Notes on 2/28 Stanford Discussion on Machine Learning
Last night, I attended a pretty cool panel discussion at Stanford’s D-School on “Designing Machine Learning.”
Stanford offers a class on machine learning and design and it was really intriguing to listen to the juxtaposition of usability and this emerging technology. The panel included:
- David Gilmore, Co-Founder and CEO, Health Bricks
- Prabhas Pokharel, Co-Founder and CEO, Reduct
- Borui Wang, Co-Founder and CEO, Polarr
- Ryan Philips, Entrepreneur-in-Residence, SAP.iO
I thought I’d share the raw, unedited notes I took from this fascinating discussion. They’re not grouped into any particularly coherent organization but are simply presented in the order I took them. Perhaps a machine learning algorithm can organize them for me!
- The power of machine learning means input devices will be changing. The qwerty keyboard will increasingly become less critical as learning fills in for what had to be previously asked explicitly.
- Increasingly, the focus of machine learning is on interactive machine learning versus simply automating away laborious chores. The objective is to create interactive, assisted experiences to augment yourself and your effort. How can we go beyond anticipating need to anticipate output?
- With machine learning we need to focus on how can we guide people to do things they didn't think they should do? Can we move beyond anticipating needs to actually anticipating and accomplishing goals.
- To do this, our software has to learn how we learn.
- One big change coming is the phenomena of machine learning can take control of the User Experience process away from the designer. What does this mean for the design of future user experiences? How will we adapt user experiences for Machine Learning enabled environments?
- A really powerful new tool is Google "what if".
- As we progress in machine learning we have to in guiderails – sort of like designing bumpers for the bowling alley. We have to realize there are people who won't get bank loans because of our machine learning algorithm. How do we account for this?
- There is certainly a role for “design” in a machine learning environment. Design is being purposeful about making decisions. Sometimes it's about when to not employ machine learning.
- An interesting use for AI is to determine when AI has been employed. For example, can we determine how likely it is that a photo has been edited?
- The most widely experienced issues surround data. How do we collect data ethically? Where does our data source come from? There are a handful of big tech companies that have seamless access to data in a way that allows them to control their own destiny. What are the implications of this for them? How about for the rest of the companies?
- Machine learning needs to evolve from " we have a cool technology” to “what's a cool problem”?
- Our consensus is that we’re still 5-10 years from the glory years of machine learning.
- Machine learning start-ups should consider focusing on excluded groups? If you start with points of exclusion you'll find amazing things to work on.
- If you're a designer, find a machine learning engineer. If you're and engineer find a designer.
Contact us today to start a discussion.