AI-based solutions should behuman-centered. In a context of rapid technological change, it is essentialthat trust is the bedrock of societies, communities, economies, and sustainablegrowth. Trustworthiness is a prerequisite for people, organizations, andsocieties to develop, deploy, and use AI-based systems. Thus, to help itsstakeholders, and more generally, society realize the benefits of AI to addresssocietal challenges, the core agenda that ROBUST addresses is to develop abetter technical and socio-technical understanding of the principles underlyingtrustworthy AI and machine learning.
Robustness
If trustworthy AI is a prerequisite to ahuman-centered use of AI for sustainable growth, how can we ensure warrantedtrust? There has been a global effort to formulate sets of ethics principles toguide the development of AI-based systems. Trustworthy AI has three components:it should be (i) lawful, (ii) ethical, and (iii) robust, from a technical andsocio-technical perspective. The ROBUST program focuses on this thirdcomponent: robust AI.
Six dimensions
Trust can be gained in an intrinsic mannerby revealing the inner workings of an AI-based system, i.e., throughexplainability. Or it can be gained extrinsically by showing, in a principledor empirical manner, that a system upholds verifiable guarantees. ROBUSTpursues the following dimensions for which such guarantees should be obtained:
- Accuracy, including well-defined and explained contexts of usage;
- Reliability, including exhibiting parity with respect to sensitive attributes;
- Repeatable and reproducible results, including audit trails;
- Resilience to adversarial examples and distributional shifts; and
- Safety, including privacy-preserving machine learning.
In context
The ROBUST vision is to address thechallenge of trustworthy AI-based systems in a multi-disciplinary,multi-stakeholder set-up, with knowledge institutes, industrial partners,governmental organizations, and societal stakeholders, in healthcare,infrastructure, energy, and the service industry. ROBUST combines the creationof economic opportunities with targets for social progress. We articulate andoperationalize this ambition by aligning our research agenda on robust AI withthe United Nations’ sustainable development goals (SDGs).
