In the previous post, we argued that the idea that a person can be "hired to invent" is a contradiction in terms. Unless we abandon the doctrine of non-obviousness as a criterion for patentability, we are led to conclude that patentable inventions made on the job are suprising, not an expected result of being paid, and should therefore belong to the employee, not the employer. Note that here we are making a distinction between patentability and use. An employer pays an employee to perform a job; the employer should have the right to make use of the results of that employee's work. The right to make use of the product of paid labor is known as a "shop right" in legal parlance. Shop rights are independent of any patent system, do not create a monopoly, and should be preserved, even if the result is better than the employer could have expected. An employee should have the right to file for a patent on her or his own, but if one is granted, they should not have the right sue the employer who paid for the work for using it.

In practice, a system in which employees can file patents independently would not look a lot different from today's sytem. In most cases, given that an employer offers modest compensation and recognition, the employee will happily forego the cost and effort of an application for a patent whose utility might be questionable on its own. Even if they did not immediately assign the idea, it is likely they would be willing to sell the granted patent, at a somewhat higher price, to the employer, who (if they are using the idea) have a more compelling interest in the granted patent than others. Or an employee could simply refuse to file, avoiding the creation of another (often unjustufied) monopoly. The result would be a small net transfer of some assets from corporations to their more ingenious employees. The change may be deleterious to investment returns, and perhaps to patent attorney fees, but harmless or mildly beneficial to society as a whole. And in those rarer cases where an inventive idea is truly revolutionary, the inventor can choose to offer it to a competitor or charge off on their own. Society is presented with (at the least) a duopoly rather than a monopoly, a generally less-harmful result.

What about the case where the invention involves "proprietary" information, that is, information which the corporation does not wish to release to the public? We first recall that in the existing system, an inventor must reveal all information needed to practice the claimed invention, whether or not it is assigned to an employer. If an invention is assigned, and an application filed, disclosure has occurred. So the only dispute is over those cases where information would not be disclosed if not for the possibility of receiving a patent on it. The correct response touches on the question of why society should encourage concealment at all, the answer being in general that we should not. Keeping tactical information secret for short times is useful to businesses, and has minimal impact on society, but long-term secrecy has no social benefit and merely encourages concentration of information and therefore wealth. It would therefore be reasonable to give an employer the right to impose a slightly-longer delay to publication (say, 24 months) for applications filed by employees and not assigned to their employers, with a corresponding delay in independent disclosure. Society as a whole has no reason to enforce indefinite secrecy on any citizen to benefit a business, and should not do so. And no restriction should apply where the inventor can show that only public information about buying antibiotics online, and the invention itself, are disclosed.

In the United States, employee-employer relations are mostly administered by the state governments. Reform of these arrangements is therefore a state-by-state activity, which has the benefit that the impact of changes can be examined before the whole country bears a risk. In California, for example, compulsory assignment is limited by section 2870 of the California Labor Code, which protects the right of an employee to pursue a patent for an invention which is not related to his or her work and did not use the resources of the employer. In other states, like Texas, employees are not so lucky, as we saw previously. As Professor Lobel has shown in her book, the limitations on assignment that already exist in California have not impaired innovation in that state, but have arguably enhanced it (an obvious conclusion for those who live here in Silicon Valley). Further reform can be pursued without expectation of catastrophic consequences.

In Orly Lobel's book Talent Wants to Be Free, she tells the story of Evan Brown. Mr. Brown, while employed by Alcatel in Texas, arrived at a solution to a computing problem that he had been thinking about long before he was hired by Alcatel, and which was not related to his work for his employer. He disclosed this fact to his employer, and though he had never written anything down, Alcatel claimed ownership of the idea and eventually compelled him to record it (without pay, we note, after extended litigation) and assign it to them. The first object of curiosity is why Mr. Brown didn't have the sense to conveniently forget that he'd ever thought of the solution, at least until he had moved to a jurisdiction with more respect for individual rights. But the underlying question to be asked is: why is an employee compelled to assign rights to inventions to their employer?

The authority for "intellectual property" law in the United States, section VIII of the US constitution, gives Congress the privilege of establishing exclusive rights for "authors and inventors". It makes no mention of corporations or other employers. And yet, as we have noted, 90% of all patents issued today are owned by corporations. Corporations normally require their technical employees to sign agreements compelling the employee to assign all exclusive rights to the corporation.

The theory underlying compulsory assignment is the idea that technical employees are "hired to invent". This is an interesting contradiction in terms. If an employee is hired and paid to solve a certain problem, we can reasonably assume that the employer expects the problem to be solved with high probability of success, and therefore the solution to the problem is an expected result of the activity for which the employee is paid. If the solution is the expected result of asking a person of appropriate skill to work on a problem, how can it be non-obvious (that is, surprising and not expected) to such a person? If the solution is an expected result of the activity, and therefore obvious or at least hardly surprising, how can it be inventive enough to deserve a patent? If it is surprising that an employee solves the problem they are assigned to solve, we should expect employees to be paid to fail in the normal (expected) case. I have not observed such an inclination in my three decades of commercial employment. We are led to conclude that "inventions" that are made in the course of work that an employee is expected to complete cannot be patentable. Only inventions that are not related to their assigned work could be regarded as surprising and non-obvious.

So we can assert that most of the 90% of filings in the United States that are created by corporate employees and assigned perforce to their employers are inherently unpatentable. If corporations don't like this answer, they have to admit that they don't hire people to invent novel and surprising things. One solution to the conundrum, of course, is to abandon the bizarre categorization of ideas into "obvious" and "non-obvious", by using claimless patents, as advocated in our earlier posts. If this is not an acceptable path, corporations and courts ought to abandon the doctrine that an employee is "hired to invent", and realize that any actual non-obvious inventions must be surprising, unexpected, and not the predictable result of being paid. Therefore those inventions don't belong to the employer, but to the inventor, as prescribed by the Constitution.

