Erich Spangenberg has always been good at spotting opportunities for arbitrage, places in the market where there is a pricing spread that a savvy buyer can exploit for profit. He started in high-yield bonds and went on during the internet boom to turn a million dollars in patent acquisitions into a portfolio of software intellectual property worth $150 million.

By his own account, he didn’t make his IP fortune by not being aggressive. But Spangenberg describes his latest, and what he thinks will be the biggest and longest-lasting investing idea he has ever had, as something he stumbled into as a result of a friend convincing him to buy bitcoin.

“When I stumbled into it, I didn’t know it was enabled by the blockchain,” he recalled. “I made a lot in finance and what I recognized was how incredible inefficiencies create arbitrage opportunities. I realized blockchain was significant, and the more I thought about it, the more it seemed so. I jumped all over it.”

Spangenberg — who over the years has been on the receiving end of “troll” criticism for his patent fortune — recently pursued a more interesting path involving use of his patent knowledge to target what many of the populist-minded would judge to be an unfair stranglehold over pricing.

He teamed with hedge fund manager Kyle Bass to challenge the patents of pharmaceutical companies with high drug prices while shorting their stocks. But Spangenberg says he knew when he started with Bass that the drug patent effort wouldn’t hold his interest for the rest of his life. Spangenberg was also going through a divorce at the time. “I had time to think about a lot of things,” he said. Including exactly where he had made his millions and just how inefficient the patent world itself had become in allowing a few specialist niches to receive most of the riches.

“I started thinking about the space I had been screwing around in, the patents,” Spangenberg said. “There are roughly 200 national patent offices globally, and for 99.9 percent — even when we know who owns the patents — it is so hard to figure out anything. There’s not a whole lot of people positioned to say, ‘Here is where it’s screwed up.’ I played in arbitrage and the troll business, and inefficiency permitted me to make a fortune. We should eliminate that and now make money another way.”

His new firm, IPwe, is attacking two issues critical to what is really innovation’s own infrastructure: the intellectual property space. IPwe is creating a blockchain-based registry to tackle the lack of access to good information on patent transactions. And it is using machine-learning algorithms to better evaluate patent validity and worth. Specialized investments in the patent asset class exist today, but the end goal of IPwe is to turn the opaque and fragmented world of patents and patent transactions into a broad asset class that can attract much more interest from the world’s top institutional investors, as an example, insurance companies looking for pools of non-traditional investments that can generate significant returns.

In recent years the value of patents that have been monetized through acquisitions or licensing has neared $200 billion, and yet that represents a tiny portion of all patent grants. The World Intellectual Property Organization has estimated $180 billion in annual patent value derived from 2 percent of all patents. That’s an arbitrage opportunity that Spangenberg thinks will disrupt the intermediaries who today benefit the most from the opaque way the patent economy is run: primarily, patent lawyers. The end goal is to turn what has been a limited asset class, where a vast majority of profits go to a narrow group of specialists, into a major asset class receiving allocations from a broad set of investors.

Mark Schankerman, a professor at the London School of Economics, who first met Spangenberg after he donated some of his patent fortune to fund work in entrepreneurship at the London school, said patent trolls and their lawyers play a role in limiting the patent economy, especially in cases where they are demanding too much for IP — “the “holdups,” as he called them. But patent quality is also a huge issue.

“That’s the other big elephant in the room. … The vast majority of patents applied for are low quality,” Schankerman said. “A lot just should have never been patented, and no one wants the damn stuff.”

His research on U.S. patent litigation estimates that as much as 75 percent to 80 percent of patents are junk. The U.S. Patent and Trademark Office, meanwhile, is able to screen out, at best, one-third of the patent applications that shouldn’t get through. He said survey work on patent value, while less pessimistic than his own research, still shows as much as 40 percent of patents as being overvalued.

Even if 90 percent of patents are worthless, that leaves 10 percent to transact, already five times the current level that is in play, based on the World Intellectual Property data. He doesn’t think the math will be linear, but for the sake of an example, that’s $900 billion — 5 x $180 billion. “Even 2x would still be significant. Our bet is that this number is closer to 20 percent,” Spangenberg said.

He believes the critical problem is not on the large pool of overvalued patents on the output side — patent offices issue patents on what is applied for, but on the input side. “Over time the illusive patent quality will improve as a result of inventors, R&D and patent examiners having better tools. We will make more if there are more quality patents. We are capitalists at the end of the day — just creative on how we get there.”

Artificial intelligence algorithms can identify what is left in the world of intellectual property worth patenting. “We can tell you that, and that will make patents better,” Spangenberg said.

IPwe is working with the computer science team at the University of Minnesota, led by Professor George Karypis, which has previously provided patent evaluation technology to clients of IP Navigation Group, a previous company founded by Spangenberg.

“Erich has a grand vision,” Karypis said. “He always has one. We have the system in place to cover a large chunk of IP, both in terms of scale and analysis … on all U.S. and European patents and a good chunk of Chinese IP.”

The fundamental problems that algorithms can solve in the IP space relate to prior art search — how to know if a patent application is already covered by existing patents — and where “white space” exists in the patent system — how much room is left to patent an idea based on what has already been covered by existing grants.

“Existing methods of keyword searching and advanced levels of that work well enough if the prior art is obviously there,” Karypis said. “If you invented something and someone else does exactly the same thing, then prior art search succeeds.” It gets trickier when there is a combination of components from different patents, and when the terminology used is different or has changed in meaning over time.

