Providing Trust and Security in Artificial Intelligence
Trustworthy AI can be seen as a combination of (i) fairness, (ii) robustness, (iii) privacy protection, (iv) security, (v) accountability, and (vi) transparency and is a cornerstone for building reliable systems and creating user acceptance. We therefore apply our in-depth knowledge in the field of IT security to the design of secure and trustworthy AI systems.
Trust is one of the major issues when it comes to IT systems that are deployed in critical environments and it is one of the basic building blocks of IT security. In the context of artificial intelligence, trust is a key concern when developing systems that work with sensitive (personal) information or providing decision-making support to human decision-makers, thus resulting in the notion of trustworthy artificial intelligence. In other words, providing trust in such systems is also vital for gaining the acceptance of users and data subjects.
Trustworthy AI is often described as a combination of the following key attributes of a data-driven system:
- Privacy Protection
- Safety & Security
Furthermore, there is a lot of implicit trust invested in the data sources used for training models, which means that they need to be taken into consideration as well, including protection of said source data by means of data protection through hardened systems and fingerprinting technologies, as well as protection of user privacy through advanced anonymisation and aggregation technologies.
In order to build trust in data-driven systems, we also incorporate blockchains in order to construct highly distributed systems that can fulfil the above-mentioned criteria, even in the presence of hidden adversarial entities (both internal and external). Further work extends to the realm of explainability as well as transparency with respect to source data stewardship. Last but not least, we incorporate the wide range of knowledge we have gathered in the area of IT security, ranging from low-level technical fields like digital forensics to matters of organisational security and security management, in order to provide secure data-driven systems.