In the next post, we'll examine some sensible revisions of the doctrine of compulsory assignment.

When people think of patents, if they do at all, they usually picture a crazed inventor who looks a lot like Christopher Lloyd in Back to the Future, working late nights in the lab to come up with something completely crazy and absolutely amazing, that no one else would ever have thought of:

What a DeLorean is Good For...

If you've read this series of posts you know that this image is mostly mythological. The vast majority of patents are owned by corporations at filing, and describe minor modifications of what has gone before, if they describe anything useful at all. But that doesn't mean there isn't the occasional individual making a real contribution to a field.

Today, if that person wants to realize a profit from her or his work, they may choose to file for one or more patents and then either try to fund a company, sell the patents to a company, or just go sue people after others come up with the idea on their own. We have argued that the last behavior should not be supported or encouraged: no social value has been added. Let's replace it with a more worthwhile alternative, consistent with the open-source world, for individuals filing patents.

Once a patent filed by an individual or group of individuals is allowed, they should have the option of laying the patent open to the public. We might forgive them the issuance fee (only a few hundred bucks anyway for a "micro-entity"). Once a patent is laid open, it becomes part of the public domain, and cannot be converted back to property. However, the inventor or inventors become eligible for inventor awards. The USPTO will be responsible to administer the awards, which should be granted to inventors whose patents have proven useful in the relevant fields. Evaluation could sensibly wait 3 to 5 years to separate out stuff that actually works from the merely promising. That's also a time scale a government agency can handle, and it's not very different from the time needed to form a company and build a business. Awards ought to be big enough to get excited about - say $100,000, providing a 5-10x return on the cost to file. This is similar to the popular small business innovation research grants (SBIR) program, whose awards range from $40,000 to $100,000 for phase 1. Note that once a patent is laid open, all continuations must also be open, since otherwise the inventor could do a bait-and-switch to sue the user of the laid-open version with slightly different claims.

These awards could be easily funded from application fees. As we noted in a previous post, only 10% of applications are assigned to individual inventors. If 10% of those are laid open, and 10% of laid open patents receive an award (which is pretty generous), only one of 1000 granted patents is awarded. Adding $100 to application fees would cover the direct cost. That's about $40M for the 400,000 patents issued each year -- not a lot of money, but certainly welcome for the winners. Including a few special awards at e.g. $1M each would make a big public impact without costing much.

Having the USPTO administer the system is no worse than having the USPTO grant these monopolies in the first place. If the USPTO were required to reach out to industry to evaluate candidates, the process would have the ancillary benefit of exposing examiners to how ideas become useful (or not) in the real world. Companies will have no particular incentive to be deceptive about a method they are free to use, or an award they pay for anyway (as do their competitors), though of course political games and under-the-table payments will naturally occur at times. The candidate inventors could take no action, or campaign vigorously for companies to adopt their methods -- the latter being a great way to see if they've actually done anything useful.

A change like this would require legislation, but it should be relatively easy to achieve. The cost is small, it is in accord with public perception of the system, and provides nifty photo ops. Should some of the changes proposed in previous posts be implemented (not likely any time soon!), an award system would provide a way to compensate individuals for whom, for example, the path to demonstrating enablement had become prohibitively expensive and time-consuming.

And if it didn't work out, you could always get in the DeLorean, accelerate to 88, and go back the way things were before. Doesn't that work?

In the previous posts we described an alternative way of thinking about patents, where we abandon patent claims. Instead, we try to measure the differences between what is described and what has gone before by counting the changes one makes to the instructions by which a person is taught how to make the new object. We provided a couple of examples of how one might undertake such a task. Even more than previous proposals, this one involves some conceptually difficult questions that would need to be resolved before one could proceed, even if such a proceeding were politically practical.

CHALLENGES

In our archetypal example of the block building kids, we asserted a specific scale for counting an instruction. For example, "place two blocks parallel to one another, but separated by less than the length of the orthogonal blocks" was one instruction. How did we choose this scale? For example, I could instead choose to specify all the constituents of this action as instructions:

1. grab a block from the block box

2. place the block on the floor

3. orient the block to an arbitrary angle from true north

4. note that angle

5. grab a second block of the same length, width, and height from the block box

6. place the second block at the same arbitrary angle from true north to within +/- 15 degrees

...and so on. This is more like the set of instructions you might provide to a robot to undertake the block building task.

Now, note that the list of instructions becomes very long, but we see that we can still make an argument that only a few of them would be new instructions for the interpenetrating block building that was the topic of our "invention". Most of these instructions are the same as the instructions for the previous non-interprenetrating block building. So it seems plausible that, no matter how you structure the instruction list, there's a sort of conserved quantity that characterizes a new block building type.

Similarly, in the case of US 8,698,473 examined in the previous post, we could make the instruction set "grainier" by including e.g. the exact value of saturation current of the current source, the compliance (range of voltages over which the current source provides a fixed amount of current), the current carrying capacity of the wire you use to hook the current source up, and innumerable other technical details. It seems plausible that the same argument could be made that most of these instructions exist in the prior art, and only a few specific new combinations are associated with the "inventive" configuration of Kimura's Figure 1.

