LinkedIn

View James Drake's profile on LinkedIn

Thursday, August 15, 2013

Different Copyright Law for Machine Authors - NOT YET!!

In his latest contribution to Wired, Algorithms Are the New Content Creators, and That’s Bad News for Humans, author Peter Wayner bemoans the "exemption" that so-called "aggregators" enjoy over human authors in relation to copyright law and speculates that this gives non-human "authors" an unfair advantage over their human counterparts.

Mr. Wayner concedes that he is not a "practicing lawyer" and then proceeds to support his assertions based on several fundamental misconceptions and misunderstandings of copyright law. As a frame for his discussion, Mr. Wayner talks about the prohibitive cost that he would have incurred if he had tried to include licensed copies of photographs from the 1949 and 2012 Broadway productions of "Death of a Salesman" in his book Attention Must Be Paid, But For $800?

The headline in his piece caught my attention, because, if true, it signaled a new advance not only in jurisprudence, but also in the abilities of artificial intelligence.  I had not been aware that algorithms were now authoring books!  After digging further into the piece, I found that Wayner's definition of authorship was rather expansive:
Anyone who searches for “Death of a Salesman” gets search results with a nice sidebar filled with a few facts and some images that Google scraped from websites under fair use. In this way, they can do things that I, a lowly human, can’t do. And while I had to pay $10,000, they could “get” them for free.
One could make the case that, in the act of assembling graphics and written materials together, the aggregator algorithm is, indeed, creating content.  Such content is hardly the equivalent of a book, however. Further, the fair use defense to copyright infringement has some well-defined general categories, as established in 17 U.S.C. § 107:

Notwithstanding the provisions of sections 106 and 106A, the fair use of a copyrighted work, including such use by reproduction in copies or phonorecords or by any other means specified by that section, for purposes such as criticism, comment, news reporting, teaching (including multiple copies for classroom use), scholarship, or research, is not an infringement of copyright. In determining whether the use made of a work in any particular case is a fair use the factors to be considered shall include—
(1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
(2) the nature of the copyrighted work;
(3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
(4) the effect of the use upon the potential market for or value of the copyrighted work. 
Moreover, the fair use defense is not why these aggregators are free of liability.  Copyright protects the tangible form in which a work is fixed, not the information contained in that form.  The classic example of this is the phone book.

As decided in the case of Feist Publications, Inc., v. Rural Telephone Service Co., 499 U.S. 340 (1991), the Supreme Court of the United States established that information alone without a minimum of original creativity cannot be protected by copyright. According to the facts on record in the case, Feist had copied information from Rural's telephone listings to include in its own, after Rural had refused to license the information. The Court rejected Rural's copyright infringement law suit and stated that information contained in Rural's phone directory was not copyrightable and that therefore no infringement existed.

This is the rule that search engines and other aggregators have relied upon for years in relieving them of liability for their search results.  Although the information cited by Wayner is an example of integrating written and graphic search results, it cannot be said to be copying the original material and representing it as the work of the aggregator.  No one who comes upon such a page as the result of an internet search looks to it as an original work of authorship.

Mr. Wayner seems to be aware of this hole in his argument, because he follows up his complaint with the following:
We already know these companies make a profit on the ads. But what we don’t know is that the algorithms they use are acting less and less like a card catalog for the web and more and more like an author. In other words, the machine isn’t just a dumb hunk of silicon: It’s a living creator. It’s less like a dull machine and more like a fully functional, content-producing Terminator.
The fact that these companies derive commercial benefit from listing this information is not evidence of copyright infringement.  It would be a stretch to argue that the ad revenue generated by these aggregations in some way affects potential market for or value of any underlying copyrighted works.  To the contrary, one could more convincingly argue that a searcher would be more likely to seek out and/or purchase the underlying work once they were made aware of the work's existence!

To suggest that an algorithm that combines scraped materials in different and more engaging ways is an author still misses the point.  In order for these companies to be at risk of copyright infringement, they would need to package the scraped material into a format that directly copied the original material.  Aggregating bits and pieces from various sources is less fair use and more transformative use. Indeed, Wayner indicates that this is the basis upon which Buzzfeed stands in defending itself against copyright infringement accusations.

