Do we already have the solution?
Fortunately, many forward-thinking technology companies were already addressing the trust question before generative AI revealed the full implications and risks of AI. Discussions and research have been ongoing for several years, and some of the world’s brightest minds have been working on solutions to make trustable AI a reality.
Fujitsu has built tools to detect and mitigate bias in AI models and launched a private, trustworthy sandbox environment to leverage the power of conversational generative AI. Both AI Trust and Generative AI are key areas under Fujitsu Kozuchi, which consists of 7 areas of AI that can be rapidly developed, tested and implemented to deliver immediate results.
A key advantage of Fujitsu Kozuchi Generative AI is that organizations and startups can design and develop new generative AI solutions that meet their specific needs without worrying about resources: the risk of GPU scarcity is mitigated, as the necessary infrastructure exists, and no-code natural language inputs and the availability of consultancy from expert advisors minimize skills issues.
The real breakthrough, however, comes with the ability to achieve a new level of trust, created with multiple technology layers presented as a holistic solution:
• A new Fujitsu technology detects hallucinations in conversational AI models – a phenomenon in which generative AI creates incorrect or unrelated output.
• Fujitsu AI Ethics for Fairness allows training datasets and the fairness of AI decisions to be simply verified on a web browser. It includes functionalities to visually assess bias mitigation techniques to select the best one.
• Fujitsu Intersectional Fairness, integrated into Fujitsu AI Ethics for Fairness, specifically addresses discrimination towards groups of people with various combinations of sensitive characteristics (e.g., gender, race, age, sexuality, religion, disability, etc.) that are not detected and mitigated by traditional bias mitigation strategies.
• Fujitsu AI Ethics Impact Assessment provides a digitized solution to assess AI systems’ ethical impact and risks based on international AI ethics guidelines and previous AI incidents. Fujitsu AI Ethics Impact Assessment can be used to future-proof AI systems to comply with ethics principles and legal requirements. It can also create evidence to engage with auditors, approvers, and stakeholders.
• A recent solution looks beyond model development to provide security technologies that prevent possible generative AI attacks through disinformation and “LLM poisoning”.
