Inside AI - January 20th, 2019

Inside AI (Jan 20th, 2019)

New blank template
Subscribe | View in browser

Happy Sunday and welcome to the weekend edition of InsideAI!  I'm Rob May, CEO of Talla.  We recently launched a very cool "learning dashboard" for our customer support automation product that you should read about if you find AI UX interesting.  This week we have Q&A with the founder of, a "swarm AI" company.  Scroll on down to learn more.

But first, lets start with the most popular articles of the week from the daily version of InsideAI:

Intelligent surveillance is being installed in schools around this country. Schools are experimenting with technology including facial recognition, license plate readers, audio gunshot detection, and patrol robots. In Broward County, Florida, where the Parkland school shooting occurred, there are now more than 12,500 cameras in schools and $11 million was spent on cameras last year. The school district is planning to spend another $600,000 on next-generation surveillance. Such surveillance is concerning to students, teachers, and privacy advocates. — AXIOS

Japan's Henn-na Hotel chain is decommissioning many of its AI robots because of maintenance issues and complaints from guests. Robots that act as front-desk workers, cleaners, porters, and in-room assistants are being replaced by traditional human staff. Dinosaur and humanoid bots at the front desk were unable to respond to certain guest questions, robot luggage carriers malfunctioned when wet and were not able to reach all the hotel rooms, and the in-room voice assistant "Churi" could not understand some accents. According to one guest, Churi thought his snoring was a command and kept waking him up by asking him to repeat his request. — SOUTH CHINA MORNING POST

Facebook AI Research (FAIR) director Yann Lecun and a team of NYU researchers are proposing a new method for training autonomous vehicles. Conventional reinforcement learning frameworks use traffic data and vehicles navigate in a digital replica of the real world, but when the training data doesn't include certain scenarios, problems can occur. The new method doesn't just reward and penalize driving behavior, it also has penalties when a vehicle strays into a situation where there is not enough training data. The effect is that the vehicle approaches new situations more cautiously. — TECHNOLOGY REVIEW

Researchers at the University of Science and Technology of China in Hefei developed an AI system that can recognize a person's emotional state. The system uses audio processing algorithms that rely on speech spectrograms (visual representations of sound frequencies) as well as a series of face recognition networks. The model was trained on 653 video-audio clips from the AFEW8.0 television and film database. The research is published in the arXiv online repository. — VENTUREBEAT

-- Commentary --

This week, I had the pleasure of interviewing Dr. Louis Rosenberg from about their "swarm AI" technology, which has so far had some pretty impressive.  In a season-long test of the Swarm AI's ability to forecast NBA games, researchers found that swarms were able to easily beat Vegas betting markets with a 25% Return on Investment. By using a deep neural network to determine only which games would be most profitable, that ROI more than doubled to 57%.  If you want to know more about Swarm AI and what it can do, read the Q&A below.

1.  What does Unanimous do, and how did the company originate?

While most AI companies are working to replace human intelligence with AI algorithms, Unanimous AI amplifies human intelligence by connecting networked groups of people together into super-intelligence systems.  In layman's terms, we build "hive minds" by connecting group of people together in real-time, moderated by AI algorithms, enabling them to converge on solutions that are significantly more accurate than the individuals could do on their own.  

Our technology is called Swarm AI because it's modeled on the biological principle of Swarm Intelligence - the phenomenon that enables flocking birds, schooling fish, and swarming bees to be significantly smarter together than alone.  We have empowered networked human groups to achieve the same thing by connecting together online, enabling significantly more accurate forecasts, assessments, decisions, evaluations, and insights. 

2.  What was your background prior to Unanimous?

My background is very technical, having earned a PhD from Stanford in Robotics and Human Computer Interaction.  My doctoral work was conducted at US Air force Labs in the early 1990s and produced the first augmented reality system.  Here is fun picture on Wikipedia running an experiment back then (  After that, I founded Immersion Corp, one of the very first virtual reality companies and the only one to have gone public on Nasdaq.  After that, I took a few years off and became a professional screenwriter.  This might sound like an aside, but it was while working on science fiction screenplays that I started to get very interested in collective intelligence and hive minds, and I started to realize that mother nature has been optimizing group intelligence for hundreds of millions of years.  And it made me realize that people have been using polls and surveys and prediction markets to optimize the intelligence of groups, but it's not the right way to do it.  The right way is to connect groups together using AI algorithms modeled after natural swarms.  So... after selling a few screenplays, I went back into tech and founded Unanimous AI to explore this very simple question: if birds and bees and fish can get so much smarter by thinking together in systems, why can't people do it?  And it turns out, we can.  That's what Swarm AI technology is and it works better than we ever expected. 

