Scale Transform: AI Leaders Showcase Research Breakthroughs and Real-World Impact

If you were not able to join Transform live, the keynote and breakout sessions are now available on our website for on-demand viewing.

Artificial Intelligence is now a part of our everyday lives. This technology is no longer theoretical, but is making a direct impact on businesses and their bottom lines. Nearly every industry has woken up to the potential of AI to transform their operations, whether it’s leading the charge of an entirely new category, like autonomous transportation, or modernizing legacy industries, like finance, real estate, and retail. AI is poised to be as transformative as the internet, and with that, organizations are learning that data is the new code. If they’re not investing in this critical technology now, they’re at risk of being left behind.

Today we hosted our inaugural conference, Scale Transform, to showcase the state of AI today, from the latest research breakthroughs to the real-world impact across industries. As I shared in my opening remarks, “AI continues to surprise us with its new capabilities, primarily enabled through advancements in the core resources of data, compute, and algorithms.”

We brought together more than 9,000+ leading researchers, practitioners, and executives to hear from the people and companies leading our industry forward. It was a day full of thoughtful conversations and presentations that examined the strides made in advancing these core resources, the challenges that still lie ahead, and what we can expect in the years to come.

The Latest Advancements in Research

AI research has been crucial in advancing the industry and our understanding of what’s possible. We’ve made significant progress across countless areas that have propelled the industry forward, yet there is still plenty more to be done. Some of the latest research discussed includes:

• Artificial General Intelligence (AGI) - Sam Altman, Co-Founder and CEO of OpenAI, touched on breakthrough research such as GPT-3, CLIP, and DALL·E, bringing us closer to achieving AGl. He said, “We may not even know when AGI actually emerges - it won’t be a single moment in time. The Turing Test for achieving AGI will be when AI is able to complete a percentage of human labor really well.”
• Machine Learning Frameworks - The ecosystem of ML frameworks is evolving quickly. Google Software Engineer and Keras Creator Francois Chollet discussed how the roadmap for Keras in the next five years includes developing an ecosystem of reusable models such as KerasNLP and Keras Applications, increasing automation, enabling larger-scale cloud-based workflows, and supporting real-world deployment. Separately, Soumith Chintala, the Creator of PyTorch, shared what he considers the three personas in ML Development: The Modeler, Prod, and Compiler and discussed how these personas interacted with each other pre and post-deep learning. He closed his session by highlighting the need for PyTorch and other mainstream frameworks to maintain a high velocity to stay relevant and keep up with the latest innovations to enable Modelers.

Soumith Chintala, Creator of PyTorch speaking on the future of ML frameworks.

• Autonomous Vehicles - Drago Anguelov, Distinguished Scientist and Head of Research at Waymo discussed the current state of perception, highlighting that, “Perception works very well for the vast majority of cases, but there are still some interesting long-tail problems like rare events that still need to be addressed.” Despite the remaining challenges, Anguelov discussed how the self-driving industry is beginning to “move up the stack” to tackle behavior prediction and planning, including the recent release of the updated Waymo Open Dataset to support further research into behavior prediction and motion forecasting. Anguelov also discussed how Waymo has successfully deployed L4 level autonomous vehicles in Chandler, AZ.

This research and advancements made by others in the industry, including our conference partners, Stanford Institute for Human-Centered AI (HAI), PyTorch, DeepLearning.AI, OpenAI, Pachama, and ELC, are the catalysts for meaningful use cases of AI.

Challenges Ahead

One of the biggest challenges for our industry is scaling AI systems from proofs of concept to production. A common theme heard from speakers throughout the day is that data is the critical foundation that companies must invest in to unlock AI’s full potential: Andrew Ng of DeepLearning.AI discussed how models can improve with more compute, data and parameters, however, it's not just about how much data you have, but the quality of that data. Andrew brought up a critical point here, because developing AI with bad data means it won’t live up to the potential and quite frankly it’s dangerous. It can lead to:

• Safety risks (e.g. self-driving cars not accurately identifying pedestrians or obstacles)
• Biased algorithms (e.g. facial recognition bias)
• Inaccurate results / recommendations (e.g. AI-driven healthcare decisions)

Andrew Ng, speaking on the need for good data to develop AI.

Bart Nabbe, VP of Strategic Partnerships and Corporate Development at Aurora, reinforced this point, stating, “You’re not building ladders to get to the moon, you need rockets.” And data is the fuel for those rockets. He emphasized that in the early days of training ML, every piece of data counts, however over time it’s increasingly important to be precise and focus on the long-tail, differentiated data to improve models.

The importance of creating ethical and unbiased AI has been a focus in the industry for years, and as the technology has gone mainstream, so has the conversation. Global leaders and technology experts across the world are tackling challenges in developing AI that is ethical, as well as ensuring the benefits of AI can be equally distributed. Our speakers provided passionate advice on the topic:

Kevin Scott, CTO of Microsoft at Scale Transform.

• Kevin Scott, Chief Technology Officer of Microsoft, enumerated some of the challenges to ensuring that the benefits of AI are equally distributed. “We have an enormous amount of work to do. […] We’re learning a lot about biases in datasets, how to manage potential harms, ensuring the development of AI is inclusive, and that AI is being used safely.” Scott then spoke about how his teams at Microsoft are developing the tools (e.g., PowerBI, autoML) to enable developers without machine learning expertise to develop ML applications.

Dr. Fei-Fei Li, speaking on the need for diversity in the field of AI.

• Diversity in the field of AI and greater interdisciplinary collaboration is key to minimizing the unintended consequences of AI, and Dr. Fei-Fei Li, Sequoia Professor of Computer Science, Stanford University, Denning Co-Director, Stanford Institute for Human-Centered AI (HAI) believes that as we continue to build AI, the industry needs to take inspiration from and embed human cognition, intention, and emotion in the technology for ML to achieve its full potential to augment human capabilities.

Real-World Impact

Perhaps most importantly, we also heard inspiring work that’s happening in the real world – tangible examples of how AI is already improving the way we live and operate:

• Facebook reaches billions of people across the world and often needs to deploy products quickly, at scale from translations to social recommendations to generating descriptions of images for the visually impaired. On an average day, Facebook is processing 6 billion transactions a day. Head of Applied Research at Facebook AI, Srinivas Narayanan, relayed the importance of “getting the most bang for your buck” for your labeled data by augmenting it with various techniques, and creating AI responsibly by viewing fairness as a process and not a binary notion in order to implement the best AI across a variety of use cases.
• The financial industry has complex workflows filled with massive amounts of paperwork that often includes tedious manual processing. During a panel, Brex’s Co-Founder and Co-CEO Henrique Dubugras discussed how applying AI technology like document understanding has helped the financial world increase efficiencies and lower costs while freeing up humans to focus on more fulfilling work.

What’s Next

The continued advancements in AI have created incredible applications in today’s world but there is an exciting future that awaits - from the potential to create safer transportation to agricultural benefits like increasing crop yields and improving healthcare outcomes. At Scale, we’re excited to have a hand in creating trusted, ubiquitous use of AI and we can’t wait to see what’s next.

A special thank you to our conference partners: Stanford Institute for Human-Centered AI (HAI), PyTorch, DeepLearning.AI, OpenAI, Pachama, and ELC. We are also grateful for all our incredible speakers representing organizations including Aurora, Blend, Brex, Cruise, DoorDash, Etsy, Facebook AI, Fast.AI, Flexport, Google, Hugging Face, iRobot, Microsoft, Qualtrics, RedFin, University of Toronto, and Waymo.

If you were not able to join us live, the keynote and breakout sessions are now available on our website for on-demand viewing.