From theorem to market through multiple valleys of death and beyond
This is a personal story about our real-world experience, which contains little resemblance to most of what is written about entrepreneurism and technology commercialization. While our journey has been longer than most, scientific commercialization (aka deep tech) typically requires two decades or more from theory to market. Even more rare in our case is that the R&D journey has been self-funded and very lean. Although my route was different, my peers in R&D have been scientists in a handful of labs—primarily universities, a few corporations, non-profit institutes and national labs, this model has allowed us to develop one of very few unified AI systems in pure native form free from institutional and other conflicts that too-often kill or ruin much-needed technology and companies based on them.
I’ll begin in the Puget Sound area where my wife Betsy and I met in 1980 while working at Mt. Rainier. A couple years later we started a traditional business. After selling our business Betsy went into banking and I started a consulting firm that worked with a variety of different clients across the Pacific Northwest. We moved to Arizona in 1992 in part due to consulting work cleaning up the S&L crisis for private owners. In 1995 we decided to test the emerging web with a self-guided management system that was distributed in hard copy. That effort became one of the leading networks for small business. It was a first so we experimented with all models. The venture grew rapidly in organic form but needed significant growth capital to reach sustainable maturity. Unlike San Francisco and the Seattle area where nearly every good scalable business was funded, flyover states had little infrastructure to support scalable businesses, even when risk had been mitigated in sustainable form, so we sold prematurely.
The lean KS lab
The experience with our first online venture was so intense with such profound implications that I converted our consulting firm to one of the original lean venture labs. I retrained in computer programming and built a lab and data center in the building above on our property in Northern Arizona. Our specialty was in knowledge systems (KS — arm of artificial intelligence / AI). Stanford has a well-known KS lab—one of few at the time. Although hundreds of billions USD were wasted on me-too dotcoms in the 1990s, AI was still in an ice age (aka AI winter).
Two decades ago this year I was working in the lab operating a learning network I designed called GWIN (Global Web Interactive Network), which was the most advanced of several experiments we developed from scratch. Primitive by today’s standards, GWIN was a cutting edge network at the time that attracted an impressive membership of leaders in science, business, NGOs and government. Tech CEOs and VCs were among our closest followers, though we had entire boards of in the Fortune 500, intelligence agencies, and hundreds each of professors, investment houses, analysts, NGOs, and editors. Log activity from Air Force One was not uncommon. A nun reporting from the Amazon jungle was one our most interesting members.
The most promising program in GWIN was ‘Lookout’, which was a primitive early digital assistant that delivered personalized news clips sourced from the web containing brief human analysis accompanied by discussion. Although we offered web discussion and chat, email lists were preferred at the time.
GWIN was a fascinating experience that was also producing enormous value. One of many examples was a network-wide warning on hurricane Mitch—second most deadly Atlantic hurricane. A life-long weather geek, I typically had a monitor running radar and sat loops so I watched as Mitch grew into a dangerous slow moving cat 5 heading right for high risk areas, so I issued a warning. Between a few members in Central America, media, corporate and government members with operations in the region, our warning on Mitch spread rapidly. I can still recall the satisfaction in receiving messages from members conveying that by distributing a few lines of text in GWIN we helped save lives and prevent unnecessary losses. Prevention has been one of my personal passions all along. When planned and executed well, prevention can provide the highest possible ROI—in dollars and lives.
Most of the GWIN members didn’t realize that although a team of remote developers helped build the network, I was operating and improving it solo 24/7/365 from my office in our onsite data center. My wife Betsy and I were paying for almost all of the efforts personally other than a small investment at the time from my late partner and friend Dr. Russell E. Borland. By that time a tsunami of capital had arrived in Silicon Valley and Wall Street causing the infamous dotcom bubble, resulting in enormous levels of predation, subsidies, and losses, much of which I considered fraud. Few would pay for online services because of it. It was the largest consumer price war in history so we focused our efforts on deep tech and business rather than consumer.
