Original post → https://goo.gle/2RopzLu
This week, Jon Foust and Michelle Casbon bring you another fascinating interview from our time at Next! Michelle and special guest Amanda were able to catch up with Paco Nathan of Derwen AI to talk about his experience at Next and learn what Derwen is doing to advance AI.
Paco and Derwen have been working extensively on ways developer relations can be enhanced by machine learning. Along with O’Reilly Media, Derwen just completed three surveys, called ABC (AI, Big Data, and Cloud), to look at the adoption of AI and the cloud around the world. The particular interest in these studies is a comparison between countries who have been using AI, Big Data, and Cloud for years and countries who are just beginning to get involved. One of the most interesting things they learned is how much budget companies are allocating to machine learning projects. They also noticed that more and more large enterprises are moving, at least partially, to the cloud.
One of the challenges Paco noticed was the difference between machine learning projects in testing versus how they act once they go live. Here, developers come across bias, ethical, and safety issues. Good data governance polices can help minimize these problems. Developing good data governance policies is complex, especially with security issues, but it’s an important conversation to have. In the process of computing the survey data, Paco discovered many big companies spend a lot of time with this issue and even employ checklists of requirements before projects can be made live.
In his research, Paco also discovered that about 54% of companies are non-starters. Usually, their problems stem from tech debt and issues with company personnel who do not recognize the need for machine learning. The companies working toward integrating machine learning tend to have issues finding good staff. Berkeley is working to solve this problem by requiring data science classes of all students. But as Paco says, data science is a team sport that works well with a team of people from different disciplines. Paco is an advocate of mentoring, to help the next generation of data scientists learn and grow, and of unbundling corporate decision making to help advance AI.
Amanda, Michelle, and Paco wrap up their discussion with a look toward how to change ML biases. People tend to blame ML for bias outcomes, but models are subject to data we feed in. Humans have to make decisions to work around that by looking at things from a different perspective and taking steps to avoid as much bias as we can. ML and humans can work together to find these biases and help remove them.
For more GCP Podcasts → https://goo.gle/30x1aYn
Subscribe to the Google Cloud Platform channel → https://goo.gle/GCP
Publisher: Google Cloud
You can watch this video also at the source.