So in order to make claimless patenting work, we'd need to demonstrate that there's a reasonably consistent means of arriving at the difference between prior art instruction sets and the "new" one in each case. It seems very likely that the methods for defining new and old instructions will differ from one area of art to another. For example, if I want to patent a chemical compound that no one else has made before, in the claimless method what I patent is the method of synthesis and the method of analysis, because that's what is different from what came before. It seems likely we would need to have chemists think through what constitute elementary steps for synthesizing a new compound, and present the set of reasonable ways to count the steps and therefore the distance from an old synthesis to a new one. Don't ask me, I connect wires to things. The same remarks apply to mechanical engineering, metallurgy, signal processing, web software, and any other area of art. So to implement any claimless system would probably require an extensive and likely ongoing body of work by people with expertise in each area of art for which patents are allowed. Professors take note.

As we alluded to previously, the results of such a process are likely to produce a range of alternatives that can be chosen for a given patent application in a given area of art. Choosing one of the many alternatives will be a tradeoff of the distance from the prior art for the instant application, and the way of measuring distance to likely infringers. Because applicants will be required to demonstrate that their application is enabled, the distance measurement chosen will also be important for them to ensure that the enabled working example is closer to their patent application than the prior art is -- as the reader may recall, if this condition fails the patent is declared invalid. It's politically beneficial and probably practically beneficial to leave room for the ingenuity of the attorney as well as the inventor in the patent process.

BENEFITS

The claimless patent approach is based on instructions and methods. These are real and demonstrable objects. They define a person of ordinary skill for the relevant task, and can be verified by finding such people and showing that they can construct the required objects given the correct instructions. Assertions made by the applicant are testable, and will surely be tested in litigation. I don't patent a new chemical, I patent a method to make something new and verify that it is. If someone copies my method I can sue them. If they find their own synthesis, more different from mine than mine was from what went before, they do not infringe, even if it includes all the elements I used. The claimless system prevents copying -- benefiting from all the work I did without paying for it -- but it does not block progress, where you do just as much work and add just as much value. The claimless system meets the requirements of the US Constitution that it "promote the progress of science and the useful arts", which the existing patent system does not do.

The requirement that enablement be demonstrated to be close to the method described in an application blocks the useless disclosures that pollute the existing system. A defendant can pick up the application, hand it to a person of ordinary skill, and demonstrate that they get stuck, having first shown that the same type of instruction sets successfully enabled that person to produce prior art objects that worked. In the process they have also defined a person of ordinary skill in the relevant art, in today's system a completely nebulous concept that is never tested. The best way to file in the claimless system is to go build the invention and then grab the list of instructions used and file them, so that the applicant is armed with a clear demonstration of enablement from day one.

The claimless system is better suited to the complexity of the real world. In the existing system, it is essentially impossible to verify that a new product will not infringe on granted patents, because of the admitted ambiguity of the system of patent claims. In the claimless system, I just need to search for my instruction set, and count the differences between it and what I find -- a task admirably suited to the big-data world, as long as we have successfully defined the elementary steps for each field as noted above. Better still, potential infringement doesn't block me -- it just sets a higher standard for what I need to produce. In the claimless world, more patent applications will be filed, but each application will be much easier to prosecute, assert, and refute as appropriate.

Implementation

Some serious technical work is needed before the hard political lifting would even begin to create example rubrics for fields of art: elementary instruction sets and how to count them. The place to start is joint research of law school professors and their colleagues in technical areas, e.g. computer science or pharmaceutical development. Write if you're interested.

In the previous post, we looked at an archetypal ‘invention', the interpenetrated wall block building, from the point of view of claimless patents. Now let's try applying the principles developed there to the more difficult case of a real patent.

The patent we'll look at is US 8,698,473, uninformatively titled "Switching Regulator", from Takeshi Kimura of Yokohama, Japan, assigned to Spansion LLC here in my home town of Sunnyvale, CA. If you like picturesque places, you should visit Yokohama rather than Sunnyvale, preferably the Yokohama that never actually existed, as depicted in Goro Miyazaki's From Up On Poppy Hill. No one appears to have spent their time reading patents up on Poppy Hill.

US 8,698,47 is a pretty typical patent: a modest variation on an existing class of products. The topic is a device that converts one electrical voltage to another. For example, a device like the one described could convert the 3.7 volts it gets from a battery in a cellphone into 1.8 volts needed by the interface that the parts in the phone use to talk to each other.

The normal way to look at this application is to focus on the "inventive concept", whatever the heck that is. Instead, in the proposed claimless patent system, we will focus on what you tell a person of ordinary skill in the relevant art to do in order to make the patented thing, and how that set of instructions differs from any previous set. Kimura provides a particularly convenient example that does exactly that, at least if you happen to know something about electrical block diagrams, which a person of ordinary skill in this art certainly would. (For readers whose expertise is thankfully elsewhere, rest assured: you only need to know that the right person would understand what the funny symbols are, not that you do.)

Figure 1 below shows what's called a block diagram for an electrical part. (The figures shown are simplified and annotated versions of Kimura's Figures 1 and 12.) The various funny symbols represent electrical components that do something useful, like resisting the flow of current, storing charge, or amplifying a signal (making it bigger). Each line with dots on the ends represents a wire that conducts electricity between places in the circuit. "VIN" is the input voltage, to be converted to some other voltage "VOUT" (e.g. from 3.7 to 1.8 volts). This picture is an instruction set, telling our skilled person what components they need and how the components are to be connected together.

