Mark Montgomery is the founder, CEO, and Chairman of the Board of KYield, a pioneer in AI systems with a special focus at the confluence of human and machine intelligence. I had a chance o talk to Mark who is the originator of the theorem ‘yield management of knowledge’, and inventor of the now patented AI system that serves as the foundation for the KYield OS. Mark began training in software development in 1996 and network engineering in 1997, becoming proficient in object-oriented programming and several languages then progressed to more advanced analytics, search, and artificial intelligence. Mark was a frequent visitor and invited guest at the Santa Fe Institute from 2009-2016.
You can read the complete interview below.
In your recent presentation at the ExperienceITNM technology conference in Albuquerque, you spoke about the importance of adopting large, unified AI systems like KYield OS that can achieve higher results, compared to small AI projects. Tell us about KYield OS. Why is it called ‘an advanced distributed AI operating system’ and ‘Yield management of knowledge’?
Several of the most important and valuable functionalities in the KYield OS can only be executed with an enterprise-wide or network-wide system. Examples include behavioral cybersecurity, prevention of human-caused or enabled crises, insider risk, enhanced productivity, and continuous learning across the organization, including individuals, teams, and groups. Organizations must have a unified systems approach to optimize these and other functions.
I certainly didn’t plan it that way as it’s a much more difficult, expensive, and long journey for a founder and entrepreneur than simply taking small projects. However, we followed the evidence from inception, and this is where it took us.
We call the KYield OS, an advanced distributed AI OS for technical accuracy. The CKO (Chief Knowledge Officer) Engine is essentially a semi-automated bot that allows senior management with high-security clearance to administer the settings for governance, security, access, and curriculum, among others. The original system is a modular patented AI system architecture, so, for example, the CKO Engine can be extended to partner organizations, and the individual modules can be extended to individuals in partner organizations, citizens, customers, etc. Much like the Internet construct, the KYield OS modular system can be scaled infinitely with proper engineering.
The advanced description is really for two reasons. First is that the technology within the system is state-of-the-art. We employ what I call data physics to deal with entropy, compression, security, and encryption. This is the primary reason for the long voyage in R&D (23 years). When the system was still primitive and young I communicated with many leading scientists to deal with these issues—one example is Vint Cerf. Vint is the one who found the misspelled word in the patent, so I left it there as a reminder—we had several PhDs in addition to myself review prior to, but no one caught it. Scalability was the primary obstacle in the early days.
The second reason the word advanced applies is for the competitive power for organizations in the KYield OS. As you could see in the video of the presentation at the ExperienceITNM conference, the goal of the system is to achieve a CALO or a Continuously Adaptive Learning Organization. Productivity growth can be exponential. I shared a chart showing the difference between three organizations over a ten year period, reflecting the outcomes from three different approaches to adopting AI systems. The organizations all start with a $100 million investment. The lower-performing organization declined over the ten year period, ending with $86 million. The organization that achieved the inflation target of 2% per year in productivity growth remained neutral in real dollar terms, ending the decade with $100 million more, or about $186 million, and the third organization ended with $426 million. The productivity performance reflects the knowledge yield curve and execution thereof, which is driven by AI systems.
Yield management of knowledge is the underlying mathematical theorem for the KYield OS. I first developed it conceptually in the late 1990s in my small knowledge systems lab and incubator, then located on our property in Northern Arizona. We were operating multiple networks at the time. My brother was diagnosed with ALS, and I could see the eventual potential in using data techniques and algorithmics to accelerate R&D. It would take a long time and couldn’t help in his case, but that was the original inspiration. A few years ago I finally matured the math in the theorem, though I haven’t shared it yet.
Some large companies, especially banks, mostly because of security and compliance, try to build their own AI systems and infrastructure all internally from scratch. And, we have countries like China, building everything for their own and not necessarily worrying about what happening in the rest of the world. What are the pros and cons of building big AI systems from scratch?
Yes, there are pros and cons for certain, most of which are obvious, some of which are not so obvious, particularly to chief decision-makers who may not have the technical chops to be able to recognize conflicted advice, whether from internal or external sources.
The primary benefit of building a large system from scratch internally is that the organization doing so understands every aspect of the system, assuming they document it well so that churn in key employees doesn’t leave them in a vulnerable position. Very few organizations are in the position to do so, but theoretically, if a large enterprise has the talent and budget, they could potentially compete with the best of big-tech and emerging companies. In practice, that is primarily theoretical to date. Many influencing factors contribute to the actual outcomes, including core competency, legacy architecture, culture, and talent.
It’s not necessarily due to security or compliance, though often cited as the reason. For example, one of the primary risks in building internally is conflicts of interest for insiders. In large and systemic human-caused crises, it is not unusual for senior engineers to be involved in the cause or ensuing cover-up. From an enterprise risk management or regulatory oversight perspective, I often make the case with explicit examples of why especially large organizations should not build internally. You don’t want to empower bad actors, whether internal or external. So risk management is one reason to license or buy our KYield OS. For example, in our HumCat program (Prevention of human-caused crises), we provide the option for independent oversight tied to financing or bonding. This is a pragmatic solution to real-world problems we’ve encountered. This method shares limited data to underwriters, regulators, or other stakeholders that reflects a much more accurate risk profile. Critically important when dealing with large or system risk.
