Inside AI - June 21st, 2017 |

Inside AI (Jun 21st, 2017)

Inside AI Premium Edition: Why A.I. Will Replace Knowledge Workers First, Not Last

Subscribe to Inside AI

New blank template
Subscribe | View in browser

Welcome to the first edition of Inside AI Premium.  This first one is going out to all free and premium users as a sample, so those of you not on premium can decide if you want to sign up.  After just 1 week, we are the #1 Premium subscription newsletter on the network, so thank you!  And if you aren't premium yet, consider signing up if you want more mid-week content.

This mid-week piece will consist of the top 5-10 things I've seen so far this week, with some summary, and once a month will include a special piece of commentary not available to free subscribers.  This month's commentary is about why, contrary to popular opinion, A.I. will replace knowledge workers first, not last.  Premium subscribers will also get special reports and analysis, and a monthly look at a handful of A.I. companies.  The first 5 coming in a few weeks are in the NLP space.

I want to thank for being our first corporate sponsor.  Please check them out.  As this newsletter starts to monetize, the sponsorships and subscriptions will pay for a bunch of original research we are starting.  Stay tuned for more awesome stuff.

-- Industry Links --

When A.I. Can Transcribe Everything.  The Atlantic.  Human level accuracy is almost here, which will increase the demand for, and availability of, transcriptions of everything.  What could it mean?  " Captioning on YouTube videos could be standard, while radio shows and podcasts could become accessible to the hard of hearing on a mass scale. Calls to acquaintances, friends, and old flames could be archived and searched in the same way that social-media messages and emails are, or intercepted and hoarded by law-enforcement agencies."  Exciting.  And scary.

Google Open Sources One Machine Learning Model To Rule Them All.  ZDNet.  Neural networks have tended to have a consistent topology.  You hear about convolutional neural nets or recurrent neural nets or other examples.  Google has combined a bunch of different ideas into a single neural net.  "They claim that the single model can concurrently learn a number of tasks from multiple domains and that the model is capable of transferring knowledge. It is able to learn from tasks with a large amount of training data and apply that knowledge to tasks where data is limited."  Pretty awesome innovation.

The National Science Review has a great interview with Thomas Dietrich.  NSR.  Dietrich is the President of the Association for the Advancement of Artificial Intelligence, so he has great insights into what is going on in the ecosystem.  When asked about the impact of A.I. on jobs, he notes, "I think it is similarly very difficult today to predict what the jobs of the future will be. There will certainly be jobs involved in creating AI systems, teaching them, customizing them and repairing them. I suspect that it will not be cost-effective to completely automate most existing jobs. Instead, maybe 80% of each job will be automated, but a human will need to do the remaining 20%. That human thereby becomes much more valuable and will be paid well."

Berkeley A.I. Research now has a blog.  BAIR Blog.  The first post, Learning to Reason With Neural Module Networks, is technical, but excellent.

Inside Microsoft's A.I. Comeback.  Wired.  A great inside look at how Microsoft wooed and signed deep learning pioneer Yoshua Bengio.

Ginni Rometty says that IBM "woke up the A.I. world".  CNBC.  Rometty makes some excellent points about how differently IBM is positioned compared to the other tech giants with powerful A.I..  IBM is the only one solely focused on business use cases.  In this weekend's newsletter, we will have a look at a new report on machine intelligence that will come out later this week, and one point of analysis in that report is why vertical A.I. is so important.  It seems IBM is well positioned if that is true.  I've asked the author his thoughts on IBM, so stay tuned Sunday to see what he says.

A very philosophical piece on A.I. - is the concern intelligence or autonomy?  NPR.

Many thanks to Inside AI's corporate supporters:


- Why Knowledge Workers May Be the First To Succumb To the A.I. Revolution -

For my first premium commentary, I want to challenge your the conventional wisdom a bit today and throw out an idea that hasn't been mentioned very frequently.  There have been a bunch of reports in the past year about which jobs will be replaced by A.I., and when.  The approach to do this analysis is typically to guess at when A.I. is capable of performing that task, and then assume shortly thereafter A.I. does it all.  But such analysis is shallow.  Simply creating the capability for A.I. to do something is only one of many steps to replacing a workforce.

As an entrepreneur, I think a lot about how to roll out innovation.  Simply creating something isn't enough.  You have to think about production, behavior change, and so much more.  The conventional wisdom is that jobs lower on the cognitive ladder will be replaced first, and more cognitively complex jobs last, but that really hinges on what it takes to execute such a replacement once the technology is available, and how long the gap is between A.I. solving a cognitively simple problem and amore cognitively complex problem.

Here is an example.  Say you solve the problem of an A.I. accountant.  There is little physical work an accountant has to do, so, you can replicate that A.I. agent or some variant of it really quickly.  Now say you solve a problem that is less cognitively complex but has a physical component - say a construction worker.  Rolling that out to replace all construction workers is much more complex.  You have to build robots.  They require parts, and assembly, and shipping.  They probably have more constraints for how they work with existing parts of the ecosystem built for humans.  Rolling out these A.I. robotic construction workers will have many more real physical challenges and impediments than rolling out a fully digital accountant, because most accounting information is already digital.  It's more plug and play.

So the way this plays out in the real economy depends on the time lag between solving the A.I. construction worker problem and the A.I. accountant problem.  If the latter is solve 8 or 10 years after the former, then yes, maybe construction workers are replaced first.  But if it's 2 years, accountants may fall first while we work through all the physical requirements of rolling out A.I. construction laborers.

You can add other variables into this.  For one, the U.S. government may decide that certain roles need humans for national security reasons, regardless of whether A.I. can do their job.  Certain labor unions may negotiated anti-A.I. deals with large companies.  Roles that pay at the bottom of the wage scale to begin with may not be worth the effort to replace with A.I. because the cost savings is negligible.  Some jobs may need to wait on legislation to define how A.I. issues in that job (errors and mistakes, for example) are treated before they can be rolled out, and companies won't want to roll out A.I.s when they don't understand their liability.  Every job that could be automated away will have a different set of variables impacting it's actual roll out.

I don't have a clue how all this plays out, because it is too complex to predict.  What I am trying to point out is that the notions we have right now about how A.I. will take jobs are very very simplistic, and basically tied to "when can an A.I. do the work?"  That is such a small part of the equation actually.

There is a book I wrote about in a very very early edition of this newsletter, almost 2 years ago, called Manna.  In that book, mid level managers, the ones who instruct others what to do, are the first layer to be eaten by A.I.  It starts in a restaurant where employees wear an earpiece and the A.I. can tell them every next step to their jobs.  The story is one reasonable future of how A.I. advancement could play out.

When you think about rolling out A.I., think about all the parts, not just the cognition.  And keep your eye on legal trends, social acceptance, and implementation costs because they will matter as much, if not more, than solving the actual problem of creating an A.I. for that task.

That's all for this mid-week premium edition of Inside AI.  Thanks for reading, and let me know if you have specific topics you want to see covered in the newsletter.


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