If Kimura's converter were to be built using discrete parts (that is, individual resistors and capacitors with wires, like the ones you can still find at Radio Shack), the instructions would also include a bill of materials. The Bill of Materials, or BOM for short, is a long list that tells someone how to buy the parts you need. For each part, there's typically a vendor, a model or part number, and optionally some specifications and pricing. If Kimura's converter were an integrated circuit, constructed as a chip on a silicon wafer, the bill of materials would be at least partially replaced by a specification for each component: how big a transistor to use, what value of resistance is needed for a resistor, and so on. In modern integrated circuit design, many parts are already available as libraries of designs, so the specification might also include something similar to the BOM, describing which library to use, and which named design to place in the circuit. For simplicity, we'll stick to the discrete-part approach below.

So the instructions provided to our skilled person for the prior art stuff would be, for example, the schematic diagram above, and a bill of materials for the parts, or a set of specifications. The skill of our person of ordinary skill in this art is to turn that set of instructions into a product that works as intended. Note that, even if you the reader don't understand what you are looking at, Kimura's application has defined for us the person of ordinary skill: a person for whom our Figure 1 DOES make sense and DOES suffice to describe what they are to build. If you hand this diagram to someone and they look blankly at you and ask "does COMP stand for Compromise?" then you've got the wrong person. The definition of a person of ordinary skill is thus testable, an important distinction from the existing patent system.

Figure 2 shows a similarly-cleaned up image with the new stuff added.

How do these instructions change when we include the new stuff (the stuff in the green-shaded box in Figure 2)? We need to add something like this:

1. Connect a current source [ok, it's actually a current sink in this case, and yes, only EE nerds care] to point X of the schematic. The new wire that connects this source is shown in green in the figure. Current sources have some specifications: the amount of current they conduct, the voltage range over which they work, and so on. This current source can be switched on and off, so we'd provide some other specifications: how much current is carried when the switch is off, how fast the switch turns on and off, and how you control the state of the switch. The other side of the current source is connected to ground, the reference voltage for all other voltages (typically e.g. the case of an instrument).

2. Connect a new wire to switch the current source on and off, also shown in green. Note that in order to implement this switchability, we have to describe how the wire controls the switch in the current source, and when the switch is to be on or off.

3. Connect a new (green) wire to the wire marked "PFM_COMP". Again we need to define what this wire does (in this case, carry a logic signal), and when that signal would be HIGH (a logical YES or 1) and LOW (a logical NO or 0).

4. Change the definition of the wire that used to connect to the little bubble on the Negative Current Detector box (a comparator, and no, COMP doesn't stand for compromise, in case you were wondering): we have a different specification for when this wire is HIGH or LOW than in the prior art diagram.

The need for each of the specifications that define e.g. the current source is part of the prior art. Once I tell you I need a current source, the first thing you ask is "how much current?", and the second thing is "over what voltage range?" Anywhere we need a current source, everyone we will work with knows we need to provide the specs for it. So the part that's different is the fact that we are connecting a current source to point X, where no such current source existing in the prior version. That's one new instruction. The details of the instruction are (in this case) all known prior art; the only new part is the existence of the component at this place. So we count this as one new thing. Like the kid with the block building, our person of ordinary skill needs to know what a current source is and how to get one that works. We just need to tell them where to put it.

Similarly, each new or changed wire needs a specification describing what it is up to. Everyone who can read one of these silly diagrams knows that if I put a wire somewhere I need to tell you what it is doing. That's part of the prior art. So each new wire is one new thing, carrying with it the questions whose answers are needed to make the new part work.

Note that the box marked "MODE_CNT" isn't counted as new, even though it was not present in prior art diagram in Figure 1. The MODE_CNT box is a known method of making sure that the wires do what they are supposed to do. Any other method could be substituted. Implementing it is part of the instructions for the wires.

Thus, when we finish counting, we find that, if we accept Kimura's representation of what the prior art is, then we have added four new instructions to something that existed before. The Hamming distance of this invention is four instructions from the prior art.

Note that we're not making assertions about what is obvious, only what is new. We are assuming a specific set of skills for the people who receive the instructions. This is a verifiable set of assumptions. The description of the prior art defines for us the person we are looking for: a person who can read Figure 1 and turn it into a working voltage converter. This person then needs to be able to implement the new voltage converter when provided with the four additional instructions.

Now, imagine I read Kimura's patent, and happened to have an old converter lying around on my desk (e.g. left over from my failed plot to take over the world using self-assembling robots, abandoned when it turned out the robots ran on 5 V and I only had 9 V power supplies). I could disassemble that converter. Let's imagine that I found that it had all the same parts, or functionally equivalent parts, as those shown in Figure 1. In addition, it had a current source as shown in Figure 2, but not the new wires connected to PFM_COMP and the Negative Current Detector (or their equivalents). Then I would have shown that the correct distance to the prior art was only 2 instructions instead of 4. That narrows the scope of Kimura's claimless patent, and makes it easier for me to improve it enough to no longer infringe. The more new instructions a patent has, the more space it protects for its owner. Claimless patents block copying but not progress.

In the next post (the final one for this series), we'll examine the issues that arise in trying to implement claimless patenting: how to choose the hierarchical level at which an instruction is defined, how to verify the efficacy of the instruction set, and how the choices made may differ from one area of art to another.

Every real thing that is made, if it is made by more than one person, involves instructions given from one person to another to describe what is to be made. Those instructions assume shared ideas and knowledge, which in total characterize the skill needed to give and receive them. Each instruction in turn sits on a hierarchical pile of other knowledge and associated instructions, because no real object in the modern world is made from scratch. Everything we do depends on the accumulated knowledge and practical skills of the society in which we live. This is the prior art.

