Learning Alignment: An AI to Amplify and Extend Human Intelligence
The concept of AI technology that enhances our daily lives is not new. We have reaped the benefits of AI, starting with early successes in fraud detection to recent advances in interacting with machines in natural language. The pattern is clear, expanding the frontier of automation afforded by AI leads to improved living standards, massive job creation, and wealth generation. But, where would we be without an AI that organizes information (Google search)? AI-based search feeds our human need to know. AI has facilitated our desire to connect and be recognized (Facebook). Yet, the same technology creates information silos that divide us. An AI trained to optimize clicks to affirm our silos results in fiery fragmentation. Without noticing it, we took the highway to hell.
Identity is complex. When in-person, we consciously and unconsciously relate to others on multiple levels. The person who may disagree with us politically may have other interests that create a bond, bridging our differences. Organizational consultants understand this, which often inspires a wide variety of events to discover and expand the depth of relationships to clear the way for building focus and strength in strategic planning. In a world of virtual connections, can we boost the dimensions of interaction and collaboration?
This article applies AI to learning the emergent ideas and beliefs that bring groups together. We conduct business and live in a highly complex adaptive ecosystem. Over 20 years ago, Gladwell’s Tipping Point, How Little Things Make a Big Difference, made it clear; discovery of emergent patterns of behavior have substantial business and social significance. For that reason, we must apply the best technologies that match the challenge of the situation. Learning how little things get big in business and in social movements in a way that embraces the power of diverse ideas is foundational to our path forward.
The messy, complex entangled world we live in today is connected, and changes are fast and frequent. The unmet need is an AI that discovers and empowers shared ideas. There is a need for an AI that builds our collective intelligence based on shared knowledge and shared imagination. From a business perspective, this allows learning preferences in products and services. From a civic government perspective, this allows collaborative shared ideas on solving problems. The same technology that can learn alignment for social good works to find the next opportunity for constructive market disruption. The new fundamental unit of economic development is not the individual. It is the complex adaptive network of people, products, services, policies, investors, and companies. The new reality is a complex ecosystem with emergent, intricate behavior patterns. Before there was an AI solution for this, I was CEO of a company, Informative, that demonstrated business value in learning how little things get big.
One case study from Informative was Intuit Turbo Tax. The task was to determine what would turn passives into promoters for Turbo Tax (Net Promoter Score). The process engaged customers in an open-ended conversation, and the software learned customers’ collective desires to enhance the customer experience. The software revealed alignment on what the customers wanted based on a bottom-up process of alignment discovery. We learned the emergent importance of suggestions by customers prioritizing where they agreed with other customers. The result was a list of 3 or 4 action items representing most of the customer base. Rather than comb through hundreds of pages of research documents, the TurboTax GM aligned his internal team on the 3 or 4 issues and acted. The result was a 17 point increase in NPS year over year. Intuit is one of the case studies in “The Ultimate Question” by Fred Reichheld.
Learning alignment enables focused action. Unfortunately, the traditional tools used for learning needs and preferences fail in the climate we find ourselves in today. Interviews, reports, focus groups, and surveys fail to address a fundamental sea change precipitated and intensified by the internet and e-commerce. Connections between people and organizations create a dynamic, complex adaptive system. A person’s intention to purchase a product or service can change in a New York minute. A company’s supply chain strategy becomes obsolete overnight because of a pandemic.
The complexity and dynamics that we observe today signal a new normal. It is bursting with potential because connectivity and dynamics can benefit the agile, fast-moving company or organization that quickly learns new market opportunities and meets them with speed and agility.
Learning alignment is as complex as understanding the movements of a murmuration of starlings. Don’t be discouraged! It is not impossible, and we can put AI to work for our benefit. Emergent and beautiful behavior out of what appears to be chaos has a robust mathematical foundation.
The most significant advances in AI result from learning alignment of patterns in data. Deep learning has its roots in the mathematics of alignment. The mathematical underpinning of deep learning is identical to the theory of magnetism in physics. In physics, emergent alignment is called cooperative phenomena. The math of cooperative phenomena applies to social phenomena, biochemistry, immunology, and flocking patterns in birds. I mention this not to confuse you but to signal that when something has deep roots in how things work, one should think, “where else might I apply this principle”? What if we could turn the power of deep learning, the power of cooperative phenomena, to the discovery of alignment in product preferences, government policies, forecasting, planning, and decision making?
The algorithm underpinning the software, discussed in the Intuit case, gave a promising indication that this was indeed possible. However, at the time (~2006), we knew it worked but did not understand the fundamentals of why it worked. Intuit was one among many companies that benefited from learning alignment based on customer collaboration. For example, LEGO used the technology to learn the desired building experience and used that knowledge to radically shift their strategy to the big sets, the LEGO Star Wars series. The concept of learning preferences from open-ended conversations worked, but the AI technology, particularly scalable natural language processing technology, was not yet available.
The experience of applying the technology in a wide variety of use cases inspired a quest to build a system that could replace the adaptive learning algorithm with a robust AI platform. We tested and evolved the technology from 2016 to 2020, testing a specific hypothesis: “Can we learn if there is an investor market for a very early stage startup by taking a group of people consisting of individual investors, technologists, domain experts through a process based on learning alignment? The rise in the popularity of collective intelligence paralleled the rise of deep learning technologies. Our technology development efforts focused on integrating the findings of collective intelligence with AI. Collective intelligence demonstrated the power of the collective IQ. Together we are more intelligent. The diversity prediction theorem mathematically states that diversity reduces errors in judging outcomes.
