Inside AI - November 20th, 2016

Inside AI (Nov 25th, 2016)

Happy Friday and welcome to the latest edition of Technically Sentient!  For those of you who are new, I'm Rob May, CEO of Talla and active angel investor in the A.I. space.  I'm sending this issue out on Friday because I'm sure many of you in the U.S. have the day off, and may enjoy filling your Friday with some great A.I. reading.
-- BIG IDEA --
Benedict Evans has a nice blog post about Cameras, Ecommerce, and Machine Learning.  What I find most interesting about it is this passage:

Now, suppose you buy the last ten years’ issues of Elle Decoration on eBay and drop them into just the right neural networks, and then give that system a photo of your living room and ask which lamps it recommends? All those captioned photos, and the copy around them, are training data. And yet, if you don’t show the user an actual photo from that archive, just a recommendation based on it, you probably don’t need to pay the original print publisher itself anything at all. (Machine learning will be fruitful grounds for IP lawyers.)

Many people are saying that "data is the new oil" and I think this use case is particularly interesting.  What will I.P. law say about data training sets?  As machines get smarter, will we treat them the way we treat humans with respect to these laws?  Lots of interesting questions to think about here.

Feature Engineering Is Easier.  Lab41Blog.
A great look at feature engineering vs architecture engineering and how it impacts the problem domain in machine learning. 

Cutting Through The Machine Learning Hype.  RRE Blog.
Great introductory post by VC Jason Black, with lots of clear examples to back up his thinking.  A good read if you are new to the space.

What A.I. Can and Can't Do Right Now.  Harvard Business Review.
Andrew Ng writes about the promise and problems of the current state of A.I.

Four Steps To Turn Neural Dust Into Reality.  IEEE Spectrum.
This is important because technologies like this are the most likely way we augment ourselves to keep up with machines.

Siri Architect Grapples With How To Make A.I. Less Artificial.  PCMag.

Could A.I. Erase Fear From Your Brain?  Mashable.

Good Robot Design Needs To Be Responsible.  ReadWrite.

Vinod Khosla Says 80% of I.T. Jobs Will Be Replaced By Automation.  Futurism.

How To Hold Algorithms Accountable  MIT Tech Review.

SAP is finally getting into machine learning.  WSJ Blog.

Panasonic has invested $60M in a new robot that sorts and folds laundry.  Telegraph.

Has China surpassed the U.S. to be the new leader in A.I.?  Medium.

Why Nature Is Our Best Guide For Understanding A.I.  Techcrunch.

What Problems Will A.I. "Solve" in the Next 5-10 Years?  Medium.

Key Questions in ML and AI in the Enterprise.  ZDNet.

Humanity and A.I. Will be Inseperable.  The Verge.

Chatbots 101:  A Primer for Developers  IBM Watson Blog.

The World's First Photonic Neural Network.  MIT Tech Review.

How Machine Learning Can Create a Smarter Electric Grid.  Energy Collective.

Scholars Delve Deeper Into The Ethnics of A.I.  NPR.

A.I. Can Now Lipread Better Than Professionals.  NewScientist.
A great video of Daniel Shank (Talla Data Scientist) on Neural Turing Machines.  Link.

Generating Machine Executable Plans from Natural Language Instructions. Link.

Local Minima in Training of Deep Networks.  Link.
I'm going to start accepting a few A.I. job postings in the email, next week will have a sign up form, so stay tuned.
A few months ago I was chatting with a partner at Matrix who told me his concern about investing in A.I. companies is that it was becoming "the sport of kings."   He said this because, to really do big A.I. things at the moment seems to require lots of data, lots of compute, and lots of PhDs - a trio of things only found at the top technology companies in the world.  

A similar idea was recently posited by Professor Enrique Dans where he asks if A.I. is the new digital divide.  He writes:

In short, very soon, companies will be divided between those who are able to take advantage of artificial intelligence and machine learning for their day-to-day operations, and those that continue to operate as they always have, making them much less productive and much more unpredictable. We are talking here about the emergence of a new digital divide, a virtually Darwinian event in terms of competitiveness.

