Jeremy Barnes
From a Career in ML to AI for Good

15 Apr 2018



"Real value is in creating new knowledge. And knowlege is ... the basis of all competitive advantage."

Listen to this episode on iTunes/Apple Podcasts

Listen to this episode on Android/Google Play

Download .mp3 (85.3M)

Player by Podigee

Show Notes

Topics: Ad Tech, Alpha Go, Amazon Web Services, Applied Mathematics, Churn Modeling, Kubernetes, Ethics of AI, Google, Hydrology, Machine Learning, Managing Engineers, Natural Language Processing, Neural Networks, Open Source, Product Development, Prof. Robert Williamson, Real-Time Bidding (RTB), Social Responsibility, Teamwork and Team Leadership, TensorFlow, Trust, UX (User Experience) Design

Companies

Projects: RTBkit

Introduction

As a boy, Jeremy Barnes learned from his hydrologist father who was collecting data from small devices in remote terrain that small amounts of data can support useful models and make lives better. These themes shaped his career, as he spent the first two decades studying machine learning and creating systems that created value by creating new knowledge. After founding and selling two startups, and learning the power of open source software to “put ideas into the public record” and influence an industry with RTBkit, he is embarking on a new challenge. As Chief Architect at ElementAI, he is part of applied research and product design efforts shaping a future of AI that accounts for human users, promotes transparent AI decision making and human trust in AI systems, and is explicitly pursuing AI for Good, starting with a foundation by that name. Jeremy looks back on what he has learned as an entrepreneur, engineer, data scientist, team member and leader, and looks ahead to his own future and the future we are all building.

Guest Bio

Jeremy Barnes is a Montreal-based engineer, researcher and entrepreneur in the domain of artificial intelligence. Since arriving from Australia in 2001, he has founded and served as CTO and CEO of several companies in AI, including Datacratic (to iPerceptions) and mldb.ai (to Element AI). He’s currently the Chief Architect at Element AI, where he is responsible for building out the core artificial intelligence platform and ensuring that the research, product, customer projects and other activities all fit together to enable a broad, practical and humane application of recent AI techniques. Element AI advances cutting-edge AI research and turn it into scalable solutions that make businesses safer, stronger, and more agile.

Notes

  • 05:30 - AI for Good Lab - London
  • 06:00 - His role is to investigate AI as a driver for economic behavior which drive impact on society, especially for corporate actions
  • 06:30 - AI needs to solve problems which are good for society
  • 07:00 - Trying to give users a strong voice in AI in products
  • 09:00-10:00 - AI systems can discover/predict things humans haven’t found before, surprising, but won’t be trusted unless decisions are explained and system build trust. Alpha Go example.
  • Element AI encourages companies to do both of these things - humans in UX process and design products with explicable decisioning that can earn trust. Even great product if untrusted will fail
  • 10:00 - Applied research group. Consult with strong companies. Create products, structure and design there but no data. Customer data completes the product for that customer.
  • 12:00 - Inspired by hydrologist father collecting data in the field
  • 13:30 - “Data is not something that comes to you. It’s actually something that you need to go and get.”
  • 14:00 - Learned you can create good models from small amounts of data
  • 15:00 - Diabetic, realized he was learning from observation, from data, without new effort
  • 16:00 - Idilia - NL as a service
  • 16:30 - Design data collection mechanism, tepresent problems in data, and let algorithms solve problems
  • 17:00 - Learned a huge amount at Idilia and realized it would have a huge impact
  • 18:30 - Can small amounts of data still solve some problems? Or is this era where big companies and their data dominate?
  • 20:00 - AI misconception that data is fixed - “either you have data or you don’t”
  • 21:00 - It’s actually a continuum between modeling and the amount of data, you can apply models from large data sets to smaller problems or do more involved modeling and still get good results from smaller data sets
  • 21:30 - Churn modeling - zero-sum game, it’s actually about competing better. Better solution is actually to deliver a better product, rather than use AI to perpetually steal customers. Use AI to design a better product.
  • 23:00 - AI is still just another tool for product development, not a replacement
  • 24:15 - “People will do almost anything to try magic bullet solutions, just to see if they work. … That doesn’t mean they are actually solving a real problem. It just means they tried enough things that something worked.”
  • 25:00 - Real value is creating new knowledge. Neural networks don’t create new knowledge. “Knowledge … is the basis of all competitive advantage.”
  • 26:15 - RTBkit
  • 27:00 - Startups that are solving new problems, inventing new IP
  • 27:30 - Market timing very hard to get right. Idillia 7-10 years too early.
  • 28:00 - “Missionary work” - explaining that your product idea is the future. Fed his ego but recognized as an anti-pattern
  • 29:15 - Tensorflow - open source but really an extension of a corporate entity - Google
  • 29:30 - Open source is a way to put an idea into the public without needing perfect market timing. Put the idea into the public record.
  • 30:00 - RTBkit maybe a year early. Changed perception in the industry that you can and should own your own bidding process. Project not vibrant but had an impact on the market
  • 31:00 - Kubernetes as Google’s masterful attack on AWS by making workloads portable in and out
  • 32:00 - Google has become very good at using open source as a weapon and to create a moat
  • 34:00 - Kubernetes at a right level of abstraction. It reduces lock-in and actually enforces competition among the very large companies.
  • 35:00 - But big companies using open source this way hurts all the startups in a market.
  • 36:30 - What have you learned about yourself as a leader?
  • 37:00 - Started out career thinking I was best at every aspect of the work. Learned humility from those around me.
  • 38:30 - Recognizing I can have the most impact by supporting everyone else and connecting their work
  • 39:30 - How do you create the conditions for a team to thrive?
  • 40:00 - In teams that work well you need to build up trust, safety and communication. You need to make sure the team is able to take risks.
  • 41:30 - “I do struggle to have a career plan for myself … I love to learn. I love what I do because I get to learn and have fantastic people around me.”
  • 42:00 - ElementAI a chance to work with larger companies. Larger companies have entrepreneurial initiatives also.
  • 43:00 - Learn how to deliver AI innovation at scale

Links

Credits

  • Host: Mark S. Weiss
  • Intro and Outro Music: “Florida Song” Copyright 2016 by Photographs (Mark S. Weiss). All rights reserved.
  • Ad Intro Music: “Woozy” Copyright 2015 by Photographs (Mark S. Weiss). All rights reserved.
  • All Interview content Copyright 2018 by Using Reflection and Mark S. Weiss. All rights reserved.