by Dr. Kushani De Silva
15 Jan 2025
This blog aims to reflect the ethical use of AI with governance systems for making wise human decisions. Further, illustrates challenges, problems, and smart solutions for making life easier with AI and related emerging technologies.
The anthropologically fundamental fact reveals that humans are always able to do good and bad, life-supporting and life-destroying activities. However, values of dignity such as sustainability, solidarity, security, peace, participation, empowerment, freedom, community, justice, and love encourage humans to do good and life-supporting activities (Stückelberger, 2020; 2024). Therefore, it is important to inculcate such values within communities through support systems.
Among administrative, service-oriented, security, consumables, infrastructure, transport, health, and education support systems AI plays an essential role in a modern-day context. Thus, it is essential to set standards for AI governance values to make sure technology support is only to do good and life-supporting activities by humans. AI governance values can be illustrated as self-confidence, courage, compassion, generosity, gratitude, honesty, humanity, respect, modesty, and love (Stückelberger, 2020; 2024).
AI governance is linked to fact grounding aspect of AI models. Fact grounding is a challenging task of the large language modeling (LLM) model, which simply means the ability to generate factually accurate responses in information-seeking scenarios. Thus, AI governance values should essentially be a part of LLM. For example, factuality modeling (e.g. architecture, training, and inference) as well as measurement (e.g. evaluation methodology, data, and metrics) should be necessarily built on the foundation of AI governance values (Jacovi et al., 2024)
Fact grounding is the Model’s ability to generate factually accurate responses in information scenarios. It could be factuality based on input (e.g. user request, grounding document) and factuality based on external sources (e.g. general world knowledge) the main methods of fact grounding (Rashkin et al., 2023; Tang et al., 2024; Panetal, 2024). However, there are subtle cases that need clearer and better methods to analyze the combined methods of the above two fact grounding methods which are yet to be invented.
The highest grounding percentage is 83.6% associated with the Model gemini-2.0-flash-exp by Google whereas gpt-4.0 by Open AI is 78.8% (Google DeepMind, 2024). The important aspect we need to understand is why it is difficult to invent a model with a 100% grounding percentage. The answer is human mind’s decision-making capacity cannot be calibrated 100 % into AI models. Thus, it is wise to use AI as a support source of information and tool that needs to be governed for making decisions by humans without mistakenly replacing human decision-making skills as illustrated in Figure 1.
Fig 1: AI model integration with AI governance values for making wise decisions (De Silva, 2025)