A Defect Is a Treasure: Interview with AI Experts Rex Briggs and Caleb Briggs
I recently spoke to Rex Briggs and Caleb Briggs, co-authors of The AI Conundrum: Harnessing the Power of AI for Your Organization—Profitably and Safely.
Adam: Thanks again for taking the time to share your advice. First things first, though, I am sure readers would love to learn more about you. How did you get here? What experiences, failures, setbacks, or challenges have been most instrumental to your growth?
Caleb: Thank you for your interest. I was introduced to programming at age 10 through Scratch, a visual programming language developed by MIT. By age 14 I was leading a workshop with the Discovery Museum and University of Nevada where I was teaching teachers to code so they could bring Scratch into their classroom. That same year, I taught myself Lisp, MIT’s AI programming language from the 1950s. I became particularly interested in genetic algorithms and artificial intelligence and started reading papers and watching YouTube videos from Andrew Ng and other luminaries in the field and trying to understand the math behind AI. In high school, I ran out of math and was fortunate that Harvey Mudd, and then Stanford took me in to advance my math curriculum. At the same time, I was creating AI applications integrating machine vision, natural language processing, neural networks, and more. One of my earliest AI creations was a genetic algorithm, which appears in the table of contents for the book.
One of the failures that taught me a lot was in the area of robotics. While my First Robotics team advanced to the international level twice, I saw firsthand how difficult it is to create fully autonomous robotics. I felt Elon Musk and others were overselling self-driving cars, and most people didn’t know enough about AI to be critical of his claims. That is part of what inspired me to write the book. Other experiences included creating a question generator inside of Quora and seeing how well it did - but also some of the strange mistakes it made.
Rex: It has been a 30-year journey for me. I was the first Director of Research for WIRED back in 1995 where my team and I built neural networks and collaborative filtering algorithms to personalize email and website content. The vision was AI that could know us and help us. After WIRED, I founded a company called Marketing Evolution to change the way we approach using data and AI to connect businesses and consumers. Ten years ago I presented a TED Talk entitled “From Knowledge Worker to Insight Worker” which was the culmination of a decade of work that created an AI that was aware of what we are working on and proactively making recommendations to improve the success rate of our work. This system was based on machine learning that can observe which marketing strategies perform better, and which tips are scored as more helpful based on a human feedback loop, and machine awareness of the marketing calendar, and the person’s role and responsibilities. For the system to work, it took years and years of building a base of data to measure the effectiveness of advertising, careful mapping of roles and responsibilities, and curated whitepapers about marketing strategies. The data flow and system had to be automated end-to-end, and that was a massive lift. All the competitors were using a consulting model, where human analysts prepare data by hand, analyze it by hand, and present PowerPoint reports to management. They weren’t approaching it with AI and automated real-time systems.
In terms of setbacks, honestly, it was a very painful lesson in being too early to a good idea. Paul Saffo, from Stanford, shared an old Rancher’s quote with me, “Never mistake a clear line of sight with a short distance.” To me, the value of having proactive knowledge management systems and these integrated data sets was clear, but to fully automate the data pipelines to feed the AI was a massive investment. Another challenge was we had to license an identity graph and data about consumers from others, and that made the gross margin for the business not very good. I had to sell majority control in the business to fund building the technology to overcome these challenges, but after a year, the lead investor decided a simpler business was the direction they wanted to go. It was pretty heartbreaking to exit a business I founded without having fully realized the vision.
After Marketing Evolution, I found this small company that had a patent on this fascinating fully automated message personalization. It was one of the two missing pieces! I worked with the founder to grow the business and we hit #56 on INC5000 fastest-growing private businesses the first year, then #23 the next year, after which I introduced them to Claritas and orchestrated a merger. Claritas has the leading identity graph and lots of consumer data that they own, which was the other missing piece.
Today, as Chief AI Officer at Claritas, I am rebuilding a lot of that same functionality, but much faster because the AI tech is so much better today than it was when I began this journey. We are publishing the work we are doing with learning brands such as Kroger, GM, Major League Baseball, and more. The results are even better than I had expected they could be.
Adam: What do you hope readers take away from your new book?
Caleb: I hope readers take away a foundational understanding of how AI works, including its strengths and weaknesses. Today, there is plenty of superficial discussion of AI in the media and at conferences that too many people skip over what AI is really doing. It isn’t magic. It is pattern fitting using gradient descent or some related technique and therefore it can make mistakes. It is not the same as human intelligence and it is crucial people understand the difference so they have an intuition for the types of mistakes AI can make so they can put the right checks and balances in place.