A new theorem
A few months after the launch of GWIN I received a life-changing call from my brother Brett telling me that he had been diagnosed with ALS. I then dedicated as much time as possible attempting to find promising therapies or tools that could accelerate R&D. Tragically, I discovered that we were a long way from even understanding ALS or obtaining technology that could significantly accelerate effective therapies. Brett passed away three years later within a few days of the estimate by doctors at Mayo Clinic in Scottsdale who confirmed his diagnosis. My quest to find, test and develop more intelligent tools led to a new theorem ‘yield management of knowledge’, which was then followed by piecing together components of a unified AI system in our Kyield OS.
The pathway to the theorem began with a classic aha moment after an extended period of intense work on information overload in operations and research, including testing promising search engines and other methods as they became published. I’m still refining the equation, but it essentially details key factors in optimizing the knowledge yield curve given the needs and constraints of each entity. Although the human brain is amazingly powerful, it does have finite limits beyond which it begins to malfunction, which I first discovered at 30-something in the lab. We were clearly faced with a highly complex systemic problem requiring a systemic solution with the capacity to effectively manage the complexities involved. To help clarify I posed the following question:
If a hypothetical perfect Chief Knowledge Officer (CKO) existed, how would we optimally achieve his/her mission in a network environment, how would it be designed, and what essential components would be required?
That question eventually led to our CKO Engine, which provides governance and security for the entire distributed network. Administration in the Kyield OS is through a simple natural language interface with multiple security levels and methods, some of which are kept secret for security.
It was discovered that multiple obstacles could best be overcome within a single holistic architecture; and without which none of the problems can be fully overcome:
-
If we do not resolve the problem of information overload, then creativity and productivity suffer.
-
If we do not resolve the problem of ownership of original work, then innovation suffers.
-
If we do not provide accurate metrics, then meritocracy cannot function properly.
-
If we do not provide adaptability, then differentiation and continual improvement will be very difficult to achieve.
-
If we do not embed intelligence into the files, the most relevant search queries cannot be returned even by the most improved algorithms, thus negatively impacting productivity and innovation.[i]
We realized that it would be at least a decade before essential components matured sufficiently to begin to effectively manage knowledge yield over computer networks. A continuation of Moore’s Law in semiconductors in combination with rapid improvements in bandwidth and algorithmics would be required over an extended period before the theorem could be fully realized in applied form as intended. However, I was confident it would be achievable in my lifetime, even if imperfect.
We were able to test components of the standard system and verify supercomputing results of similar scale and data structure in early 2000s, but scale challenges and bandwidth bottlenecks prevented the ability to deliver functionality to individuals and devices. By the mid-2000s Kyield had matured into a distributed operating system (hence Kyield OS) and essential pieces of the puzzle began to coalesce, so I submitted my AI systems patent application “Modular system for optimizing knowledge yield in the digital workplace.” The 2006 application was granted in 2011 representing about 25% of the total IP/IC at the time.[ii] I viewed the patent as additional insurance.
Initium Capital
In late 2007 I met with Craig Barrett at his office in Chandler Arizona. Although Craig and I were both active with local universities and tech groups in Arizona we had never met, so a mutual friend Les Vadasz introduced us. I won’t go into detail on what we discussed in our one-on-one meeting other than to say it was open, honest, and friendly. Craig may have been approaching mandatory retirement age but he was impressive, helpful, and obviously still at peak performance. A few years earlier I had spearheaded a VC firm (Initium). It had proven suicidal to build high cap ventures in flyover states that depend on capital centers for growth funding. In addition to rare private efforts like our small lab, universities, federal and state governments were investing enormous sums in R&D just to see ventures copied or cherry picked primarily by California (more recently China). In New Mexico most of the spinout ventures from national labs were exported, perpetuating a long-term trend in one of the worst state economies. I warned often of an economic balkanization underway. Few seemed to understand that if that wasn’t fixed most other problems would be trivial.
Our efforts to build Initium hit a similar capital ceiling as individual ventures in the form of lack of regional support. We had one of the strongest teams ever assembled in a flyover state with an unusually large inaugural fund target of $250 million. The fund structure contained a flexible 40% dedicated to the region and 60% with no geographic restrictions. While we earned a place on the emerging leader radar, history had painfully demonstrated the need for key local support and investment. To the extent such regional investment existed it was rare, too risk averse for deep tech and/or unqualified. So we reluctantly sized down Initium and explored merger interest from Bay area firms. Betsy and I liquidated everything but our property and relocated to Half Moon Bay during the first week of 2008, just in time for the financial crisis.
