AI, NLG, and Machine Learning, Chatbots, Thought Leadership

How to Create Trustworthy AI

As society has entered a race to develop more artificial intelligence implementations in various sectors like education, healthcare, and more, many people have questioned the downsides of technology such as the ability to mimic human cognitive functions, showcase emotions and empathy, and potential mistakes and errors in implementation.

How To Create Trustworthy AI

Discover.Bot partner Adi Gaskell, Writer and Consultant on Technology and the Future of Work, shared his insights on how to create trustworthy AI, touching on AI bias and ethical AI.

Adi Gaskell - Amazon on Biteable.

Find the transcript of the recording below to follow along with the video!

As society has entered a race to develop more artificial intelligence implementations in various sectors like education, healthcare, and more, many people have questioned the downsides of technology such as the ability to mimic human cognitive functions, showcase emotions and empathy, and potential mistakes and errors in implementation.

Ethics of AI is part of a much broader ethics of technology conversation that's often specific to robots and artificial intelligent entities. These conversations are usually divided into two main categories.

These systems have real implications for business and people, and they're vulnerable to the biases and errors of their human makers. For instance, Deloitte's Beena Ammanath outlines four key forms of AI bias in her recent book Trustworthy AI.

Key Forms of AI Bias

1. Selection Bias: First of these is selection bias, which is due to imperfect data that over or underrepresents certain groups.

2. Confirmation Bias: The second main form of bias is confirmation bias, which is when we seek to develop AI systems that conform to the outcome that we want.

3. Explicit and Implicit Bias: The third kind is explicit and implicit bias, which describes the biases we know we have, but also the biases we don't know we have.

4. Institutional Bias: The final main form of bias is institutional bias, which are the biases that are so deeply rooted in society that we may not even know they exist.

When thinking of ethical AI, it’s almost impossible to overlook the different scenarios in which these systems could go wrong. Overall, research has considered our pressing issues in the ethics of AI and the concerns that arise, as well as organizations that have pledged a commitment to addressing these concerns.

1. Job displacement and wealth inequality:. Automation, or labor-saving technology, is technology in which processes and procedures are performed with minimal human assistance or interference. Automation is widely regarded as the future of work.

2. Imperfect errors and mistakes: Machine learning takes time to be useful. If trained well and fed good data, AI can perform well. Conversely, if fed bad data or made with internal programming errors, machines can be harmful.

3. Bias in algorithms: As AI becomes increasingly inherent in facial and voice recognition systems, these systems have real implications for businesses and people, and they’re vulnerable to the biases and errors of their human makers.

4. Singularity beyond our control: Singularity is one of the most interesting issues in the ethics of AI to consider. It refers to a hypothetical point at which technological growth becomes uncontrollable and irreversible, resulting in unintended consequences.