Computers are no more unbiased than people – they are designed by and operated by humans. With any automated system, we must look out for where human bias could creep in, risking unintentional unethical behaviour. This is particularly important with AI and machine learning, where computers often learn their behaviour from large amounts of potentially biased real-world data.
In procurement, that data could be suppliers’ historical responses to previous buyers’ requests, or suppliers’ past success rates in their sector. A machine learning system could then decide which suppliers to invite to buyers’ requests, based on past similar decisions. Our previous collaborations with the excellent team at the University of Exeter’s Institute for Data Science and Artificial Intelligence (IDSAI) developed machine learning tools to suggest lists of the best-matched suppliers for each buyer request.
However, this means there is an inherent diversity challenge: the machines learn from human precedent. If consciously or subconsciously buyers discriminate on sex, ethnicity or other factors then the machine will learn and copy those behaviours. An ideal, ‘fair’ process would ensure that similar suppliers had an equal chance of being matched by a machine learning tool, if their only perceived differences were sex or ethnicity based.
To reduce the likelihood of the best-matched supplier lists being discriminatory, we have once again partnered with the IDSAI, thanks to Innovate UK funding. By assessing the perceived gender and ethnicity of suppliers suggested by the AI, we can quantify how fair the best-matched supplier lists are. We can then utilise the latest research into countering discrimination in machine learning to account for any underlying biases from the historical data, and suggest more balanced lists of suppliers in future recommendations.
The question of how to make machine learning behave ethically is an emerging and important area of research in AI of which this will be an early implementation, applied to promote diversity within supply chains. Addressing this issue, as SupplyDevon aims to do, is vital for machine learning based systems to be appropriate for use in public procurement.
To find out more information about SupplyDevon or to pre-register as a Buyer or a Supplier visit, supplydevon.org.
