Sherman Lee on AI – Interview

Next up is Sherman Lee from, Rocco.AI, and

You can read more about Sherman Lee here:

Let us begin:

1. Can you tell us a little about your background and why you have an interest in AI?

I never thought that I would get into the AI field. My story is a wild one as my parents moved from Hong Kong, to Queens, NY, to Oakland, California. The liquor store below is where I spent most of my childhood. We would go there to buy drinks, food, ice cream, snacks, and alcohol every single day. But so would all the drug dealers in the neighborhood. The area was so rough. When the police came around, the drug dealers hid their stash of crack rocks in my mailbox (without my consent). They came knocking on MY door when things cooled down to retrieve their supply.

All my life I hustled hard and got into UC Berkeley where I took my first graduate course in distributed computing.

Then I got one of my first jobs at Yahoo doing machine learning on their content platform. One of my biggest projects was clustering of news articles. So if you ever searched for a news story and saw all articles around a topic, or if you ever scrolled down all the way to the bottom of a single article and viewed a section with “see related articles”, you probably saw some of my work.

T​his was an incredibly hard problem to solve. I worked with a team of 25 research scientists. People with PHDs in machine learning. My job was to figure out how to scale those learning algorithms and get it to run on 40,000 servers via a Hadoop Cluster.

2. You are involved in the AI space now, can you tell us a little about that?

Well, this was also a wild ride. While at Yahoo I was reading WAY too much Techcrunch. I wanted to be a founder too and thought it was going to be easy. I started my first company Good Audience in London and received funding from Techstars. It’s a monthly subscription based service for social media done by humans. I learned a great deal about content at Yahoo and wanted to help the world with content distribution via social media. Good Audience has worked with amazing brands like, Kettle Brand Chips, the San Francisco Federal Reserve Bank, and many more you’ve probably heard of.

My co-founder Karl and I doing a content strategy session with one of our early customers in London. Warner Yard circa 2014.

My intellectual curiosity never stopped. There was one repeated task from our customers that kept coming up. It was a request to create and post content on their social media. This was a hugely creative task, but it was not rocket science. You just needed a solid understanding of the target audience, tone of voice, branding guidelines, marketing guidelines, and you could go and do that.

A year ago, I started another company called Rocco — an AI-powered social media marketer. It’s a Slackbot that helps you curate content on Twitter and Facebook. Over 300 customers are using it and I learned a very big lesson. You do need to be a rocket scientist 🙂 People loved what we were doing and we had some of the biggest brains in AI backing us.

Speaking to a crowd at WeWork who were dying to use Rocco.

However, nobody told me that unique content generation by machine was a near impossible task. The latest advancements in research, the computing power, and large datasets would all be needed to see my vision come true. So while our customers are finding the best links and articles to post on their social media channels through Rocco today, they are waiting patiently for the day where Rocco will actually write unique content with human level intelligence.

Let me give you an example. Someone who guys by @Samim trained a neural network on ~2,000 TED talks with ~4 million words. You would think this could generate content around technology, right? Here’s what the output looks like:

To see my vision for Rocco come to life, I needed to solve a much more foundational problem in AI. How am i going to compete against the mega-corporations without the computing power? As you know, training neural networks is expensive. I called my friend Rahul Vishwakarma and asked him how to solve this.

Mate Labs founder Rahul Vishwakarma

He showed me a prototype of something and my jaw dropped. Rahul sent me a URL that performed a unit of deep learning calculation when I loaded it up on my browser. I asked if that meant if we sent this to all of our friends we can train neural networks for free and can we work together on this?!? He said yes and yes.

That’s how the Raven Protocol was born! Raven is a decentralized & distributed deep-learning training protocol. We are providing cost-efficient and faster training of deep neural networks for the world to use. This is being built as a community effort with the biggest and brightest minds in AI. Please reach out if you’re interested! Btw, if you visit the site quickly you’ll get 100 free Raven tokens to use in our platform.

3. Do you see any risks in the future when it comes to AI?

Killer robots is always a risk. Anything autonomous with no kill-switch could potentially be dangerous.

4. In the AI space, who do you admire the most and if you could ask them 1 question what would it be?

Jaan Tallinn, how do you save humanity? I know you’ve been thinking about it 🙂

5. What is your favourite TV or film with some form of AI in the storyline and why do you like it?

Her. I almost broke into tears when Samantha told Theodore she was talking to thousands of people simultaneously and has fallen in love with hundreds of them.

6. I like to sneak this one in. If a self driving car had to choose one of two separate people to run over, who should it choose and why?

Dangerous question. In a world where AI is accepted as the norm, pedestrians should fill out some sort of disclaimer about how ready they are to be the chosen one.

7. I know it’s ever evolving but, what is the most exciting AI project you have come across, academic or real world and why?

Google’s Federated Learning. Can you believe it does on-device learning on your mobile phone through the GMail app?

8. What is the funniest AI story you can think of?

While using one of those AI scheduling assistants, the recipient got upset at ME for the AI failing. Whoops!

9. What ethical issues can you see from the increased use of sex robots?

People falling in love with these robots while in another relationship.

## Ends

Thanks for your time Sherman, some great insight in there.

If you would like to get involved, either asking questions or being asked them get in touch.

Richard Young

Richard has been interested in the AI space for some years. Questions of ethics can raise some serious problems. What if AI can learn who is pre-disposed to cancer and then not give them health insurance?