield concrete solutions. They highlight how scientific advances will have direct and tangible societal impact and contribute to economy and well-being.
FCAI’s Research programs
Agile probabilistic AI
Contributes to objectives I, II and III by developing an interactive and AI-assisted process for building new AI models with practical probabilistic programming. The models will work as explainable, verifiable, uncertainty-aware, reliable tools to build and check the behavior of AIs.
Responsible coordinator: Professor Aki Vehtari, Computer Science (Aalto University)
Contributes to objectives I and III with new methods needed for real AIs to have efficient and interpretable reasoning capabilities. This requires cross-breeding modern machine learning and simulator-based inference.
Responsible coordinator: Professor Jukka Corander, Statistics (University of Helsinki, University of Oslo and Sanger Institute)
Next-generation data-efficient deep learning
Contributes to objectives I and III by developing methods that harness the power of deep learning. These methods include semi-supervised learning, few-shot learning for making use of auxiliary sources of training data, and learning models that can be reliably used in simulator-based inference. The goal is to achieve high-quality results with scarce training data and only limited human supervision.
Responsible coordinator: CEO Harri Valpola (Curious AI)
Privacy-preserving and secure AI
Contributes to objectives I and II with security and privacy research. We develop realistic adversary models to build effective tools and techniques that practitioners can use to build dependable AI systems.
Contributes to objectives II and III by developing methods for collaborative forms of AI: their ability to infer human beliefs and abilities from observations and predicting the consequences of their actions on humans. These are AIs with which people can naturally work and solve problems, and which demonstrate the ability to better understand our goals and abilities, take initiative more sensitively, align their objectives with us, and support us.
Responsible coordinator: Professor Antti Oulasvirta, HCI, Cognitive Science (Aalto University)
FCAI research programs are fixed-term and new programs will be launched yearly primed by active scouting by the researchers in the community, and by proactive planning to include missing fields by partnering, networking, and recruiting. Currently, we are preparing to initiate a new research program on Autonomous AI.
The next set of new programs being prepared for consideration involve Ethical and Social Aspects of AI, Multimedia and Computer Vision, and Industrial AI. We also launch annual open calls for the renewal process.
Artificial Intelligence is expected to become the technology with the biggest business and societal impact over the near future. Education, government, social welfare and work and employment will be drastically impacted by the transformation it brings.
The Finnish government has launched a national AI Strategy and Program that outlines a set of key actions. By successfully applying AI, Finland can reach a €20-billion and 8% annual boost in GDP by the year 2023. With focused AI-based activities in business, net employment can increase by up to 5%.
A basic requirement is to set up an international AI hub, which attracts key players around the world and supports Finland in becoming a leading country in the application of AI.
FCAI will be that hub.
FCAI enables industrial renewal and promotes the transformation of society in multiple ways, in accordance with four impact dimensions: scientific, economic, societal, and educational.
FCAI will play a key role in Finland’s AI Strategy to realize the potential for AI-led economic growth by accelerating industrial and societal renewal through effective and ethically responsible application of AI. Our research results bring on the renewal of existing businesses and generate new AI-driven ones. We will also put results into practical use in the public sector.
FCAI forms an ecosystem that enables strong international networking of academia and the private and public sectors. We create new possibilities for collaboration and synergy, technology transfer, recruitment, and knowledge exchange.
FCAI aims for three kinds of scientific results:
1) The next generation of AI solutions, which are both able to learn from data and able to plan and simulate, will open doors for developing AIs with new kinds of capabilities.
2) Each of FCAI’s research programs are solving fundamental technical problems—enabling new kinds of solutions in machine learning and modelling.
3) As AI is badly needed in a large number of other scientific fields, but data efficiency, trust and understandability are bottlenecks for current AI systems to become widely adopted.
A major hurdle for the adoption of AI in industry and society is the lack of experts at all levels. Continuous and life-long education were identified as a major challenge in a recent government report on societal effects of AI. Successful implementation of AI requires that the general public consider the new applications trustworthy and ethical.
The objective of this program is to increase the competence of both the professional workforce and the general public.
FCAI is structured to facilitate the wide dissemination of research results in society—ranging from industries to the public sector to individual citizens—across diverse fields of application. FCAI is actively participating in strategy work, policy discussions and in advisory bodies to facilitate ethically sustainable integration of AI into society.
To promote societally conscious application of AI, FCAI has also founded the FCAI Society, a multi-displinary working group to consider the socio-ethical and cultural impact AI will bring.