This week, BuzzFeed News, citing sources familiar with the matter, wrote that Facebook is developing an AI tool that summarizes news articles so that users don’t have to read them. The tool — codenamed “TLDR” in reference to the acronym “too long, didn’t read” — reportedly reduces articles to bullet points and provides narration, as well as a virtual assistant to answer questions.Read More
The European Union’s Fundamental Rights Agency (FRA) has issued a report on AI which delves into the ethical considerations which must be made about the technology. FRA’s report is titled Getting The Future Right and opens with some of the ways AI is already making lives better—such as helping with cancer diagnosis, and even predicting…Read More
Transparency, explainability, and trust are pressing topics in AI/ML today. While much has been written about why they are important and what you need to do, no tools have existed until now.Read More
Smartphone-based mobility data has played a major role in responses to the pandemic. Describing the movement of millions of people, location information from Google, Apple, and others has been used to analyze the effectiveness of social distancing polices and probe how different sectors of the economy have been affected.Read More
As machine learning (ML) becomes more effective and widespread it is becoming more prevalent in systems with real-life impact, from loan recommendations to job application decisions. With the growing usage comes the risk of bias – biased training data could lead to biased ML algorithms, which in turn could perpetuate discrimination and bias in society.Read More
Unconscious biases are pervasive in text and media. For example, female characters in stories are often portrayed as passive and powerless while men are portrayed as proactive and powerful.Read More
Zest AI, a company developing AI-powered loan decisioning products, today closed a $15 million funding round led by Insight Partners. A spokesperson says the capital will be used to accelerate Zest’s go-to-market efforts and product R&D.
About 1 out of every 9 loan applications (10.8%) for home buying — and more than 1 in 4 applications (26.4%) for refinancing — were denied in 2017, according to a nationwide analysis of lender data conducted by the U.S. Bureau of Consumer Financial Protection. Minorities were disproportionately rejected, with the overall denial rate for mortgage applications from Black Americans reaching 18.4% in 2018. (Hispanic and Asian applicants were rejected 13.5% and 10.6% of the time, respectively, compared with 8.8% for non-Hispanic white applicants.)
Zest, which was cofounded in 2009 by former Google CIO Douglas Merrill and ex-Sears VP Shawn Budde, claims its mission is to create “more rigorous” standards around debiasing algorithmic lending. To this end, the company helps banks, credit unions, and specialty lenders identify borrowers by taking into account more than credit scores. Zest claims institutions that lend using its models — including Discover, Akbank, and VyStar — have seen a 20% increase in approval rates on average and an up to 50% reduction in chargeoffs, or declarations that an amount of debt is unlikely to be collected.
Zest provides over 30 customers with resources to prep, build, iterate, and document machine-learning decisioning models for cards, auto loans, personal loans, mortgages, and student loans. Complementary tools help teams evaluate and validate the models for safety, stability, business impact, and compliance. Customers can deploy and monitor algorithms in production, or they can engage Zest’s team of service and machine learning experts for assistance with development and validation steps.
Zest claims to use a technique called adversarial debiasing to minimize potential model prejudice. The technique pits two machine learning models against each other, with one attempting to predict creditworthiness while the other second-guesses the race, gender, and other attributes of the applicant scored by the first model. Competition drives both to improve their methods until the predictor can no longer distinguish the race or gender outputs of the first model, resulting in a model that is ostensibly more accurate and fair.
Zest recently introduced ZAML Fair, which the company claims can reduce bias in loan portfolios with “little or no” impact on profitability. ZAML Fair leverages the transparency tools built into Zest’s solutions suite to rank a system’s variables by how much they lead to biased outcomes. It then attempts to mitigate the influence of those signals to produce a superior model.
Based on the mortgage lenders who tested ZAML Fair, Zest says the tool would eliminate 70% of the nation’s gap in approval rates between Hispanic and white mortgage applicants and cut the even larger gap between Black and white borrowers by more than 40%. In a blog post, Zest cited a survey conducted by the Harris Poll that found a majority of Americans would give up more personal data if it resulted in a fairer credit decision. With that in mind, Zest believes it can reduce bias by using “better math and more data to assess borrowers.”
Of course, it’s difficult — if not impossible — to completely rid algorithms of bias. Facial recognition models fail to recognize Black, Middle Eastern, and Latinx people more often than those with lighter skin. AI researchers from MIT, Intel, and Canadian AI initiative CIFAR have found high levels of bias from some of the most popular pretrained models. And algorithms developed by Facebook have proven to be 50% more likely to disable the accounts of Black users compared with white users.
But Zest claims the data proves its efforts are making a difference. Using Zest’s underwriting software platform, one lender says it was able to shrink the disparity in approval rates between white applicants and applicants of color by 30% on average, with no increase in portfolio risk. Separately, an auto lender was able to approve “thousands” more borrowers.
“The COVID-19 shock led many financial institutions to update and improve their systems for resilience and durability, which caused a significant increase in demand for our business,” CEO Mike de Vere told VentureBeat via email. “A big part of that included building new and improved underwriting systems with the latest math and software technology. This resulted in Zest’s best Q2 on record, with an eye on finishing the year with triple-digit growth.”
Los Angeles-based Zest has raised over $87 million in venture capital to date.
Sign up for Funding Weekly to start your week with VB’s top funding stories.
LinkedIn today released the LinkedIn Fairness Toolkit (LiFT), an open source software library designed to enable the measurement of fairness in AI and machine learning workflows.Read More
Researchers at the Montreal AI Ethics Institute and Microsoft propose using machine learning to build comprehensive archives that could bridge gaps in cultural understanding, knowledge, and views. They assert that including more voices in archival processes — with the help of machine learning — can have positive effects on communities, particularly those archivists have historically marginalized.Read More
Twitter researchers claim to have found evidence of demographic bias in named entity recognition. They say their analysis reveals AI performs better at identifying names from specific groups, and the biases manifest in syntax, semantics, and how word uses vary across linguistic contexts.Read More
Why you need more women in data science. Meet the initiative which helps companies set a better habit to generally consider women for data-intense roles.Read More
In the past few weeks, there has been much debate around existing biases in our day to day lives and how we should tackle them, promoting the idea of a bias-free space. The AI community wasn’t left untouched, and someone tried the existing language models for their developed biases.Read More