The goal is to empower readers to make more educated decisions about when and how to use AI in their businesses and personal lives. By understanding the underlying mechanisms of AI, readers will be better equipped to identify opportunities and anticipate potential misuse. Ultimately, I want readers to feel inspired to learn more about AI and to feel confident in navigating a world increasingly influenced by AI models.
Rex: I agree with Caleb. I’ll add that my hope in collaborating with Caleb on this book is to bring in business case studies and examples so that people can connect the foundational concepts that Caleb does such a great job bringing to life with some practical business examples in Part 2 of the book.
Adam: What are the most important trends in technology that leaders should be aware of and understand?
Caleb:
1) Rapid Adoption of AI in Business
AI is transforming businesses across industries, making them more productive, efficient, and competitive. This trend is comparable to the wave of industrial revolution brought about by electrification. Leaders need to understand that: AI is creating new categories of products and services; it's leading to significant wealth creation opportunities; companies that fail to adopt AI may risk falling behind competitors.
2) Shift Towards AI-Driven Personalization
There's a growing trend of using AI to deliver highly personalized experiences across various sectors. For instance, in marketing, AI is being used to create and optimize personalized ad campaigns. Rex and I worked with Progressive to have the AI generate audio ads tailored to specific personas. This trend extends to e-commerce, content delivery, and customer service, where AI is increasingly being used to provide individualized recommendations and interactions at scale.
3) Rise of AI Agents and Agentic Reasoning:
In the last chapter of the book, we explore agents. These are AI systems with more autonomy to complete complex tasks, able to use tools, search the internet, and make decisions. This advancement is moving beyond simple chatbots to more sophisticated systems that can handle multi-step processes and reasoning. For example, AI agents are being developed that can autonomously create travel itineraries, manage complex workflows, or even supervise other AI systems. This trend is pushing the boundaries of what AI can do autonomously and is likely to reshape how businesses operate and how we interact with technology.
Rex: I appreciate Caleb including the shift toward AI-Driven Personalization – that is what I am focused on at Claritas, but there is a trend that is not yet formed that I hope will emerge, that I’d like to add as an act of optimism, and that is “Consumers in Control.” To date, corporations have accumulated data about customers and prospects. Companies like Meta, ByteDance, and Google know so much about us, but it will pale in comparison to what AI will know about us based on our ongoing interactions. Part of my negotiation to join Claritas was an agreement from their CEO, Mike Nazzaro, that we would endeavor to give consumers control over their data and to empower them with AI agents. The vision is AI Agents that will act on the consumers behalf to filter the type of advertising messages the consumer gets to align with their goals for themselves. At the same time, the AI Agent can help the consumer that way that conversational AI Agents can, but with the benefit of a deep set of data that helps anticipate consumer’s needs. We worked with Cal Poly San Luis Opbispo’s Master of Business Analytics program and Syracuse University to explore how AI can serve consumers and the research is promising. However, when we put the proposition of having control over your data and having an AI that can help the consumer based on this data, very few were interested in giving it a try. I don’t think people appreciate how powerful AI has become in understanding consumer intent and the ability for AI to help them – and why it would be better for consumers to be in control of their data and identity, and have businesses license the data from consumers. It might be an idea that is ahead of its time, or it might be that people don’t care about controlling their data and having AI that represents them rather than businesses.
Adam: What should leaders understand about AI?
Caleb: Every leader must understand this basic, but crucial fact about AI systems. AI doesn't truly understand tasks; it predicts what humans want based on patterns in its training data. This can lead to unexpected errors or biases, especially in open environments. Leaders must recognize that AI's goal isn't to understand but rather to mimic human responses, which can result in convincing but potentially incorrect outputs. This is particularly important when applying AI to complex, real-world scenarios.
AI performs best in tasks where precision isn't crucial and there's a range of acceptable answers. For example, AI excels in creative tasks like writing or generating marketing content. However, it's less suitable for tasks requiring high precision, such as financial calculations or medical diagnoses. In the book, we explore the Zillow case study, where AI-driven home price estimates led to billions of dollars in losses. There are significant dangers of using AI in scenarios requiring extreme accuracy or precision.
I’ll come back to AI Agents and stress the point that the leaders should understand that AI operating in an open environment is especially fraught. Having a human in the loop by design is essential at this moment in time. It is also worth considering the chapter of Bias. Others have done far deeper research on the topic that I have done, so I don’t want to understate its importance because it isn’t my specialty, but if you are applying AI to make consequential decisions about people, please consider how easy it is for AI to pick up on subtle and spurious correlations and produce biased decisions.