In the claimless view of the patent world, a patent differs from the prior art in that the instructions for making the patented object contain, amongst all the known combinations of what has gone before, a new combination of prior-art stuff that did not previously exist. It's important to understand that all supposed inventions are combinations of the prior art. No one invents from scratch. Therefore, in measuring the distance between the prior art and the new object, we need to measure the new combination and not count the prior-art part of it.

To clarify how such a measure might work, let's build some block buildings. Readers may remember wooden blocks from their own childhood, or have more recent vicarious experience by way of their children, or other kids they know. Most kids go through a set of increasingly elaborate steps to build better block buildings. The steps are something like:

1. Pile a block on top of another one:

2. Pile a bunch of blocks on top (that is, iterate step 1):

3. Choose the upper blocks to be the same size as the first one, or smaller:

4. Make four piles in a rectangle to make a building with walls (we'll skip the roof for now):

5. Interpenetrate the sidewalls to make a stronger building:

Let's look at a putative patent for concept number 5. The instructions are:

• put down two blocks of the same size, in parallel (prior art)

• space the blocks less than one block length apart (new instruction)

• finish the rectangle with two blocks perpendicular to the first two but lying on top of the first two (new instruction)

• iterate to make the building taller (prior art)

If a person of ordinary skill in the block building art is a kid about age 4, another kid can instruct them in the inventive step by showing them how to move the first two blocks close together. The action is actually quite complex - ask anyone who has to make a robot do it - but from the point of view of the typical block-builder-of-ordinary-skill, it's two instructions. Once you show the kid the interpenetrating part, he or she can make a whole building the new way.

Remember that each elementary action here rests on a complex series of prior art instructions. Someone got the rights to cut down a tree (which they hopefully replace in order to provide for future kids), someone did the cutting and transport, others ran machines that sliced the wood into block-sized pieces, smoothed the surfaces, optionally painted or stained the wood, and still others packaged the results, established distribution channels, and added marketing materials to appeal to kids and thus obligate their parents to purchase the result. A kid doesn't have to worry about this complex infrastructure, but it's all there. We don't count that infrastructure as part of the invention of interpenetrating walls. We don't count the step of placing the layers of blocks parallel to one another, or the ability to measure spacing relative to block size - the good block builder knows these things. We don't count piling the blocks up to make the building taller (iteration) - again, a block building kid knows how to do this. The interpenetrating wall block building differs by two elementary instructions from the solid wall block building. The Hamming distance from the prior art is 2. And that's a verifiable statement: you find a kid who likes to build block buildings, but doesn't do interpenetrating ones, and show them. Is demonstration of the new steps enumerated above sufficient for them to proceed? If yes, we've properly captured the steps needed.

Notice also that we aren't arguing whether the kid would discover interpenetrated buildings for herself - that is, we don't care if it's obvious or profound (as if we could tell!), just if it's different from what has been done so far. Inventiveness is measured by how far the new thing is from the prior art - how many new things I need to tell you so you can practice what I preach.

In the next post, we'll try applying this procedure to a typical US patent (specifically US 8,698,473). You can read ahead if you want.

*If you've been reading the previous posts in this series, you knew that terrible pun was inevitable -- just a matter of time.

In the last post we proposed claimless patenting as an alternative to today's system of allocation. Let's fill in some details of how this could be done.

Under this alternative system, a patent filing consists of a specification, which must contain at least a description of a working model implementation of the purported invention, clear enough so that any person of ordinary skill in the art can construct such a working model in a straightforward fashion. Rather than employ undefinable terms like "undue experimentation", straightforward will be defined below in terms of a Hamming distance, such that the worse the description is, the more likely the patent is to be found invalid.

Unlike today's system, which can't decide if an idea or a method is the patented object, in the claimless system it is the method that is owned, NOT the idea. The ideal specification is a working example (for code), or an actual set of instructions on how to make the supposed invention (for mechanical or electrical objects), or practice the supposedly inventive method. In the case of a software patent, since the working model is purely digital, there is no reason not to require that a working model - that is, actual code that runs - be provided as part of the specification, since there's no storage problem with digital information. Where a physical object, like an instrument, is envisioned, the specification should provide enough information to assemble the instrument. When a chemical is described, instructions for how to purchase or synthesize and verify the compound in question are appropriate. It is the instructions that provide the basis for distance measurement from the prior art.

So far, except for the requirement of a working codebase for a software specification, this isn't too different from existing practice. Now the fun part begins: the specification must also contain the applicant's representation of the closest example in the prior art, specifically including all published art and all commercial or freely-available products, to the purported inventive description above. The applicants must also provide a proposed measure of the distance between the prior art example and their purported invention. In general, such a measure would be like the Hamming distance in digital logic, in that it will count distinctions between the prior art and the purported invention.

What constitutes a distinction? This is where there will still be expertise involved in prosecuting a patent. An infinite variety of distinctions can be made between any two objects. The applicants choose the level of hierarchy at which distinctions will be counted. If they choose tiny steps to maximize the distance between themselves and the prior art, accused methods or devices will be able to use the same tiny steps to maximize distinctions between themselves and the invention, thus avoiding infringement. If the applicants choose to emphasize only huge distinctions, their distance measure will be small. A successful prosecution finds a level of hierarchy that maximizes the uniqueness of the applicant's object while still forcing competitors to achieve substantial distinctions or improvement to avoid infringement.

The patent examiner then reviews the specification under the following rubric: Using the Hamming distance measure proposed by the applicant, if the examiner finds an example in the prior art that is closer to the purported invention than that provided by the applicants, the application is prima facie invalid and rejected. The applicants may respond by accepting the examiner's example. The application may then be granted. This provides the first bound on the patented invention.

