The Market Ticker
Commentary on The Capital Markets- Category [Technology]
2017-05-12 11:35 by Karl Denninger
in Technology , 429 references
[Comments enabled]  

Let's dispose of the commonly-spewed horsecrap right up front: There is no such thing as "artificial intelligence" or "machine learning."

What does exist is pattern recognition and the price of doing it well has gone down very rapidly as the cost and power of computing devices has gone down and up, respectively.

Decades ago doing it "well" required mainframes and then only in certain segments -- because it was simply impossible to analyze a larger data set.  AC Nielsen made their business out of analyzing supermarket checkout scanner data.  So did IRI.  Both did it using big iron and they were only looking at specific data from a specific source, which they paid for (frequently by subsidizing the installation of all that hardware in the grocery stores) and then sold back to the suppliers of those stores (food companies like Heinz.)

That was a symbiotic relationship.  The grocery store got helped twice: First with the capital cost for installation of (at the time) extremely expensive checkout line scanners, which allowed them to bust the checker's union and hire people at half their previous salary -- and then again in that the overriding goal of a grocery store is, of course, to sell more groceries.  As such empowering the food suppliers in figuring out what sold and what didn't had no negative impact on the stores and in fact it was to their benefit.  While two potato chip companies fighting over an endcap might be amusing to a store manager it didn't do violence to the store's sales numbers and actually might improve them.

Today running much more complex pattern analysis requires only some server racks full of Xeon processors and SSDs -- a tiny fraction of the cost of a big IBM or Amdahl mainframe for much more capability.

But this belies a far larger problem when it comes to companies like Salesforce, Amazon and others: Essentially no firm, even the behemoths like Amazon and Salesforce, have enough data of their own to provide the enormous data sets that lead to better and better pattern recognition.  In fact companies like Salesforce have, for all intents and purposes, no data of their own upon which to apply said algorithms at all!


To put it in simple terms: There is no symbiotic relationship, as will become clear in a moment.  In fact being a part of and plugged into these systems is actually likely to screw you as a company.

What does IBM's "Watson", for example, have to suck up to continue to process language at an ever-more accurate level?  Lots of language -- all of which has to come from somewhere.  With language it's pretty easy to come up with a huge data set on an ever-increasing basis that nobody actually owns, because speech is all around us all the time.

There are exceptions.  Visa, Master Card, Discover and American Express, for example, have very large and high-quality data sets to detect fraud from which they own because between those four they basically own the market.  Their volume is so high that they need nobody else's data to make good decisions nor do they have to share any of what they have.

But few companies -- even huge ones like Amazon -- have such a luxury.

So where do you think all these data sets come from that the "machine learning" touts are talking about?

I'll give you one place and in fact arguably the biggest placeAll the ad networks your computer is "accidentally" connected to.

Let me point out just one example: Fox News' web site, if I look at it critically, tries to transmit data to 30 different connected networks when I simply look at the top page.  And that's with transmission blocked.  With it unblocked it literally tries to connect data to over 100 different places on the Internet because each of the 30 passes on data to others.

This is not unusual.  In fact it's not only common it's universal!

So where do Salesforce and Amazon, never mind all these other sites, get their data sets?

From you, short -- both from when you contemplate buying something and when you actually buy something.

That shouldn't shock anyone because it's what data analysis is all about.

That sounds ok but in fact it's a huge problem for these "analysis" companies and their customers: Most if not all of that data winds up being massaged and then given back to others including their customer's competitors!

Essentially nobody has figured this out yet, but they eventually will.  Where do you think all those ads come from that target something you just looked at or bought somewhere?  If you didn't buy it on Amazon but was clicking around various other places like B&H Photo and similar how did Amazon get the data?

They got it from a competitor.

Indirectly, in many cases, but that's where it came from.

That sound great for Amazon but it really isn't because the same data is sold to anyone who wants it and that means that Amazon's competitors can and do buy that same data too!

This is true across the spectrum and is the dirty little secret about so-called artificial intelligence.  What is being sold as "AI" is not intelligence at all; it's pattern recognition and to feed it you need extremely large sets that can be crunched.

Since nobody owns that sort of data set on their own with damn few exceptions and none of the purveyors of these "solutions" are those exceptions the data that is going into them includes yours and it winds up being sold to competitors too!

This will eventually find recognition in the marketplace and when it does the paradigm of 100+ competitors to Fox News being given the data on what I read there who will then use it to pick off Fox's customers will run into a "wee problem" from Fox.