3.  Deep learning has been powerful by its own right, why is this combo better than deep learning alone?

There are two parts to that answer.  First, machine learning only works if you have a structured database with lots and lots of samples to train on.  There are many situations that fit those conditions, but there are countless others were the data that is important does not exist in a database somewhere - it exists as the knowledge, wisdom, insights, and intuitions of people.  With Swarm AI we tap into that "human database," building systems that leverage this very powerful, very diverse, and fully unstructured source of information.  We can then take out output of a Swarm AI system and run it through a neural network, enabling us to use the power of traditional AI to optimize the insights produced by a swarming system.  The second point that I wanted to make is that humans, in addition to having vast knowledge and wisdom, also have the remarkable ability of "inference" which enables good decision in situations that have never been seen before.  Machine learning fails in such situations, as it needs to have seen something very similar for it identify it.  This enables humans to perform well on categories of problems where ML performs poorly.  But, when you combine Swarm AI and traditional AI, you can achieve a whole that is greater than either one alone. 

4.  How does a "swarm" differ from "wisdom of the crowds"?  Can a swarm be composed of anybody or does it need expertise?

When people talked about crowds, they generally mean collecting polling data from a larger group and finding a statistical aggregation.  This works, but it's not the best way to combine the knowledge and wisdom of a population because it reduces each individual to a few simple data-points.  Swarms are very different, as they require all participants to interact at the same time, pushing and pulling on each other a system as they converge together on optimal solutions.  In this way, each participant is not a passive data-point, but an active "data processor," assessing and reassessing the problem as the group converges on a solution with the support of AI algorithms. 

5.  What's the next step for swarm AI technology?  What problems will you turn it to in the future?

Our goal at Unanimous is to keep making "human swarms" smarter and smarter.  We believe we've just scratched the surface of how smart these systems can be.  We continue to refine the underlying algorithms.  We are also looking at what happens when we form larger and larger swarms.  We've seen remarkable results with swarms of 5 people, 50 people, even 150 people, but what happens when we get to swarms of 50,0000 people?  We believe this is worth exploring.

6.  Have you considered opportunities to combine swarms with non-neural network types of AI?  Would swarm genetic algorithms make sense?

We have looked at combined use of swarms with other forms of AI.  We believe there are many approaches to leveraging the power of human intelligence with the efficiency of machine algorithms.  It will be exciting to see what works the best. 

7.  What tools and techniques do other companies need to make this work for themselves?

Unanimous has just released (in beta) a SaaS platform called SWARM that will the members of any business team to login, form a swarm, and amplify their intelligence.  We believe business teams will want to use this to quickly make better forecasts, predictions, estimations, decisions, and prioritization.  We've already worked with some of the top brands in the world, having their teams form swarms and converge on AI-optimized solutions. 

8.  What AI problem that you aren't working on at swarm, do you wish someone would solve to make your technology better?

Because our AI needs to run in real-time (processing human behaviors as part of an interactive system), we are always excited when technologies make data processing faster. 

9.  Late last year there was a big debate between Gary Marcus and a bunch of other AI leaders about how far deep learning can take us.  Where do you weigh in on that debate?

Personally, I believe Gary Marcus makes great points.  I resonate most strongly with the points related to the data requirements of  deep learning.  To solve real problems in the real world, most people underestimate how hard it is to get a data-set that actually represent the situation on the ground they are trying to solve.  The world is a very complex place.  And humans have a remarkable ability to continually update their mental model about the world.  This is why I am so interested in tapping the "human database" as a different approach to AI.  


As always, thanks for reading.  See you next weekend.


Copyright ©, All rights reserved.

Our mailing address is:
767 Bryant St. #203
San Francisco, CA 94107

Did someone forward this email to you? Head over to to get your very own free subscription!

You received this email because you subscribed to Inside AI. Click here to unsubscribe from Inside AI list or manage your subscriptions.

Subscribe to Inside AI