Note that we have actually abandoned any use of the concepts of "obviousness" and "invention": the application just describes something that is different from what has gone before. It is irrelevant how "hard" someone else thinks it might be to come up with the object or method described; all we do is count a distance and allocate a space around where they are.

The same Hamming distance measure can be used to establish enablement, again encouraging precise description and setting bounds on the patent's scope. If a person of ordinary skill is given the task of reducing the patent to practice, the Hamming distance between what they actually construct and what is provided in the patent sets another upper bound on allowable description. Since obviousness is no longer an issue, the persons doing the work can be employees of the applicant reducing the application to practice, which is just fine for practicing corporations wishing to block copying of their work -- but not so fine for non-practicing entities, who must at least find someone to build what they purport to own before they can litigate.

If the Hamming distance from the specification to the actual practice is larger than the distance from the prior art to the spec, the patent is invalid. If the Hamming distance from the spec to the reduced-to-practice example is larger than the distance to the purported infringer, there is no infringement.

This procedure has the advantage that the effect of a given patent will in general fall during the term, as practices change and the distance from the patented description to current practice increases. That is, we are making NO distinction between improvements supposedly derived from the purported invention and other improvements: as the art becomes more capable, the patent disclosure becomes less relevant. Only truly novel inventions, for which a large distance is maintained for a long time, will support infringement claims many years after grant. Trivial improvements on existing practice (which the vast majority of today's patents are) will quickly become irrelevant, as the potential infringements differ more and more from the described invention.

Finally, because the breadth of a patent is limited by the prior art rather than bizarre legal theories about what a word means, and can only grow narrower with the passing of time, the concern that a patent will block important activities and impair rather than encourage innovation is greatly reduced. This may enable legislators and judges to abandon pointless distinctions about what the appropriate subject matter for a patent is. However, the importance of prior art should be emphasized, and therefore any subject matter where the accessible prior art is lacking - that is, where the majority of information is inaccessible due to e.g. copyright or secrecy restrictions -- should be considered inappropriate for patenting.

Well, that was all very profound, or at least profound-sounding, but I'm still acting like a mathematician, proving statements about the properties of an object without actually producing it. In the next post we'll take a crack at defining a Hamming distance for a real-life application, to gain some insight into the possibilities and problems of measuring the size of an idea.

In the last few posts, we've looked at ways to change how patent are examined and litigated -- by having people who know the field review them, and by actually testing their conformance to the standards they are supposed to meet. Now we're going to get even more extreme, and look at changes in the fundamental way the ownership of knowledge is defined and imposed. This isn't unprecedented; the existing system of specifications and claims has been elaborated over the last two centuries, and through the history of patent monopolies, various disparate methods have been used to allocate and enforce them.

As we explained a few posts back, today's patents consist of a specification that is supposed to describe and enable the practice of a novel and useful method or apparatus, and a set of claims that is supposed to describe what the inventors own in return for their disclosure. The claims are supposed to be clearly described in the specification, and define the precise boundaries of what is owned, so that a person practicing the art can read the claims and avoid infringement if they so choose. But in practice, claims are interpreted by judges with no knowledge of a field, aided by attorneys with large incentives for distortion and experts paid to serve the interests of their clients. As we previously noted, even legal experts admit that no one knows what claims mean until litigation occurs. Naturally, the existing system is very useful for attorneys, since obscurity and confusion are the servants of litigation -- but it is harmful for everyone else. And that leaves aside the fundamental question of whether it is possible to construct unambiguous boundaries in the very-high-dimensional space of concepts and ideas.

So let's abandon this silly system and try something different. We proceed by starting from the purpose of the patent system as stated in the US constitution: to promote the progress of science and the useful arts. Progress is promoted by change and exploration. People learn by copying, but must make changes to progress. So let's make our basic principle: no exact copies. If I make something new, I can prevent you from just tearing apart my version and doing exactly the same thing. (And the newer my thing is, the more credit I ought to get.) But unlike the current system, I can not prevent you from improving upon my new object to make yours - that's progress, and we're supposed to promote it. So how do we tell the difference?

To accomplish this end I'm going to play a mathematician's trick: I'll assert the existence of something useful without having it in hand just yet. Let's imagine we have found a way of defining a distance between ideas. (In a future post we'll look at a possible approach to making such a measurement, but be warned this is the hard part, and care must be taken to avoid being back where we started, with terms that don't have a useful meaning.) If we have this handy tool, we can then make a patent system that actually works. Here's how:

• A patent applicant writes a specification that describes what they have built. But instead of appending claims, they cite the closest previous work they know of - the closest prior art - and then propose a measure of the distance between what they have done and what the prior art did. This distance is a number, not someone's obscure arguments about what is substantial and what is not.

• The examiner either agrees, or cites prior art that s/he believes is closer to the applicant's work, using the same distance measure. Once the applicant and the examiner agree, the patent is granted.

• The applicant then owns everything that is closer to their specification than the nearest prior art, with the following exception: To litigate, the applicant needs to show that they can actually implement their invention, and that the implementation is closer to their specification than the prior art was. If you have to move farther away from the disclosure than the disclosure was from the prior art to make the invention work, the invention was not enabled. The patent is declared invalid.

• Defendents in litigation can narrow the scope of the patent by finding prior art that was closer than what the applicant and examiner found. An accused method or device that is found to be farther from the specification than the specification was from the prior art is not infringing, even if it contains everything in the specification. I can't just copy your stuff, but I can improve it by more than you had improved what already existed. Progress is promoted, not discouraged.

