Happy Sunday and welcome to the latest edition of Inside AI! For those of you who are new, I'm Rob May, CEO of Talla, (an A.I. Driven Enterprise Service Desk/Bot for I.T. and H.R. teams) and active investor in the A.I. space. If you have an early stage A.I. startup looking for investment, I'd love to hear from you. Recent investments include Smartvid, Omelas, and Rocketvisor.
This week's issue is awesome. If you agree, I hope you will forward this to a friend so they can subscribe too.
Also, while this newsletter will remain free and continue to go out on Sundays as always, we now offer additional content through Inside A.I. Premium. It comes with three things: 1) an extra midweek issue of links and updates, 2) A monthly profile of 4-5 startups in the A.I. space, 3) An extra monthly report that is similar to my current commentary but longer and more in depth, plus occasional other bonus content. The first few reports coming out this summer will cover the business opportunities in NLP, the impact of neuromorphic chips on A.I., and an analysis of A.I. in publicly traded companies. You won't want to miss them, so consider signing up.
One last note - tomorrow (Monday 6/19) I'm on This Week In Startups. I'm sure Jason and I will have some good banter, so tune in.
-- BIG IDEA --
The most interesting thing I read this week was this post by Bradford Cross discussing why the best approach for an A.I. startup is to go full stack vertical. It isn't just one of the best posts of the week, it is one of the best business posts I've ever read about A.I. Cross highlights a conclusion we also reached at Talla based on our own strategic analysis, which is that being able to rapidly gather new types of labeled data will be the key to success in many A.I. business models. What makes this post so awesome is how he ties everything together and explains why this is the best approach. Can startups compete against the giants of tech, or is A.I. the sport of kings? We've discussed that several times here, and I think more and more it is evident that startups can have a very strong position. Go read the post now. It's awesome.
-- MUST READ LINKS --
How Facebook Uses A.I. To Fight Terrorism. Facebook.
The first time the company has talked openly about the A.I. methods they use to identify terrorist accounts and terrorist content.
DCVC's Investment In Element.ai. DCVC Blog.
You probably read the news that Element.ai raised an impressive $102M, led by DCVC. There has been some confusion as to whether Element is a company, incubator, or fund. Read this to understand DCVC's view of the investment and Element's strategy.
The Rise of A.I. Friendly Design. Medium.
The world is designed for humans. Will we soon design it for A.I.s and robots? Fascinating idea.
General Game Playing With Schema Networks. Vicarious.
Vicarious was founded around ideas similar to those of Numenta, that modeling the human brain using something like autoassociative memory is more important than current NN approaches to A.I. They've been criticized (including in this newsletter) for appearing more like a venture backed research project than a company. But this current breakthrough is interesting because it shows that a game playing reinforcement learning model may be overfitting, and this idea of Schema Networks may be a better way to understand game playing, and a big step up in A.I. Read the discussion section at the end of the post.
Forget AlphaGo- DeepMind Has A More Interesting Step Towards General A.I. MIT Tech Review.
A summary of DeepMind's paper on relational reasoning, something not achieved previously in A.I., but very important to build GAI.
Bad At Negotiating? Facebook Is Working On Bots To Do it For You. Fast Company.
Really interesting research out of Facebook on "dialogue rollouts" which sound sort of like what a chess program does when it analyzes moves ahead, but, this uses natural language.
The Economic Impact of A.I. on Customer Relationship Management (PDF report). Salesforce.
This report from Salesforce makes the wild claim that A.I. will give CRM a $1 trillion boost, which is very self serving and may be grandstanding. Nonetheless, the report contains some great data and is worth reading.
-- INDUSTRY LINKS --
A.I. Could Target Autism Before It Emerges. Wired.
Automating The Law: A Landscape of Legal A.I. Solutions. Topbots.
Why Voice Tech Is About To See Major Love From VCs. Medium.
Carmageddon is Coming. Future Crunch.
US Intelligence Agencies Are Starting to Build A.I. Spies. Quartz.
