Principles of Responsible AI
Human-centred AI
Human-centred AI refers to AI systems designed to enhance human capabilities, not displace them. It is inspired by the depth and diversity of humanity and motivated to have an ethical impact on people and society. Human-centred AI often occurs at the intersection of AI technology and neuroscience, blending the study of machine intelligence with the study of our own brains. By using AI to make models of human neural networks, we can enhance our understanding of our brains and behaviour and also use this knowledge to improve AI.
The ultimate goal of human-centred AI is to handle projects through a partnership, with humans taking on the roles of goal setting, strategising, high-level creativity, and governance. AI will augment these human abilities through creativity and low-level detail work while providing a key opportunity to design at scale.
Humans must be kept in the loop, and there are clear demarkations around what should be automated (and with what controls) as opposed to what requires human decision-making.
(Stanford HAI, 2020)
Implications for AI development
To ensure the responsible use of AI, the following must be considered:
It's predicted that AI will increasingly impact our future decision-making processes. To create responsible AI systems, they need to be designed and built by diverse people with different perspectives so they can 'learn' all facets of humanity. This will limit the possibility of bias in AI systems and allow them to use their knowledge to help all humans.
It's also important to ensure that the data collected that will be used to influence decisions are unbiased and true. People developing AI should consider checking to ensure data is from a reliable and relevant source, has not been tampered with, and is consistent and valid. Data management platforms can help developers to design and explore the sources, structure, and reliability of data.
As AI requires data to perform actions, developers must consider how this may affect data security. Data security can be optimised by limiting access to various datasets to those who need it, keeping track of employees who access, use, and edit data, and using appropriate data management tools, among other strategies.
AI will affect the workforce in some way. Consider how to use AI with human input and oversight, which may make some jobs redundant, but may also create new jobs requiring new skills.
(IBM, 2022)