Is Artificial Intelligence Bias?
This is an article I had drafted a while ago but IBM’s recent announcement to abandon their facial recognition product reminded me to post this.
Diversity or lack of, is well reported and debated in the technology field, with women making up only 11% of the developer workforce globally[1] there’s no doubt that as an industry we have a problem. When we cast the net wider to ethnicity too the picture is mixed, generally the industry is better at attracting White and Asian but poor for African and Hispanic ethnicities[2] but the picture is complicated and varied across the world.
There’s plenty of evidence that shows that a diverse workforce create more profitable and successful businesses[3]. So a lack of diversity is bad for business but more importantly and worse it’s bad for society. As more and more applications are making decisions for us and about us they are increasingly affecting critical areas of our society from banking and policing to defense and policy making.
Although no one is intentionally writing bias into code it is being increasingly apparent that ‘algorythmic bias’[4] is coming from a workforce that is predominantly white and male.
The examples of bias are numerous and well documented. MIT looked at Amazon’s Rekognition technology which mistook women for men 19 percent of the time and mistook darker-skinned women for men 31 percent of the time[5]. Similar results are found across facial recognition technology from IBM, Microsoft and others.
Software developers largely blame biased training data which is probably true but more concerningly these technologies are being released into the world without their developers realizing there is even a problem, would that happen if the development team contained more women or more ethnic minorities? Wouldn’t these issues have been spotted before we start to see them being used ‘in the wild’. These issues may not seem too serious, it’s not like they are mainstream yet but even with these known error rates the Greater London Police are about to deploy facial recognition technology on the streets of London as a matter of course[6].
The move towards AI and Machine Learning is the foundation upon which much of a our digital and technology future is being built. It promises an unbiased and fair future however it has to learn and it’s learning from us, bias humans. At best this is unconscious but in some instances we see this come through as open discrimination such as the Microsoft’s Tay Chatbot, triumphantly released on Twitter in 2016 it allowed Tweeters to engage and ‘chat’ but within 24 hours people had turned Tay racist and sexist[7].
But it’s not all bad news. The fintech sector has long been aware of bias and with AI and ML driving this sector forward they are especially aware of the bias that can amplified within their systems[8]. AI’s are increasingly being deployed to assess credit worthiness, assess access to loans and fraud detection. AI has the potential to offer unbiased assessment in an area that has traditionally been unfair in its treatment of applicants[9]. The sector is so aware of their responsibilities in fact that they take a transparent and refreshingly open approach in their communications, for example Barclays publish their stance on AI bias openly[10].
Fintech offers us a model that other sectors should take note of. Taking an active and transparent stance will help us navigate the road ahead together. Being conscious about the issues we face is the first thing we can do. Actively recruiting diverse team to work on new innovation is vital, bring more people in to widen our horizons. And if you want to become an agent of change then take a look at the Algorithmic Justice League[11].
[1] https://www.pearsonfrank.com/blog/female-developers-workforce/
[2] https://www.techrepublic.com/article/5-eye-opening-statistics-about-minorities-in-tech/
[3] https://www.forbes.com/sites/annapowers/2018/06/27/a-study-finds-that-diverse-companies-produce-19-more-revenue/#4b1e2cda506f
[4] https://www.theguardian.com/technology/2019/jun/29/ruha-benjamin-we-cant-wait-silicon-valley-become-more-diverse-prejudice-algorithms-data-new-jim-code
[5] https://www.theverge.com/2019/1/25/18197137/amazon-rekognition-facial-recognition-bias-race-gender
[6] https://www.met.police.uk/live-facial-recognition-trial/
[7] https://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist
[8] https://www.bankingdive.com/news/artificial-intelligence-lending-bias-model-regulation-liability/561085/
[9] https://www.zillow.com/research/black-white-mortgage-denials-19616/
[10] https://home.barclays/content/dam/home-barclays/documents/citizenship/our-reporting-and-policy-positions/policy-positions/20190614-CDEI-CP-Bias-in-Algorithmic-Decision-making-Barclays-Response-FINAL.pdf
[11] https://www.ajlunited.org/take-action#REQUEST
annaxmalina
blakekathryn
stormborns
origamiaround
jbbartram-illu
2087