We enjoyed many aspects of living in the Bay area, not least living a block from the ocean after 15 years in the desert, though we found the economic situation troubling. Home prices were several times the cost of where we lived in Arizona and all other costs were much higher as well. It was quite clear why VC investment was so high in SV, contributing to sharply increasing failure rates. The number of homeless served as a constant reminder of just how out of whack the local economy was. Betsy took her first year off work to pursue a hobby in art and wound up working for non-profits as a volunteer attempting to fill some of the massive unmet social needs.
We had a one-year window during which time the financial crisis became increasingly worse and the future of the other firms and investors increasingly uncertain. We were also in discussions with market leaders for OEM-type relationships, but they were clearly not yet prepared for AI systems or Kyield. So after the most costly year of our financial lives other than not investing in pre IPO Microsoft or pre investment in Google (among others), we walked away from a merger that teed up a significant investment in Kyield. Hindsight suggests that our instincts were functioning well as Kyield and the markets were still premature a few years later. It’s unlikely that Kyield would have survived in the SV VC model at the time. Machine learning really took off in 2015 with investment in the tech stack that improved performance and value for majority of use case scenarios.
The city different in the land of (serendipitous) enchantment
Upon arrival at our property back in Arizona in early 2009 we discovered that the caretakers had trashed our property, so we took another financial hit and turned it over to a management company. We then decided to go on a road trip to find a rational place to ride out the financial crisis while maturing Kyield R&D. The plan was to do a loop starting in Tucson, then through New Mexico (NM) to Colorado, perhaps Wyoming and Montana and back through Utah to Arizona. My expectation was to lease a place in Colorado, but fate intervened in the form of a car pulling a u-turn right in front of us outside of Albuquerque on the way to meet a realtor for a house showing. The ensuing collision almost totaled both cars but no one was injured and the driver was very nice as were the police. We were on a schedule, however, so had to rent a car and move on to Santa Fe where the first house we looked at seemed perfect for us and our dogs, so we took it.
We have history in Santa Fe dating back to our first trip in 1985 and also an informal relationship with the Santa Fe Institute (SFI) from our GWIN days that share many others. I also had some interaction with national labs due to Initium. We performed consulting work in NM that included market audits in the 1990s and also covered in VC, so I was familiar with the strengths and weaknesses. One of the world’s leading research centers—more so than most realize, NM is also famously difficult for growing scalable businesses of the type that occasionally emerge from that investment. Despite hundreds of billions of dollars invested in research within NM and large numbers of spinouts, the state has never produced a significant business success in tech. Suffice to say that accidents normally occur with far more frequency.
I spent quite a bit of time at SFI over the next several years meeting with leading scientists from around the world working on similarly challenging problems in physics, computer science, biology, economics and sociology, which helped indirectly in ways difficult to capture or fully understand. SFI is unique in the world in many respects.
In early 2012 we began presenting Kyield to management in the few organizations that had a supercomputer, sufficient budget and the internal talent to even consider Kyield in organic fashion at the time. Significant progress has since allowed us to steadily expand our focus to mid-market and government markets. When the managed services model is completed as originally intended most markets should be viable.
Byproducts of the voyage (not including R&D pipeline)
IoE (Internet of Entities)
Since the early days of our R&D I have looked at networks as being comprised of entites, not things. The reasons should be self-evident—to the degree they aren’t speaks to the influence on structural issues in the network economy we are working to resolve, some of which are causing serious economic and social damage—namely the business models applied to the web.
Our old colleagues who designed the Internet are the first to admit that it was never designed for many of the tasks required of it today, including commerce or security. Public networks involve many different legal entities, including individual humans and organizations, each of which has unique needs and legal rights. The data carried over networks represents those rights (or should). Even sensors on the network are owned and governed by entities, and they are rapidly becoming more intelligent, hence the need to view networks as entities that contain appropriately engineered governance structure to manage relationships between entities.
Today we offer a suite of IoE options built upon the Kyield OS to manage an enterprise network easily extendable to partners, customers and things (sensors). This is the wisest path from my perspective for managing networks in government, industries, homes, autos, ships, planes, etc. The Kyield OS offers critical elements for optimizing intelligent networks.
