&l;span style=&q;font-weight: 400&q;&g;E-commerce giant Amazon is finally stepping up efforts to audit its A.I. facial Rekognition tool that has been called out for built in bias. They plan to work with the National Institute of Standards and Technology (NIST) to develop standardized tests around this issue and improve accuracy. Will this move help stymie concerns over whether or not Amazon is exacerbating racial profiling and bias? &l;/span&g;
&l;b&g;The Breakdown You Need to Know&l;/b&g;
&l;span style=&q;font-weight: 400&q;&g;Amazon has been selling its facial Rekognition program to police departments across the U.S. for quite some time now. &l;/span&g;&l;a href=&q;https://www.culturebanx.com/&q; target=&q;_blank&q;&g;&l;span style=&q;font-weight: 400&q;&g;CultureBanx&l;/span&g;&l;/a&g;&l;span style=&q;font-weight: 400&q;&g; noted research shows commercial artificial intelligence systems tend to have higher error rates for women and black people. Some facial recognition systems would only confuse light-skin men 0.8% of the times and would have an error rate of 34.7% for dark-skin women.&l;/span&g;
&l;span style=&q;font-weight: 400&q;&g;This is &l;/span&g;&l;a href=&q;https://aws.amazon.com/blogs/machine-learning/thoughts-on-recent-research-paper-and-associated-article-on-amazon-rekognition/&q; target=&q;_blank&q;&g;&l;span style=&q;font-weight: 400&q;&g;Amazon&a;rsquo;s first attempt&l;/span&g;&l;/a&g;&l;span style=&q;font-weight: 400&q;&g; at partnering with a government backed lab to develop tests. Working with an organization like &l;/span&g;&l;a href=&q;https://nvlpubs.nist.gov/nistpubs/ir/2018/NIST.IR.8238.pdf&q; target=&q;_blank&q;&g;&l;span style=&q;font-weight: 400&q;&g;NIST&l;/span&g;&l;/a&g;&l;span style=&q;font-weight: 400&q;&g;, the industry leader for facial recognition perhaps finally shows they care about the damage the technology is capable of. Getting a central tool that evaluates how effective all facial recognition products are is vital to stamping out bias.&l;/span&g;
&a;ldquo;We welcome the opportunity to work with NIST on improving their tests against this API objectively, and to establish datasets and benchmarks with the broader academic community,&a;rdquo; Amazon wrote in a statement. NIST has put out three reports on facial recognition software starting back in &a;nbsp;2010, with a database of more than 26 million photos.
It will be interesting to see how Amazon plans to develop a one size fits all test, because its Rekognition can&a;rsquo;t be &a;lsquo;downloaded&a;rsquo; for testing outside of AWS. This is something that would need to happen in order to get independent results that accurately reflect how potential customers would use the service in the real world.
&l;b&g;A.I.&a;rsquo;s Big Business&l;/b&g;
&l;span style=&q;font-weight: 400&q;&g;While critics call for justification for the use of the technology and in some cases an all-out ban, artificial intelligence continues to be a billion dollar industry. A report published by Market Research Future found the global market for facial recognition is &l;/span&g;&l;a href=&q;https://globenewswire.com/news-release/2018/08/16/1553082/0/en/Facial-Recognition-Market-to-Touch-US-8-Bn-by-2022-at-19-68-CAGR-Industry-Leaders-Analysis-Safran-SA-NEC-Corporation-Gemalto-Cognitec-Systems-Aware-Inc-Daon-FaceFirst-Ayonix-Key-Le.html&q; target=&q;_blank&q;&g;&l;span style=&q;font-weight: 400&q;&g;set to surpass a valuation of $8 billion in 2022&l;/span&g;&l;/a&g;&l;span style=&q;font-weight: 400&q;&g; up from $3.04 billion in 2016. They expect the sector to expand at an impressive compound annual growth rate of 19.68%.&l;/span&g;
Other companies like tech giant IBM want to better train facial recognition systems in order to increase diversity and lower bias. They have released a new dataset, Diversity in Faces, with hopes it will help developers tackle gender and skin type biases.
A.I. needs to be directed on how to interpret and utilize data to establish a result. If the data being presented doesn&a;rsquo;t include enough information for a particular background (i.e. Black, Latinx), then the result isn&a;rsquo;t as reliable. Police departments across America still struggle to improve relationships with the black community. African Americans are already incarcerated 5 times more than other races. Black women are incarcerated at twice the rate of their counterparts.&l;/p&g;
No comments:
Post a Comment