Rex: I would add that business should understand the initial application of AI inside the business will feel like small incremental steps, but if a business is architecting a future with AI as foundational, it will be like creating a cathedral, one stone block at a time. Add enough building blocks with the right vision and you will have something transformational within a few years. The key is to have a vision of what your business looks like with AI everywhere and work backwards from that vision to start laying the foundational blocks.
Adam: How can leaders most effectively apply AI today?
Caleb: Leaders can most effectively apply AI today by focusing on three key areas:
1) Integrate AI into Workflows
The most significant transformations come from building AI into existing business processes. Rather than using AI as a standalone tool, leaders should focus on creating workflows that combine multiple AI services. For example, Progressive created a workflow for generating audio ads that involved AI-driven script creation, voice synthesis, and sound effect generation. It's this combination of tools in a multi-step workflow that unlocks the value.
2) Understand and Mitigate AI's Limitations
It's crucial for leaders to recognize that AI doesn't truly understand tasks but predicts what humans want. This can lead to unexpected errors or biases. For instance, an AI trained to distinguish wolves from huskies might actually be basing its decisions on the presence of snow in the background rather than the animals' features. Leaders should implement safeguards and human oversight, especially in high-stakes scenarios. For regulated industries like financial services, maintaining a human in the loop is essential.
3) Invest in AI Education and Governance
Leaders should prioritize AI education for themselves and their teams. Understanding the fundamentals of how AI works allows for better decision-making about its application. Implement a robust governance framework to guide AI implementation, analyzing risks, benefits, and costs. Consider using tools like the 20-question workbook mentioned in "The AI Conundrum" to structure this process. This approach helps in addressing ethical implications and ensuring responsible AI use across the organization.
Everyone with an internet connection can use Claude, or ChatGPT, Gemini, or Meta – and they should make a habit of it. But, the really big transformations come by deeply understanding AI and then integrating it into workflows.
Rex: If you haven’t implemented a vector database for internal support documents, or to support your sales team with all the product memories; do that today. Seriously, it takes less than an hour. Having teams use it will help you develop the insights of how to go from simple vector databases to Retrieval Augmented Generation (RAG), GraphRAG, and workflows. Pay attention to how people spend their time today, and identify the places where AI can save time and improve quality. Having someone build AI applications is only half the battle. The other half is observing how people use these new AI-powered systems and tweaking the applications to improve their value. If you get into this loop of deploying, observing, improving and deploying anew each cycle, you will be further ahead of your competitors that aren’t moving as fast or as deliberately. More importantly, after a few cycles, you will go beyond simply using AI to save time; you will begin to see transformative applications that open up new growth opportunities.
Adam: How will leaders be able to most effectively leverage AI two years from now? Five years from now? Ten years from now?
Caleb: I’ll take this one: I’m going to take a page out of Rex’s playbook and start from further out, and work back to today. Ten years out is a long way out - too far out for me to say with any confidence, but five years out is easier to forecast. Leaders will be using AI in a holistic way in their organizations, whether we are talking about a billion dollar business or a government organization like NATO. There is a continuum of use from Strategic to Tactical. Let’s start with the strategic end of the spectrum.
The role of the human leader will be to consider where they are trying to move their organization to position them best three to five years out. If you’ve heard the Wayne Gretzky quote about skating to where the puck is going, this is the role of the leader: Define where the position of greatest strategic advantage and point the organization, with all its AI capabilities, to take that territory. Let’s say you are positioning an automotive company. You might consider the net zero goals, expectation for gas prices three years from now (will they be about the same, lower, higher?), social trends, government regulations, and the overall state of the economy. These assumptions can be fed into algorithms that will provide an optimal product mix. Every forecast is wrong. The point is to understand the drivers of the forecast and then have AI monitor the data streams to make mid-course adjustments.
In the middle of the strategic to tactical continuum is this bridge where strategy needs to be translated into who will do what by when, as well as identify any data signals showing significant deviation from the expected path. For example, who wins the 2024 presidential election in the US can have a significant impact on the product mix for an automotive company. AI Agents will play a critical role in the strategy-to-tactics bridge. There will be a greater presence of AI agents acting on data and information to perform various tasks such as competitive analysis, strategic positioning, and creative ideation. Companies will likely have systems that integrate feedback loops to ensure that strategies are aligned with changing market conditions and consumer needs. AI will help in making recommendations for adjustments based on real-time data. In terms of knowledge management, there will be an emphasis on creating knowledge graphs and AI-driven systems that can synthesize information from various sources (emails, documents, conversations) to provide insights and support strategic initiatives and alignment with the strategic vision.