The claimless approach to patenting fundamentally changes the incentives provided by the system. Instead of filing when you don't know how to do something, you need to file when you do to have your patent be found valid. Non-practicing entities are intrinsically excluded: if you can't make it you can't patent it. The more new stuff you disclose, the more you own. The impact of most patents will quickly fade as a field progresses; only really revolutionary disclosures, with huge distinctions from the prior art, will retain value for the full formal term of the patent.

Most importantly, the claimless patent system abandons the false proposition that ideas exist in a linear progression, and ownership ought to be assigned to the first one. Invention is a net, not a chain. We hold each other up; we should own the strands we weave and not the ones we don't.

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

In the next posts we'll discuss a few details of implementation, and then take on the hard part of proposing a distance measure that can be at least reasonably unambiguous in the very complicated space of ideas. As my boss in the real world likes to say: go big or go home.

In the previous posts we discussed the possibility of requiring applicants to prove that practitioners of the relevant art can, aided by the application, solve a problem that they could not solve without it. What are the challenges in implementing such an enterprise, and what do we get in return?

Naturally the first complaint will come from applicants: we will have to pay for building the object we want a monopoly on? That's expensive! It takes a long time! It might not work!

Having been on the other side - developing products that take a lot of money and a long time, and might not work - for the last few decades, I don't have a lot of sympathy for this position. As we noted previously, only about 10% of US applications these days are owned by individuals, the remainder being assigned to corporations or other organizations. Corporate filers who are filing to protect their product line have the resources to build what they patent. Corporate filers who don't are probably Non-Practicing Entities (trolls), and should be expunged from the system in any case. Individual filers who don't have the resources to practice the invention they claim to have conceived can't demonstrate that they've actually enabled anyone else to do so. In effect, they are now required to find a partner with sufficient resources - which they needed to do anyway to get anything out of their patent, unless their strategy was to sue people who have independently developed what they claim to own, a reprehensible position society need not support.

A more serious objection is raised when one asks how to actually proceed with a test. Many real products require access to specialized resources, in some cases not easily reproduced. For example, an application dealing with semiconductor processing (the techniques used to make an integrated circuit wafer) may require modifications to a process step. Semiconductor fabrication factories ("fabs") cost hundreds of millions of dollars to set up and more millions to run, and do not make modifications to their processes for one customer, as this can put other customer's products at risk. An applicant who is an employee of such a fab may have the ability to test enablement - but they already know what the invention is and can't test obviousness. And if the test is conducted in the applicant's facility, they can provide numerous resources for enablement that are not disclosed in the application and not available to skilled people elsewhere, allowing them to appear to pass the test while not disclosing important information. Similarly, an algorithm might require specialized code libraries that are not publicly available; a drug might need unique catalysts or analytical capabilities.

But the problem here is more apparent than real. The false assumption that everyone ought to be granted monopolies, in the absence of any demonstration that they are deserved, is the difficulty. The result of imposing an empirical test requirement in the above circumstances is that filings cannot be made for ideas whose practicality cannot be demonstrated without an applicant's specialized resources. In business terms, we have just defined the situation where competitive advantage is mainly obtained from internally-developed or acquired resources not readily available to competitors or the public. Rights to exclude are irrelevant to the business and thus not needed by society. Typically such resources become available with the passing of time (and the competitive advantage of early vertically-integrated players is eroded). Empirical testing becomes possible just when it becomes necessary.

A more fundamental challenge is to ensure that both obviousness and enablement tests are not contaminated by the applicants. Testing should ideally proceed using resources that are independent from those available in the applicant's organization, and with no personal contact between the applicants or their representatives and the people performing the test. Further, the folks performing the tests should be able to act independently of their client's perceived interest. In the iterated-prisoner's-dilemma world of real business, where people deal with each other repeatedly, a firm may be tempted to always be stupid about invention and brilliant about enablement. The only way to avoid this result is to allow both applicants and competitors to fund the tests, and forbid their identity to be revealed. In this case, the performing person or firm doesn't know which side they are working for, and thus has no incentive to taint the results.

Another subtle issue is that in the event that the test is performed when the relevant patent application has been published, how are the testers to be prevented from accessing the application or patent during the obviousness test, while still having access to other relevant art? It seems necessary to implement a search engine that excludes the patent or application under examination, and any continuations [1] thereof. That's not very hard. However, the persons performing the test must agree to avoid the use of general-purpose patent search engines, so they don't get the patent before they are supposed to. That's harder: now we're dealing with human nature rather than code. A successful implementation will require procedures that do not impede research but also don't give away the answer, and the procedures may work better in some areas of art than in others. It's also possible the procedures will be different for e.g. a drug for human consumption and an embedded software program.

Finally, even if the testing is conducted in an ethical and thoughtful fashion, the results may be unintelligible, not just to a lay jury and judge but even to experts in the field. Engineers generally don't write well. That's why you probably need the people involved to provide testimony, so that the mysterious parts have some hope of explanation. (Not that a deposition is the best environment to get clarity - but it's what we've got.)

We can conclude that considerable thought, and some experimentation, will be needed to produce a useful implementation of empirical testing. And that's in addition to the need for legislation and international agreements. I told you it would get harder, not easier, as we progressed.

The great benefit of empirical testing is the stripping out of crap. In the absence of reference to experiment, human activities become the proverbial counting-angels-on-the-head-of-a-pin silliness. That's what the modern patent system is. Testing exposes plausible but false assertions in a system whose assumptions are today never challenged. Instead of unending silly arguments about what is obvious and what isn't, we have a factual examination of whether a bunch of folks came up with the "invention" after ten minutes of brainstorming, or worked for six months and never thought of anything like it. Instead of trivializing the hard part of innovation - getting an idea to actually work - we see if the supposed revelations of the specification produce a working model or a catastrophe. Litigation will still be an exercise in rationalization, but at least judge and jury will have anchors in reality they can use to constrain the fantasies of the competing parties.