Having attempted to construct a problem out of a perceived inequity in the application of fair use, Mr. Wayner concludes by proposing the use of an algorithm to fight an algorithm.

This approach assumes that an algorithm can be sophisticated enough to determine what is fair use and what is excessive borrowing.  Many years of jurisprudence have not been able to provide any bright line rules about when borrowing crosses over into infringement, and, without binary rules that establish what is and what is not infirnging, such an algorithm would be hopelessly inadequate, much as decades of pornography filters have been flummoxed in trying to determine what is or is not prurient.

While the questions posed by Mr. Wayner are pertinent to the threat posed by more and more creative AI systems, he has used a strawman as the basis for an inadequate solution to a currently nonexistent problem.  Copyright law is not unequally applied between man and machines -- at least not yet!

Tuesday, April 30, 2013

Security Awareness Training is a Waste of Time

Schneier on Security: Security Awareness Training:

Schneier focuses his essay primarily on IT security training, but the lessons are instructive when it comes to IP security issues such as confidentiality and document management training as well.  Let's look at the parallels.

As Schneier points out training succeeds or fails for a finite variety of reasons:

One basic reason is psychological: we just aren't very good at trading off immediate gratification for long-term benefit. A healthier you is an abstract eventually; sitting in front of the television all afternoon with a McDonald's Super Monster Meal sounds really good right now
This is applicable to IP security as well.  Training everyone in your organization to stamp "Confidential" on every piece of paper that is covered under a confidentiality agreement is, in the abstract, a very good way to protect proprietary information - especially if you keep the information as a trade secret.  However, a customer-facing employee tasked with getting critical information to a customer is likely to place the tangible immediate benefit of an improved customer relationship ahead of the intangible benefit of possibly preventing a leak of the information through lax security procedures.
Another reason health training works poorly is that it's hard to link behaviors with benefits. We can train anyone -- even laboratory rats -- with a simple reward mechanism: push the button, get a food pellet. But with health, the connection is more abstract. If you're unhealthy, what caused it? It might have been something you did or didn't do years ago, it might have been one of the dozen things you have been doing and not doing for months, or it might have been the genes you were born with. Computer security is a lot like this, too.
IP security and management of proprietary information is also a lot like this.  IP is, after all, intangible property.  You can't see somebody walking away with it and it is oftentimes very difficult to pinpoint the source of a leak, especially if it was inadvertent.  As a result, employees make judgments at a personal level regarding the value of the information that they disclose, the risk of making the disclosure and the likelihood that the risk will be realized.  Often, these risks appear minimal when compared to a more immediate, tangible benefit.

One area of health that is a training success is HIV prevention. HIV may be very complicated, but the rules for preventing it are pretty simple. This is important: most lay medical expertise stems from folk models of health. Maybe they're right and maybe they're wrong, but they're how people organize their thinking.  
Another area where training works is driving. We trained, either through formal courses or one-on-one tutoring, and passed a government test, to be allowed to drive a car. One reason that works is because driving is a near-term, really cool, obtainable goal. Another reason is even though the technology of driving has changed dramatically over the past century, that complexity has been largely hidden behind a fairly static interface. You might have learned to drive thirty years ago, but that knowledge is still relevant today. On the other hand, password advice from ten years ago isn't relevant today.

Schneier's examples of training success are similarly instructive, since they involve either a minimal effort in exchange for offsetting an enormous, understandable and personal risk (HIV) or achieving a short-term attainable "cool" goal (attaining the privilege of being permitted to drive a car) that has long-term persistence once achieved.  As important as the underlying commonalities of successful training programs, is the method used to achieve that training success.  As alluded to by Schneier earlier, it has to do with creating a limited decision-making framework comprised of simple rules that do not seek to improve the "average" performance of the organization until we have removed systems and processes that are vulnerable to the behavior of the weakest link:

Even if we could invent an effective computer security training program, there's one last problem. HIV prevention training works because affecting what the average person does is valuable. Even if only half the population practices safe sex, those actions dramatically reduce the spread of HIV. But computer security is often only as strong as the weakest link. If four-fifths of company employees learn to choose better passwords, or not to click on dodgy links, one-fifth still get it wrong and the bad guys still get in. As long as we build systems that are vulnerable to the worst case, raising the average case won't make them more secure. 
The whole concept of security awareness training demonstrates how the computer industry has failed. We should stop trying to teach expertise, and pick a few simple metaphors of security and train people to make decisions using those metaphors. We should be designing systems that won't let users choose lousy passwords and don't care what links a user clicks on. We should be designing systems that conform to their folk beliefs of security, rather than forcing them to learn new ones. 
IP security should look to similar models.  IP managers must assume that employees with access to proprietary information are not experts in handling that proprietary information and that they nonetheless have a need to share information with potential and existing customers.  An overall organizational strategy based on limiting the number of decisions employees are allowed to make with respect to proprietary information will require long-range planning and cross-functional support (IT, legal, commercial), but it will be well worth it.  In the next post, I will look at ways that this can be accomplished.


  

Wednesday, April 17, 2013

How Jayne's Hat got entangled in Trademark Law

Threatening Those Who Create the Rights:

More important than the trademark implications is the wildly myopic reaction of Fox.  As Pamela Chestek puts it:

Fox had a community most companies would kill for. But rather than embracing and fostering this passionate community, Fox took the short view. In knee-jerk fashion it exercised legal rights of questionable merit, acting autocratically and failing to recognize that it had something infinitely more valuable, the allegiance of a passionate community.

Monday, April 15, 2013

Responding to USPTO's new Maintenance Fee Regime - Patent Law Blog (Patently-O)

Responding to USPTO's new Maintenance Fee Regime - Patent Law Blog (Patently-O):

On March 19, 2013, the USPTO's new fees became effective. The most notable changes were the maintenance fees due at three intervals: 3½, 7½ and 11½ years after patent issuance. At each interval, the fees increase with the new top-fee set at $7,400 -- a 50% increase over the old fee.
Hard to predict the overall effect of the increased fees, but it is a good bet that it will decrease the number of unexploited patents that are still active after 7.5 years.

Did You Pay To Name A Planet? You Got Scammed | Popular Science

Did You Pay To Name A Planet? You Got Scammed | Popular Science:

Just a reminder that the Copyright Office has no authority to confer naming rights or ownership in any extraterrestrial body, be it star, moon, planet or black hole.  All that they do is register creative ideas that have been fixed in a tangible form.

The most (in)famous example of this as an ongoing scam is the execrable "International Star Registry", which, very truthfully, advertises that "your star name will be registered in book form with the Copyright Office".  Most people think that this means that this process somehow conveys official status to their name.  In fact, all it does is register a book with the Copyright Office that has the name in it.  Cost of registering your star name with the ISR?  Anywhere from $54 to $489 (including a "certificate" and "heirloom frame" at the higher levels)!!!  Cost to ISR for registering a copyright in your name? $35 (online only filing) to $105 (hard copy).

Saturday, January 19, 2013

New USPTO Fees Announced

     Lots of changes coming with the new fee schedule, effective March March 19, 2013. This is, of course, just three days after the March 16, 2013 date on which the First-to-File changes of the America Invents Act become operative.

     Several changes have been highlighted by Dennis Crouch over at Patently-O, including highlighting the 75% microentity discount that will become effective on March 19th. As recommended by Professor Crouch, since they qualify for the discount as “Institutions of Higher Learning" under 35 U.S.C. 123(d), it might benefit universities to file provisionals now and then convert to utility application once the new fees become effective.

     The PTO is doing its best to encourage applicants to file early by lowering the cost of filing through issuance, as shown in the following Table:



     At the same time, there will be increased pressure to monetize patents or otherwise come to an early decision regarding maintaining patents as part of a larger portfolio, since the maintenance fees have increased dramatically, as follows:

Sunday, January 6, 2013

EPO - EPO and SIPO launch Chinese-English machine translation for patents

From the European Patent Office:

As a result, the collections of the two largest languages in patents are now united as full-text documents on the same website through the EPO's global patent database, Espacenet, and linguistically accessible for innovators from both regions using a single tool, Patent Translate. Under the new arrangement with SIPO, Espacenet has grown by 4 million Chinese language documents, adding to the 75 million documents already available.