The results speak for themselves. Every startup had a different cognitively diverse team matched to the startup’s needs. Results were over 80% accurate in predicting the presence of an investor market for a specific startup. Four and a half years later, the highest-scoring company is valued at over $350 million. We measured diversity by analyzing scoring patterns. Diversity here is measured not by your profile but by how you score and the reasons for your score. Low diversity is equivalent to groupthink. A boardroom or executive team of people who think alike will produce a low diversity score.
We discovered a direct correlation between accuracy and diversity scores. Learning alignment resulted in significant value creation. Over 40% of the funded founding teams were female-led, about ten times the rate of VC investment in female-led teams! Using a process focused on issues and knowledge-based discussion radically reduced systematic bias. That was an unintentional and welcomed benefit of applying the principles of collective intelligence.
The system uses a collective reasoning process that learns the alignment of reasons for scores. Separating the “idea” from “personal identity” opens the door to a bias-reduced process. Participants did not see each other’s scores but only their reasons for a score. With collective reasoning, we build a model of “Why” for every analysis using preference learning and NLP technologies. The model for “Why” follows concepts of causal reasoning. Every evaluation of startups resulted in a persistent AI model of the entire evaluation process (a probabilistic graphical model). These models allow simulation and replaying every step of the collective reasoning process.
An alignment of preferences forms a market. Our experiment learned and analyzed a collective set of reasons driving scoring action that led to investment outcomes. It was clear that knowing the underlying thinking and beliefs driving scoring behavior produced promising results. Discovering and understanding the intention and motivation of groups creates a powerful system for predicting future outcomes. There is real economic and social power in models of the collective mind of groups.
A case in point is Syntiant. In the summer of 2017, the Syntiant team sought their first round of seed capital. Based on a research grant, the technology team had developed a way to build chips that ran deep learning AI software on very low power at very low cost. The question was: Would this team and technology be able to attract the amount of investment needed to take it to market? Most major technology companies were interested in cheap, low-power, deep learning chips for AI applications.
The team of people brought together to assess the investor market potential for Syntiant included experienced private investors, technology specialists, and a few product designers. Syntiant came into the engagement seeking $1 million in seed investment. The initial thought was to go after a niche market (hearing aids). The hearing aid market seemed like an excellent way to avoid stirring up the giants. During the collective reasoning session, it became clear through the real-time analysis of the interactions that investors believed the team had the experience to go after mainstream markets like speech and vision. Syntiant pivoted, closed the seed round, and 90 days later closed an A round with Intel Capital, soon to be followed by investments from Microsoft and Amazon.
What happened in the Syntiant reason session? The entire process was guided and facilitated by the AI-based collective reasoning process. Synergistic interaction between the startup team and a cognitively diverse group of experts resulted in a new path forward. Syntiant scored high but more importantly, collectively reasoning together resulted in a new path forward.
In late 2020, we turned the technology experiment into a flexible, cloud application that is now commercially available from CrowdSmart.ai. Throughout 2021, interest grew significantly in the power of learning alignment across many application areas.
Let’s summarize. We conduct business and live in a highly complex adaptive ecosystem. Over 20 years ago, Gladwell’s Tipping Point made it clear; discovery of emergent patterns of behavior have substantial business and social significance. For that reason, we must apply the best technologies that match the challenge of the situation. Dynamically learning alignment from customers, groups teams in business, and social initiatives embraces the power of cognitive diversity is foundational to our path forward.
Google organized information on the web and benefits from our human quest for knowledge. Facebook organizes a way for people to connect and be seen, creating a pathway to align influencers and group identities. Organization and aligning around ideas separate from identity is an unmet need. The missing bit in all idea-sharing sites such as Linkedin, Reddit, etc., is peer-reviewed preference. It opens the door to discovering emergent preferences that form markets and movements, giving answers to questions you did not think to ask. The CrowdSmart system is a proven engine for learning peer-reviewed preferences, fostering a new way to create and learn together.
CrowdSmart technology taps into a fundamental principle found in nature. Cooperative phenomena have deep roots in the natural world, from magnetic behavior to the flocking of birds. Deep learning and magnetism, for example, share the same mathematics. The principle is this: emergent alignment grows from microscopic to macroscopic phenomena. The CrowdSmart system detects emergent alignment of shared human opinions and beliefs and summarizes the implications of that alignment. It catches the spark that may lead to a “Tipping Point.”
We demonstrated the power of learning alignment of customers on what must be done to improve customer promotion of a product or service in the case of TurboTax. In addition, we showed that learning alignment of consumer preferences leads to disruptive innovation (LEGO).
We demonstrated the power of learning alignment of investors and experts to select the best path forward for a game-changing new technology with Syntiant.
Innovation of new products and services is a complex problem. 90 % of new consumer product initiatives fail. How much would this be improved by learning the alignment of innovations from early concept through launch?
Recently, I was part of a group discussing how we might use technology to reduce the fragmentation and division in our country. For example, what would happen if we could share our opinions without revealing our political or cultural identities?
If you have a use case you would like to explore, such as innovation strategies, future of work, or social impact, email me at tom@crowdsmart.ai.