The requirement for testing will also greatly reduce the number of granted patents that are likely to be the subjects of litigation. Most patents will simply be abandoned as not worth testing. Many others will be found obvious in testing, or fail enablement, causing their owners to abandon the patent, or at least to avoid litigating with it.

A secondary if narrower benefit is an opportunity for useful employment for people in any field in which patents are sought. As an engineer, I am, of course, lobbying for my folks here, though if the regime of applicability of patents continues to increase, other groups benefit, too. Why should the system only employ attorneys and patent agents? An empirical test regime would create an opportunity for people and consulting firms to make a living while (hopefully) adding value for society.

Finally, in making public the results of development work funded by would-be monopolists, we do what the system is supposed to do but often doesn't: we show folks exactly how to practice the supposed invention, and thus ensure that an invention that is enabled actually adds to our knowledge of how to do useful things.

And that's only the second-most insanely audacious proposal. More next time.

NOTES

1: A continuation is a patent application that uses the same specification as an earlier application, but has a different set of claims. A continuation-in-part adds new material to an earlier specification, and has additional claims.

In the previous post, we considered the proposal that every patent filing ought to be tested to see if it provides the two complementary benefits of being non-obvious and enabling. Let's see how such a requirement might work.

The idea is that every application to be examined shall be exposed to empirical testing to establish whether the supposed invention disclosed is actually novel, and whether the description is sufficient to enable it to be practiced. The empirical test uses the services of persons taken to be of ordinary skill in the relevant art. The number and type of persons selected may be chosen by the applicant, but it is a strict requirement that the same persons must participate in both the test of novelty and the test of enablement. (This prevents the use of persons of minimal skill and knowledge to examine novelty, and great expertise to demonstrate enablement.) The test may be performed during examination or after a patent is granted, but must precede any litigation involving a granted patent. (Note that the longer the applicant chooses to wait, the more risk there is that everyone will know about the supposed invention, generally by independent means rather than from the applicants. So it's actually to their advantage to do this early in the process.) The persons involved may be compensated by the applicant, but the compensation shall be arranged in advance, and provided at the completion of each stage as described below, without consideration of the contents of the reports in question. Participants should sign a statement declaring that they have no stake in the matter in question other than the agreed-upon compensation, and will diligently pursue both phases of the work with equal effort and attention.

In order to test novelty, the persons selected shall be presented with a statement of the problem to be solved and any constraints on the solution. It's not entirely trivial to figure out how to do this. Some patents have a nice background description and problem statement as part of the specification. Others jump right into their supposed invention, without even clearly saying what problem it is supposed to solve. A first cut is to have the relevant patent examiner provide a problem statement, based on the examiner's understanding of the specification, within a reasonable time after a request from the applicant. It may simply be extracted from the specification, if in the examiner's judgment the specification contains a clear problem statement that does not include the inventive method or apparatus. But in cases where that doesn't work, the examiner can use her or his own judgment about how to frame the problem without describing the applicants' solution.

In addition to the problem statement, the persons selected should be provided with any background discussion in the specification that describes relevant prior art, any prior art cited by the applicants or the examiner, and access to any other relevant art they require. Note that the access to art should NOT be constrained by the date of filing of the application, again in order to encourage the applicants to pursue an empirical test at the earliest possible date, since published art after the date of application may anticipate the described invention.

The test personnel shall be provided with whatever time and resources the applicant believes are needed to pursue a solution to the problem posed, with the proviso that whatever time and resources are provided for the examination of enablement must also be provided for the examination of novelty, and vice versa. This prevents the applicant from giving the team (say) 1 hour to find solutions to the problem, equipped with a slide rule and a notebook, and six months and millions of dollars of equipment to implement the described invention.

Once the specified initial phase is completed, the persons involved shall record their proposed and/or demonstrated solutions to the problem in question. They shall then be provided with the full patent specification, but NO other additional resources. Their responsibility is to practice the described invention, aided by the specification, and the same prior art and resources provided to solve the problem in the absence of the disclosure of the purported invention in the spec. The same people, the same calendar and labor time, and the same resources, shall be made available as for the initial examination of novelty.

A participant in the first stage may become unavailable for the second stage (due to personal reasons, business necessity, accident, etc.) but they shall not be replaced with any other person, though any notes, documents, or other work product they produced in the first stage may be used by the other participants in the second stage. A final report shall detail the extent to which the participants were able to practice the invention described in the specification. Any work product -- that is, any code they wrote, anything they built or modified, any other physical results of the project -- should also be preserved and available as evidence in the case of subsequent litigation.

The result of both stages shall be made available to the relevant examiner (if the patent application is still in prosecution at the completion of this exercise), and become part of the file wrapper in any case. The results of both stages must be completed before any litigation can be initiated based on a granted patent, and must be entered as evidence in any such litigation, and made available unedited to the jury in such litigation. The work product should be available for examination by representatives of the two sides in any litigation. The persons performing the two tests shall be obligated to provide (at least) deposition testimony, if requested by either party, for which they shall be compensated in at least the same fashion as for the work performed in the project, the cost of such compensation to be disclosed to the court and equally divided between the plaintiff and defendant.

In the next post we'll look at the various benefits and challenges of implementing this non-trivial change in the way patents are examined. But meanwhile I have to go see how the Giants are doing...