The biggest issue for most is, of course, cost. The average amount of redundancy in large AI systems today is running about 40-50% and climbing. This is similar to the office environment in the early 1980s, and as everyone knows in my generation, MS won that battle. It makes no sense whatsoever for every organization to attempt to build their own custom system as the redundancy grows. The majority of issues in organizations are universal. Very few if any have been studying these issues as long or intensely as we have. It’s frankly foolish not to tap our R&D.
Another issue to consider is the IP. Although it hasn’t yet manifested in the courts much, we’ve seen an explosion of IP in AI systems over the past few years. Most of it, including our own, are defensive in nature to protect our own R&D. However, if organizations knowingly infringe on the IP of others, they are taking enormous operational and financial risks. As we can all see, IP is one of the factors driving the trade war between China and the US today, reflecting how important it is.
Frankly, most organizations that are building internally are influenced by turf battles. Although it has improved in recent years, there is still a lot of fear of being replaced by AI, including professionals and even senior IT management. I don’t believe that fear is rational as I see little evidence to support it. Indeed, we make the case that their risk of being replaced drops sharply by adopting the KYIeld OS as we seek to enhance the competitiveness and value of individuals and organizations, not replace them. However, in dealing with the largest corporations and government entities, it appears that turf battles are a key factor in most decisions surrounding AI.
As I understand, AI systems are the most critical and valuable technology of our era. What are the common challenges that you confront while dealing with organizations?
The challenges have changed over the years. Two decades ago, the challenge was in basic science and whether the theorem was even viable. We’ve since won the scientific argument. A decade ago it was still “what is AI?” in the board rooms. Many of the large and sophisticated organizations had some competency internally, but it was restricted to one or two functions—drug discovery in pharma for example, or anti-terrorism in defense. No one had ever considered employing across an entire network or organization as we do. I believe the way organizations adopt and with whom will determine who wins and loses.
Today our biggest challenge is that the KYield OS is a large system, and it’s not yet in a turnkey format, so we still need to do more education than a demonstration. In dealing with board rooms that don’t have the technical depth to understand all these issues, we still spend more time educating than showing. It’s impossible when running large organizations to spend sufficient time studying all of the areas we cover by necessity, so understanding is a big problem still. Many costly errors continue to be made in the adoption of AI. Some have cost management their jobs already. Multi-billion dollar losses have occurred that likely would have been avoided with more competence on the board.
Another challenge we face is that we are simply a small company. Even though we have a very low-risk profile intentionally, self-funded entire voyage, and no debt, some large organizations wrongly believe their risk is lower going with a large well-known company. That is not necessarily the case at all.
We can boil today’s most AI innovations down to two concepts: machine learning and deep learning. These terms are often used interchangeably, though they have many differences. In your presentation, you support the use of machine learning rather than deep learning, saying ML is transparent. Can you elaborate on this?
Although backpropagation most often used in deep learning was a brilliant discovery that can do some things very well, it does come with challenges, one of which is that it is not transparent even to AI researchers. We can demonstrate through trial and error what the DL algorithm can achieve but not necessarily precisely how it was achieved, so that creates the so-called black box problem. It’s also one of the reasons we have a talent war in AI — very few of the top researchers have proven that they can move the science forward. A lot of money is invested now in overcoming the transparency challenge with DL, but it’s much easier to be transparent with ML and well-structured data than DL. The data structure is critically important for most of the functions in the KYield OS, as is the more advanced technology we employ to manage the data.
Another reason we use ML over DL when possible is cost. The financial, computing and energy costs are much higher with DL than more traditional analytics and ML. That’s why my criteria are to employ robust data physics with integrity to perform as much as possible with ML. However, quite a few critical functions where DL is required, so we employ DL where necessary.
Tell us about your new invention — the Synthetic Genius Machine and Knowledge Creation System, that captures and synthesizes genius features from the published works of great individuals in the past like Leonardo da Vince and then creates new knowledge. How does it enhance KYield OS and the operations of companies in the current market?
The synthetic genius machine is a new invention – I just filed the patent application about a month ago. Some of the technology involved goes back a long way. This system brings it all together in one refined AI system that can be deployed independently or integrated with our KYield OS.
By focusing on the work of proven masters, we accomplish several things simultaneously. First is that we are tapping into the most powerful supercomputers to have ever existed in the human brains of masters, and then we do some high-level modeling of their work. These contrasts say with the boil, the ocean big data approach that attempts to process all available information. So the synthetic genius machine is first and foremost a more pragmatic direct route to superintelligence with proven sources. We then combine that with our proprietary technology that includes synthesizing the genius features, or components, and converts to the symbolic representation of genius features. This serves a dual purpose by providing a form of encryption and significant data compression. For example, a single symbol may represent a large mathematical equation, a series of processes, or a mechanical drawing. The combination is quite powerful in distributed systems.
It’s important to understand that while we do focus on the masters, it’s not limited to them. We also include contemporary experts in individual disciplines and perform cross-disciplinary modeling, so it can be as broad and deep as required of the specific query. An example might be in modern surgery techniques, accelerating R&D for nuclear fusion, drug discovery, or space travel. The potential of the synthetic genius machine is only limited by the number and type of genius features and the ingenuity of the people constructing the queries.