You've been sold a bill of goods from the likes of Mark Cuban and others in the tout investment game folks.  The entire concept of broad-based "machine learning" is utter and complete horse**** and so are all of the firms engaged in sustaining themselves on it at exponentially ever-higher valuations.  It is nothing more or less than pattern recognition and in order to have a data set large enough to provide an "advantage" it must inherently take data from other people and sell it on, including to the data source's competitors which inherently damages the very firm who is their client and the firm that provided the data itself!

That's the opposite of symbiosis -- it in fact is parasitic!

When the market recognizes this and the sources revolt against giving their trade secrets away to competing firms while paying for the privilege besides this entire segment of the so-called "Internet craze" will collapse.

View this entry with comments (registration required to post)

2016-10-31 07:00 by Karl Denninger
in Technology , 295 references

Give me a break.

A task force of more than 30 major technology and communication companies said they have made progress but have not found a solution to eliminate "robocalls" or automated, prerecorded phone calls, but a top U.S. regulator urged faster action.

Throw some people in prison and you'll get their attention.  Yes, right here in the US, and yes, I'm talking about carrier executives.  Why?  We'll get to that:

Wheeler wrote major companies in July urging them to take new action to block robocalls, saying it was the top source of consumer complaints at the FCC. Scam artists often times based abroad try to appear to call from a bank or a government phone to trick consumers into disclosing confidential financial or account information.

How do they "appear" to call from a bank or government phone when they're not in the United States?

Ah, now see, there's the fraud and the US carriers are complicit in it.

Along with a call setup request (from one carrier to another) comes some information, which includes the "originating" number.  The carriers do exactly nothing to validate that for other than 800 (free to calling party) numbers.

But they could very trivially prevent, for example, foreign calls from appearing with US numbers.

How?  Refuse to route a call that comes from the UK unless the "originating" number is in the correct format including the country code prefixfor example.

That would stop instantly any of these calls that are originating outside of the United States.

As for those within the United States the FCC has jurisdiction, and can require that one of two things be the case:

1. The "originating" number be the actual originating number.  This will be the appropriate setting for all individual lines; simply do not allow an overridden number from a consumer account -- period.

2. For those that are overridden require, under penalty of law, that the party overriding accept both civil and criminal legal responsibility for the authenticity of their override under existing criminal fraud statutes.

There are very good reasons to allow such an override on outbound calls.  For example, at MCSNet we had outbound trunks that were all "rolled up" into high-capacity circuits (at the time DS1s); each of those trunks had a "real" phone number, but it was unpublished.  We then had DID mapping for certain people who needed "private lines" and in addition we had our "main" number (312) - 803-MCS1 that would ring into the PBX on the next available trunk in the group.  If you dialed out from our PBX those trunks (set up for bidirectional signalling) were configured to show 312-803-MCS1 as the "originating" number even though technically it was not.  That's fine, because we owned the originating number, it was "real", and it really was our number.

It would not be difficult at all to require that all such entities that purchase service from a telco provider in the United States and wish to provide "originating number" overrides do so under a contractual requirement, carrying criminal criminal penalties for lying, that any such number they put through be truthful and belong to the actual originating party of the call.

If you were to do this and at the same time hold carriers criminally responsible for accepting "foreign" calls that have originating numbers that violate the country code format of the originating nation, a software check they could easily implement, this problem would disappear instantly.

Of course there are "telco providers" (such as the SIP folks) that would scream about such a requirement -- but let's face reality here.  Enabling fraud as a business model makes you an accessory before the fact and recognizing that along with appropriate criminal sanction would go a long way to draining this swamp -- quickly and permanently.

Instead we "accept" a bunch of handwaving nonsense that comes from the FCC and various telcos.

View this entry with comments (registration required to post)

Main Navigation
MUST-READ Selection:
The Bill To Permanently Fix Health Care For All

Full-Text Search & Archives
Archive Access

Legal Disclaimer

The content on this site is provided without any warranty, express or implied. All opinions expressed on this site are those of the author and may contain errors or omissions.


The author may have a position in any company or security mentioned herein. Actions you undertake as a consequence of any analysis, opinion or advertisement on this site are your sole responsibility.

Market charts, when present, used with permission of TD Ameritrade/ThinkOrSwim Inc. Neither TD Ameritrade or ThinkOrSwim have reviewed, approved or disapproved any content herein.

The Market Ticker content may be sent unmodified to lawmakers via print or electronic means or excerpted online for non-commercial purposes provided full attribution is given and the original article source is linked to. Please contact Karl Denninger for reprint permission in other media, to republish full articles, or for any commercial use (which includes any site where advertising is displayed.)

Submissions or tips on matters of economic or political interest may be sent "over the transom" to The Editor at any time. To be considered for publication your submission must include full and correct contact information and be related to an economic or political matter of the day. All submissions become the property of The Market Ticker.