The standard Kyield OS
CALO (Continuously Adaptive Learning Organization) is the manifestation of the original modular system invention as applied with state-of-the-art components and algorithms. Recent improvements in machine learning combined with more sophisticated statistical processes and algorithmics within the distributed Kyield OS enable customer organizations to achieve a CALO. The Kyield OS operates substantially in the background with semi-automated controls for each organization, group and individual. Unlike earlier management concepts, CALO is executable.
Health Management Platform
First unveiled in our diabetes use case scenario paper in 2010 still in futuristic form, which has since been downloaded in the seven figures, the U.S. has yet to deal with the healthcare fiscal time bomb. The sector has evolved over decades to build resistance to efficiencies, cost management and/or patient-centric systems, resulting in the highest cost healthcare system in the world, which provides less quality than others at half the cost. Little progress can be made in U.S. healthcare until reformed by Congress, without which we are limited primarily to the self-insured in the U.S.
HumCat (Prevention of Human Caused Catastrophes)
After many years of focused R&D we announced our HumCat program powered by the Kyield OS. The HumCat program pioneers new territory at the confluence of distributed AI systems, risk mitigation and prevention. By bundling more powerful computing and algorithmics in the Kyield OS with financial incentives and risk transfer through bonds, reinsurance, and other vehicles, we can significantly improve the risk profiles of individuals and organizations and thus lower costs.
It is now possible to prevent many if not most human-caused crises, including accidents, fraud and/or malintent, whether in physical or cyber form. While each organization has unique characteristics requiring bespoke structuring, it is possible to offer select clients limited upfront guarantees that finance and cover the cost of the entire program over a defined period (1-5 years). Higher risk organizations can likely reduce costs significantly and may be able to improve ratings over time as reduced risk is demonstrated with more accurate analytics offered by the Kyield OS. As interest rates rise ratings will become even more critical for corporations and governments.
The HumCat program targets the highest possible ROI events while bundling the individual functions in the Kyield OS such as enhanced security and productivity, representing a significant breakthrough in value to clients and society. We have a great deal of interest in the HumCat program for what are hopefully obvious reasons.
Knowledge Currency
A byproduct of the architecture necessary to execute functionality within the Kyield OS is deep intelligence on workflow and work products from each entity. While Kyield makes no claim on the data ownership or control beyond required by law and as pre-agreed with customers for specific needs, that intelligence does allow us to create and manage an exceptionally valuable digital currency, or knowledge currency. The creation and offering of Kyield’s knowledge currency (KYC) opens up many positive benefits for and between customers, including more accurate valuation of individual, team, and corporate knowledge capital, the ability to be compensated fairly for knowledge work, and the ability to transact and trade intellectual work products in a more rational and accurate manner. In addition to knowledge products created, KYC can be used to value and transact knowledge about an entity, such as health information. At large scale the KYC could have profound economic benefits by substantially overcoming the serious problems across our society caused by the dominant web ad model. KYC has been in our R&D pipeline since the early 2000s.
Where is Kyield today?
We are in discussions and negotiations on various options to build out and scale the Kyield OS in the hybrid managed services model as originally intended. While the system can be installed on top of the infrastructure of others such as AWS, Azure, Google, IBM, and Oracle, we have some proprietary technology that must be installed on our own hardware for an optimal unified Kyield OS. The hybrid configuration typically includes an installed custom computer within the client data center, private cloud or a multi-cloud scenario. This allows us to offer the pre-engineered Kyield OS and additional products while protecting our security as well as customers, reduce unnecessary and costly integration costs, and reduce or eliminate redundancies.
Happy Thanksgiving 2017!
Mark Montgomery
Five related articles
[i] Montgomery M “Unleash the Innovation Within” Kyield, November 2008
URL: http://www.kyield.com/images/Unleash_the_Innovation_Within_-_A_Kyield_White_Paper.pdf
[ii] Montgomery, Mark. “Modular system for optimizing knowledge yield in the digital workplace.” US Patent 800577823 August 2011. http://www.google.com/patents/US8005778
One thought on “A 20 year trek in AI systems”
Comments are closed.