Artificial intelligence is a tremendous asset to companies that use predictive modeling and have automated tasks, but AI still has a ways to go to completely eliminate bias. Just like humans can show bias, so can AI, because it’s trained on human-generated data. Leading insurance tech provider Zelros knows this all too well and is leading the way when it comes to responsible and ethical use of AI.
If we consider an AI system where the human interaction is at the center, and a system open for self-criticism and continuous improvement, by design, we would then have a proper basis for an ethical AI system, fit for responsible use.
When human data is involved in teaching AI systems, unintentional bias happens. One of the most noticeable is gender bias, occurring when the AI system does not behave the same way for a man or a woman (husband or wife), even when the only variable presented to the system is gender. This can lead to individuals who should be in the same risk category being offered unequal policy advice.
Making an AI system goes much deeper than having it just being self-sustainable. It needs to be open for self-criticism and ongoing improvement through the design itself. Analyzing different metrics at various subgroup levels within a specific model feature is one key to identifying and keeping biases from occurring before going to market. With any AI, having it avoid falling for Simpson paradoxes through these methods ensures that the personal data isn’t used against the client/consumer it is actually supposed to be benefiting.
Companies should recognize that biases may be hidden in unexpected places. There can be a bias in the training data (hence in the associated trained model), or in the data later at inference/prediction time — or even both. One of the things Zelros encourages is companies use open-source strategy to be more open and transparent in their AI initiatives. Zelros does this, as do other companies, including Google and IBM.
Companies that want to use AI responsibly can collaborate with organizations like Isahit, which is dedicated to helping other organizations become competent in their use and implementation of responsible and ethical AI.
Christophe Bourguignat is founder of Zelros, a technology company transforming the way insurance companies serve their clients and the way consumers search for insurance and find policies in general by offering the first-ever AI software dedicated to insurance distribution.
