Artificial Intelligence: Underpinnings of a Disruptive Wave – San Francisco

Artificial Intelligence: Underpinnings of a Disruptive Wave – San Francisco

As you know, Chetan Sharma Consulting has been taking a deeper look at Artificial Intelligence in our Mobile Breakfast Series. This year our AI caravan stopped at Seattle, Vancouver and San Francisco. Each panel brought out different viewpoints but there are some common themes that started to emerge. I will talk about some of the common elements in a separate post.

Our SF event was in partnership with Mobile Monday Silicon Valley for logistics/outreach and Google Launchpad who generously provided the space and food. Thanks.

The panelists were:

 

Dr. Tsvi Achler, CEO, Optimizing Mind

David Gerster, VP – Data Science, BigML

Diomedes Kastanis, VP of Innovation, Ericsson

Barak Turovsky, Head of Product Google Translate and MI, Google

Chetan Sharma, CEO and Founder, Chetan Sharma Consulting (moderator)

 

Tsvi Achler has a unique background focusing on the neural mechanisms of recognition from a multidisciplinary perspective. He has done extensive work in theory and simulations, human cognitive experiments, animal neurophysiology experiments, and clinical training. He has an applied engineering background, has received bachelor degrees from UC Berkeley in Electrical Engineering, Computer Science and advanced degrees from University of Illinois at Urbana-Champaign in Neuroscience (PhD), Medicine (MD) and worked as a postdoc in Computer Science, and at Los Alamos National Labs, and IBM Research. He now heads his own startup Optimizing Mind whose goal is to provide the next generation of machine learning algorithms.

David Gerster is Vice President of Data Science at BigML, where he promotes the idea that data science is easy by speaking at conferences and teaching workshops. Since joining BigML in July 2013, he has spoken at CERN, Big Data Spain, Papis.io, DataLead (UC Berkeley), DataBeat (VentureBeat), and more than a dozen other venues. At Groupon, he built an elite data science team that trained the first machine-learned models for mobile deal relevance. At Yahoo, he led the project to collect billions of URL clickstreams in Hadoop and use them to improve Yahoo’s main web search algorithm. He holds an MBA from the University of California at Berkeley and a Bachelor’s degree from Harvard University.

Diomedes Kastanis leads Ericsson’s long-term technology vision and innovation across media, OSS, BSS and m-commerce. He is also responsible for university collaboration, industry thought leadership and advanced technology incubations. Previously, Diomedes held the roles of researcher in the Institute of Computer Science of Crete, head of enterprise architecture BSS/OSS for Middle East and Africa and most recently, global head of product line and content delivery. Diomedes holds an advisory role in multiple forums, including the technology innovation forum for Greece. He has played a crucial role in the evolution of Ericsson’s media delivery and identity efforts with a vision towards a networked society. Diomedes holds several patents and has been published in a range of scientific publications. He earned an MSc. in Computer Science in Electrical Engineering and an MSc in Mathematics.

Barak Turovsky is responsible for product management and user experience for Google Translate. Barak focuses on applying advanced machine learning techniques to deliver magical experiences and to break language barriers across web, mobile applications, Search, Chrome and other products. Previously, Barak spent 2 years as a product leader within Google Wallet team. Prior to joining Google in 2011, Barak was Director of Product in Microsoft’s Mobile Advertising, Head of Mobile Commerce at PayPal and Chief Technical Officer in an Israeli start up. He lived more than 10 years in 3 different countries (Russia, Israel and the US) and fluently speaks three languages. Barak earned a Bachelor’s of Laws degree from Tel Aviv University, Israel, and a Master’s of Business Administration from the University of California, Berkeley.

As you can see, it was a pretty deep panel with excellent and diverse backgrounds. They also have hands-on experience in shaping the AI/ML/DL world. As you would expect from SF, we had a very engaged audience who were precise in their questions and opinions.

We have talked about AI in the context of the “Connected Intelligence” technology Wave. It is a key building block of the industry architecture.

