AI Talk: Bias in AI and 2019 innovations

December 6th, 2019 / By V. “Juggy” Jagannathan, PhD

This week’s AI Talk…

Bias in AI – Three perspectives

The New York Times published an article on bias in AI as part of their Women and Leadership section. They interviewed three prominent scientists to get their views on this topic.

The first was Professor Daphne Koller of Stanford University, one of the co-founders of Coursera and founder and CEO of Inisitro, a company developing new drugs using machine learning. Her example of bias: If you do a search for an image of a CEO, you will get pretty much all white males. Machine learning is simply reliant on the data you feed it and can latch on to any quirk in the data. It’s extremely important she opines, to test all models in a realistic setting to determine if the model is free of bias.

The second woman interviewed was Professor Olga Russakovsky from Princeton University. She is currently working to reduce bias in the ImageNet dataset, which practically launched the revolution of machine learning. She argues that bias stems from three factors: Dataset is biased (need to do better sampling of the data); the algorithm can amplify whatever bias exists in the dataset (need to tune the algorithms better); and, lastly, humans are biased! Human bias cannot really be eliminated, but can be mitigated by focusing on diversity.

The third interviewee was Timnit Gebru, a research scientist at Google focused on ethical AI and co-founder of Black in AI, an organization promoting people of color in the AI field. She successfully managed to relocate a machine learning related conference to Ethiopia after half the black speakers were refused visa to attend the conference in Canada in 2018. Talk about bias! She echoes Professor Russakovsky’s position that the problem of bias is not just technological, but also social. We need to address the human, social and cultural factors which are more often than not the root cause of bias in the end product one sees from AI systems.

2019 Innovations

Time recently published a big list: 100 best innovations of 2019! The article itself is symptomatic of the times we live in, one of hyper acceleration of investments and innovations And AI is sprinkled throughout the list. A few innovations that caught my eye, some of which can be good gift ideas for your Christmas list this holiday season:

  • TytoHome: A tool that helps a doctor examine you (or your kids) remotely: Examine vitals, lungs, ears, skin, and throat and converse using video link. You can purchase this now from BestBuy for $300!
  • Osso VR: An educational tool for doctors to learn to perform surgery using a virtual reality headset.
  • Robyi Robot: An AI-powered educator for kids ages 3 to 7 which teaches STEM skills.
  • Stevie: This friendly, AI-powered robot converses with seniors, tells them stories and keeps them engaged.
  • Temi: Yet another robot, which looks like a vacuum cleaner with a face, rolls around your house avoiding obstacles and is generally intended to amuse you for an expensive price tag.
  • Denim unspun: This company addresses an age-old problem—how to get a pair of jeans that fits you and not some mythical, non-existent person! A 3-D scan of your body using the app they provide guarantees the denim you get in your mailbox fits you perfectly. I might even try that!
  • Omron HeartGuide: This wearable watch not only monitors your activities and sleep, but also monitors your blood pressure!
  • Scotch Flex: I will be remiss if I don’t call attention to an invention from my employer, 3M, that made this list! This one, a combination of tape and bubble wrap, is perfect for the holiday season to package anything you may want to ship as a gift!

I am always looking for feedback and if you would like me to cover a story, please let me know. “See something, say something”! Leave me a comment below or ask a question on my blogger profile page.

V. “Juggy” Jagannathan, PhD, is Director of Research for 3M M*Modal and is an AI Evangelist with four decades of experience in AI and Computer Science research.