3 Steps To Embedding Artificial Intelligence in Enterprise Applications. Forbes.
Make Pharma Great Again With A.I.: Some Challenges. Hackernoon.
Deep Mind Now Learns From Human Preferences Just Like A Toddler. New Scientist.
What Deep Learning Can't Do. Bharath Ramsundar Blog.
Social Robotics - Using Robots To Bring People Together. Magenta.
How Machines Learn: A Basic Machine Learning Pipeline in Production. Eliza Effect.
Wouldn't You Like Alexa Better If It Knew When It Was Annoying You? IEEE Spectrum.
A.I. and ML in Startups. Medium.
How An Artificial Brain Could Help Us Outsmart Hackers. World Economic Forum.
Let's Be Honest About the Achievements of A.I. Medium.
A.I. Can Comb Through Your Data To Create More Compelling Customer Experiences. Harvard Business Review.
How Do You Evaluate A Conversational Interface In The Buying Process? Talla blog.
A.I.s $37 Billion Market Is Creating New Industries. Venturebeat.
A Solar Powered Weed Whacking Robot. Robohub.
Applying Deep Learning To Real World Problems. Merantix Blog.
How To Be a Winner In The Consumer Robotics Revolution. Entrepreneur.
Understanding Deep Learning Requires Rethinking Generalization. KDNuggets.
-- JOBS --
The Berkman Center at Harvard is hiring an A.I. project coordinator.
-- RESEARCH LINKS --
Sobolev Training For Neural Networks. Link.
Meta Learning Framework For Automated Driving. Link.
Towards Grounded Conceptual Spaces in Neural Representations. Link.
The Bradford Cross article from the "Big Idea" section above is very timely because it highlights something that isn't necessarily obvious - that A.I. is going to drive big changes in user interfaces for enterprise software. If you look at recent history, enterprise ux has been through a few different phases. The first was moving from on-prem to cloud, where the ux was about making the software work in a web browser, and being really easy to setup and deploy. The second was to ride the consumerization of I.T. wave and make enterprise software more like consumer software - mobile friendly, more social like, and easier to use. To understand how A.I. is going to impact ux, we have to understand the advantages of A.I..
So far, training models effectively has required vast amounts of data and thus the value of A.I. has accrued to the big tech companies that have such large data sets. But improvements in A.I. and techniques that use smaller data are slowly eroding that advantage. As better models and ideas like one-shot learning lower the amount of data required to create useful A.I., the axis of value shifts from having a large training set to having a training set in the first place. There are so many areas where we don't even have a small data set to train on.
If you look at areas like finance, where some of these A.I. techniques have been applied longer than they have anywhere else, and everyone has access to lots of data, the value chain shifts. Speed and creativity with the data becomes important. I think a similar trend will happen in A.I. in a few years.
What does that mean for enterprise A.I. companies? It means the key to staying competitive will have to do with how fast you can get data, to train a model for a new product feature. This will filter into ux design, because the key to good enterprise ux/ui won't be first and foremost about easy of use, or making it seemingly consumerish, but instead, the key will be how well it collects data points to train a machine learning model. What does this mean in practice? Well, it's one great use of chatbots, which I wrote about a year ago in InsideBigData. But conversational or visual ux/ui will both need to be thought of from a machine learning perspective first.
You don't have to build for that immediately, because the small data tactics aren't here yet, and you really should start with some enterprise use case that builds on an existing data set. Otherwise, the bootstrapping data problem may be too difficult. But this need for training data will become a standard way to think about UX/UI, so put some thought into it early, and it will be easier to extend your product and keep it defensible.
That's all for this week. Thanks again for reading. Please send me any articles you find that you think should be included in future newsletters. I can't respond to everything, but I do read it all and I appreciate the feedback.
-- ABOUT ME --
For new readers, I'm the co-founder and CEO of Talla, I'm also an active angel investor in A.I. I live in Boston, but spend significant time in the Bay Area, and New York.