Karipas said the underlying technology to evaluate patents has been validated at IP Navigation Group over the last four to five years but only on a case-by-case basis for clients. The end goal is to provide a public and unbiased way to rescore IP assets and discover where there really is opportunity for innovation. “We need to be able to figure out what people think about a particular technology. Forecasting the future is not a trivial exercise,” he said.

“If they can screen out bad patents, the application becomes more attractive to potential licensees,” Schankerman said. It also could help to make those filing for patents think twice before applying. “The value of his system is how much noise there is. I know there’s lots of bad patents out there.”

IPwe is far from alone in applying machine learning to IP.

“The traditional exam with a single examiner doesn’t work anymore,” said Pedram Sameni, president and CEO of Patexia, a crowd-based intellectual property company. “I believe in the patent system, but I believe the implementation should change substantially and it needs to embrace technology.”

The U.S. Patent and Trademark Office worked with artificial intelligence start-up AI Patents to implement its prior art search technology between July 2016 and December 2017. “We have provided substantial evidence that A.I. techniques can be applied effectively to identify similar ideas even if the ideas are described using distinct text,” Liat Belinson, founder of AI Patents, wrote by email. But she said the project is on hold and future plans are unclear as a result of the change in USPTO leadership after the presidential election.

A USPTO spokeswoman said, “The Office continues to investigate uses of emerging technology for various applications.”

There are also IP research firms that run investment portfolios based on algorithmic analysis of patent portfolios at publicly traded companies, including CNBC’s partner on the IQ 100 Index, MCAM-International.

“We are not taking a position on who has the best patent analytic tools,” Spangenberg said. “Our tool is free (it always will be) and we will continue to spend significant R&D dollars on improving our tools and hope to enlist the IP community to make the tool better.”

When I first came across IPwe and Spangenberg’s own initial, broad brushstroke blog post about the “misfit trolls, geeks and wonks” getting into the blockchain, I assumed he was amassing patents in the space to repeat his software strategy — and so did some patent experts whom I consulted. I was wrong. While firms that have a history of monetizing patents they didn’t create are in fact major players in cryptocurrencies and the blockchain — Intellectual Ventures, which has had its fair share of critics over the years, is among the top five holders of bitcoin patents, according to Patexia research — Spangenberg views IPwe’s own patents in this space as a secondary aim.

“We have patents filed but it will be years before they are even issued and that’s not how I’m planning to make money. This will be a wildly profitable business,” Spangenberg said. But the profit potential is tied to a long-term bet on fundamental change in the patent economy. IPwe will make money by charging a fee — in basis points — for use of its technology and IP transactions that end up occurring on it.

IPwe’s research on the ownership side of patent records shows it to be no better at being accurate. As much as 30 to 40 percent of patent ownership records are currently not identifiable or misidentified. Putting licensing and acquisition transactions into a private blockchain registry — an effort that Spangenberg said will require chief financial officers to see the value in sharing deals details and as a result may take time to be embraced — will ultimately provide all buyers and sellers in the IP market with a better infrastructure for dealmaking.

There are multiple blockchain-based efforts to manage intellectual property rights, including a surprise announcement this week from Kodak about a blockchain system for photography.

“Right now patent transactions are primarily bilateral, and that almost certainly is high cost and non-transparent,” Schankerman said. “There might be other buyers out there who haven’t been contacted. It’s not a competitive market with transparent prices.”

“Transaction history is an important signal,” Karypis said. “You have an intelligent party making a rational decision to do something.”

Once the registry is complete, IPwe will consider giving it away to governments, among others, to get economies of scale. “Someone will want it, some consortium. Our model is that as more transactions occur in the space, we will grab a piece of the transaction value. We have to take the $180 billion occurring today and get it up significantly.”

IPwe plans to launch its platform on February 1.

How long it will take, if efforts like IPwe are successful, is a forecast few experts want to make, especially with slow-moving governments reluctant to make major changes. But Spangenberg seems to be in it for the long haul.

“I am sure ’20 years’ is what ‘patent people’ say. Patent people tend to be conservative and linear in their thinking. Those are extremely valuable attributes for some applications, but not for unlocking the value of an asset class that has been poorly managed for over 100 years,” he said.

“Maybe he is getting more mellow,” said Schankerman, who has has spent a lot of time discussing his research on reforms needed in the patent system with Spangenberg. “He obviously still wants to make money and is good at it, but he wants to improve the system. I think he does care about that.”

The London School of Economics professor said true reform in the patent world won’t come from technology alone. It will also require policy changes to lower the cost of litigation. He said the most effective way to blunt trolling is to make it less costly to see patent litigation through to court so companies will not feel pressed to settle claims quickly. In addition, there needs to be a way to make it more costly for those who file weak patents to maintain them without incurring costs. “He is not pushing that yet, the front end policies, and we need to push on two fronts,” Schankerman said.

Schankerman said that Spangenberg is asking the right questions: What is happening with all of these patents, and is the patent system generating bang for its buck? IPwe and other efforts to apply technology to patents may be able to democratize the patent trade. “Right now it’s hard because we don’t have the information.”

“It’s not that I’m getting more mellow,” Spangenberg said. “My mind is expanding. This is a much bigger opportunity than being a patent troll.”

“As opposed to bitching, I wanted to do something. We can come come up with a better system. There’s a fortune in the fortress of patents, and every now and then someone jumps in and makes money. I want to be able to talk to any asset manager and say, ‘If you understand the patent asset class better, would you deploy capital?’ We can make it possible for a normal human being to understand what patents are doing.”