It is a great point, and raises some good questions.  Are there opportunities for startups?  Or, is A.I. the sport of kings?  It is probably a little bit of both.  

I've seen a fair number of startups doing things for consumers that are "similar to the Netflix recommendation engine, but for X" where X can be travel, events, music, etc.  These all seem like things the existing tech consumer giants should do, and things that will be tough to win.  But there are reasons to be optimistic.

First of all, the B2B side of A.I. is much more open.  Since most data sets there are private, and more difficult to access, and more specific to the owner/company, those markets are less likely to be the winner take all markets we've seen in consumer tech.

Secondly,  there are plenty of areas the existing tech companies will not touch.  I've seen some cool A.I. for the construction industry, and my guess is that Google, Facebook, Amazon, and Microsoft are all staying clear of that space.  The nice thing about A.I. is that it's going to be more widely applicable to industries than software is/was, because of the predictive abilities and the fact that A.I. can be physically manifested in robots.

The third reason to be optimistic is NLP.  As natural language understanding becomes a solved problem (still a few years away, but, close, I believe), this will open up language as a data input mechanism for A.I., which will have explosive impacts.  While the big consumer tech companies will benefit from this, there will be so many opportunities to gather so much data through natural language that they can't possibly pursue them all.

This last point is key.  I think if  you are building an A.I. company, you start by thinking about data sets that you can access, or build, that don't currently exist.  I've seen, and think we will continue to see, a lot of A.I. business models that are build X to gain Y and use Y data to build Z.  These can be tough to get funded unless X is valuable in and of itself, so, you almost have to find two valuable businesses instead of just one, to be successful.  But plenty of people are doing it.  I believe many of these opportunities will initially be too small for the big tech companies, and they will miss some of these markets and be forced to pay up and buy into them.

To get back to the article from Professor Dans, I think he makes another great point, that the real barrier to entry is not really technology, but ignorance of its possibilities.

This is an incredibly interesting point, because depending on the level to which it is true, it means that the most successful companies in A.I. may not be big companies or startups, but rather, the first movers.  But, when you think about first movers in the A.I. space, very very very few people have a strong enough grounding in all the various pieces (tech, data, business) to understand what is possible and where to look for opportunities.  Just knowing you have a data set doesn't mean it is useful for any new business opportunities.  And unless you understand the data and algorithms on a strong conceptual level, you may not have an idea of what is possible.  And if you do know what is possible, you may not know whether or not that is a good line of business to get into.

So the problem is not just that A.I. is nascent and not as well understood as traditional new product development in software.  The problem is also that predicting whether or specific problem can be solved with A.I., or what problems might be solved with certain data sets, is also not well understood.  The people who have good intuition about all aspects of this stuff are few and far between.  It's the reason you see so much junk A.I.

I guess what I am saying is that, yes, I think for the clear obvious applications of A.I., it probably is the sport of kings.  But I believe there is a whole series of applications we are currently blind to, which are probably the "sport of explorers."  What I mean is, those who spend the most time exploring this new world, doing the experiments, and developing an intuition for the real opportunities, are the ones that will win.  And it is hard to predict who that might be and where it will come from.  What I do know is, it will come into focus fast, given the number of people pushing into the space, and the pace of innovation.
That's all for this week.  Thanks again for reading.  Please send me any articles you find that you think should be included, and A.I. angel investment deals that look interesting, or just your general thoughts.  I can't respond to everything, but  I do read it all and I appreciate the feedback.   Enjoy the rest of your weekend.


-- ABOUT ME --
For new readers, I'm the co-founder and CEO of Talla,   I'm also an active angel investor.   I live in Boston, and spend about 30% of my time in the Bay Area (Talla has a Palo Alto office) and 10% of my time in NYC.  
Inside is powered by Mailchimp. Also, we ❤️  Mailchimp.

Copyright © 2016 Inside, All rights reserved.

You're receiving this email because you are subscribed to Technically Sentient. If you don't want to receive it anymore, go ahead and unsubscribe – or just hit reply and tell us how to make it better.

Subscribe to Inside AI