The salient points discussed during the panel were:

  • There is a lot of hype around AI. Anyone with a few data points in an excel spreadsheet is running around claiming that are doing AI. One has to just drive down 101 in the bay area as the billboards have morphed into AI ads.
  • Despite the hype and noise, real work is getting done in AI. One of the biggest projects to date is Google Translate which translates 140 billion words on a daily basis. AI performs well when there is a lot of good data and a well-defined problem. Some of the interesting areas that have shown promise are by training the algorithms for English to Chinese and English to Japanese and the AI engine automatically figures out the Chinese to Japanese translation.
  • The two key areas where AI is doing well: operations and performance.
  • The best use case for AI today is unstructured data and plateaued statistical data.
  • The biggest gating factor for good AI is compute power. Combine that with the data needs and there are few dominant players who have the resources for computer power and trove of consumer data. For startups, enterprise and specific domains might be the best way to get started.
  • Securing AI IP is going to be challenging, an evolving area for sure.
  • How do we protect AI from going berserk? What safe guards can we place? This was a common theme in a number of audience questions and the answer lies in how much power does the human relinquish to the AI and automation. When stakes are high like in health and transportation, one has to be careful before full confidence is gained for giving up some of the responsibilities.
  • The big issue around AI/ML talent shortage has be dealt with education in the long-term but with tools in the short-term so that one doesn’t have to be a PhD to get better at AI
  • The wireless industry is planning to use AI in a number of ways, firstly, in the network by making networks more predictable, open, and adaptable. And this getting baked into a lot of the 5G work that is going on. Networks should be able to anticipate problems and fix them without even human interventions but again if there is some critical change that needs to happen that has the potential to bring down the network, better get a human involved in the process. These safeguards are an absolute must for safe practice of AI.
  • There are three key dimensions of AI – Autonomy, Reasoning, and Learning.
  • Domain expertise is a key competitive differentiator. Define the problem well, just don’t throw algorithms at the problem.
  • Human brains are great at big pictures but suck at small details. AI is great at details and lack the big picture. Perhaps, we can marry the two worlds.
  • The power of AI is in transfer learning and knowledge accumulation.
  • Startups should focus on specific domains, computing power, and network operations.
  • A must needed area of research and development is “explainability” of AI. Currently, AI is like a magic black box which does its thing and produces outcomes but there is a lack of toolsets that can help explain the decisions and the decision tree. It is not only essential for research but also public policy, for explaining things when they go right or wrong, to help consumers get more comfortable with the technology.
  • EU’s General Data Protection Regulation or GDPR is looking at this fundamental issue of explainability in AI and that AI will have to explain every action and decision. This is clearly a fertile ground for exploration.
  • There is a lot of emphasis on Zero shot learning – the ability to solve problems or tasks without having to train on the new dataset. The ability of AI to adjust when data is updated or is missing is crucial. This also allows it to handle some the data bias issues.
  • The biggest mistake a lot of startups or teams that are just starting with AI make is not realizing the cost function for the project – properly defining the scope of the problem and figuring out the data costs including the costs related to labeling and supervised learning.
  • Taking AI from supervised learning to unsupervised learning is a big area for research and further exploration.
  • Privacy is a big concern in the age of AI. With all the data being produced by individuals, the opportunity for abuse is enormous. How will the industry rise to be the stewards on consumer privacy rather than just the opportunists to exploit the vulnerable.
  • Obviously, the holy grail of AI is mimicking human brain but we are still far away from reaching that goal.
  • Eventually, AI will become the underlying layer of the Connected Intelligence framework as we envisioned in our Connected Intelligence series of papers.

You can watch the entire session, presentation, and the panel at https://www.youtube.com/watch?v=nmA7KuatfhM&feature=youtu.be

My thanks to all panelists and the attendees who made it lively. I had high hopes for audience to engage in dialogue and they didn’t disappoint. There were great questions and the discussion following the panel was rewarding as well. I know we could have gone till mid-night and still not exhausted the set of questions on AI. We know it is a deep topic with broad implications so have devoted quite a bit of our time and energy to explore the topic. We will continue the discussion with much in-depth discussion at our upcoming summit on Sept 7th.

Our annual Mobile Future Forward will tackle many of these themes in much detail. Hope to see you then.

 

Thanks

Chetan