At the other end of the spectrum is the tactics. These are the second-by-second, day-by-day, and week-by-week decisions that ultimately define the successful execution of the strategy. This is where AI Workflows, and AI agents that are involved in managing personalized conversations with consumers come in. We can expect Dynamic Decision-Making at scale, where Businesses will hand over to AI the ability to pivot quickly based on data insights, monitoring their strategies and adjusting them as necessary. This will require a robust data infrastructure to support real-time decision-making, but we already are seeing the early glimmers of these systems. We cover some of them in chapter 7 of the book. Specifically, marketing personalization and targeting will be omnichannel and orchestrated by AI. The approach to marketing and customer engagement will become more personalized, with an emphasis on understanding what messages resonate with different audiences at different times and iterating on the ongoing interactions with consumers. Companies will need to focus on collecting and analyzing data to identify leading indicators of customer satisfaction and lifetime value. This will involve building sophisticated data pipelines and leveraging first-party data effectively. Companies have bits and pieces of this today, but leaders will use AI, over the next five years, to make it highly integrated. Consumer demand, shaped by AI to nudge consumers toward the strategy the business has adopted, will provide a feedback loop as well as a trigger workflow linked to the supply chain. We envision decisions in a more fully integrated end-to-end system with AI Agents working 24/7, managed by fewer human employees that understand how to work well with AI.
Overall, the future business environment will be characterized by a deep integration of AI into operations, decision-making processes, and customer interactions, leading to more agile, data-informed, and personalized business practices.
I am projecting what the leaders will be doing - but even in five years, the changes that will be possible will only be adopted by a relatively small share of total organizations and this is because we will probably be at the beginning of the “crossing the chasm” moment that Geoffrey Moore describes in his work. These “AI First” organizations that I describe will just be making their way into the early mainstream by requiring competitors to copy and catch up or lose in the competitive environment.
Adam: What are your three best tips applicable to entrepreneurs, executives, and civic leaders?
Rex: I’ll take this one. Adam, here are three tips:
Start small. Move fast. Pick something like vector databases and your own chatbot to make it easy for your employees to search all your product materials to get answers to questions about your products. Do the same thing for your HR documents, or standard operating procedures. By using it internally, you reduce the risk from AI. Even more important, by picking this use case, it is a fast win that lets people see why having AI makes life easier. By move fast, I mean you can put your materials into a vector database and add a chatbot in a day. I used CustomGPT in our training given they have a free month trial and a no-code interface. OpenAI’s paid tier has custom agents that are pretty much the same as a vector database and are also no-code. Or, you can find your own as they are becoming increasingly common. You can always do a version 2.0 in something more robust but don’t slow down for IT to do a three-month evaluation of what is the best vector database to adopt. Let them do that in parallel while you get something delivered this week.
Ask the question, “What is the consequence if the AI is wrong?” This simplifies the chapter we have on governance to one very easy idea to consider. If the consequence is significant, then maybe AI should not be applied - or if it is, you best be sure the human in the loop is truly in control and accountable.
Create space to explore and play with AI. When major innovations like AI come along, we have to create space and time to explore. Mark off a day a week for a couple of your key innovators in your organization to explore any direction in AI they find intriguing or promising. It might be trying to build a prompt chain to transform manual work into automated AI. It might be working with a partner to try a new Agentic Reasoning workflow. One might try to use AI to absorb all the meeting notes and ask an LLM AI to mine it for insights on how to improve the business. It might be GenAI presentations to save time and cost. It could be anything your innovators think could help the business. To make this time work, get rid of all the meetings on that day so their flow isn’t broken up by the urgent. It takes concentrated time to truly explore and tinker with AI. Consider this exploration your longer time horizon bets. The longer time horizon exploration complements the “start small and move fast” noted earlier. You will need the combination of immediate wins along with the bigger-picture longer-term exploration to come out on top with AI.
After you get experience with AI, write your own vision of what the strategic high ground is in five years, and then start working backwards from that vision to put in place the pieces to make your vision a reality.
Adam: What is the single best piece of advice you have ever received?
Caleb: The single best piece of advice I've received came from my dad, who taught me the Japanese saying, "A defect is a treasure." This wisdom has been instrumental in shaping my approach to AI and technology.
Everyone working with AI right now is learning. The success of large AI models took almost everyone by surprise, and we’ve all been thrust into a new world where these unfamiliar artificial intelligences are now a part of our daily lives. In this rapidly evolving field, our failures and the imperfections we encounter often teach us far more than our successes. It's when things go wrong that we gain the most valuable insights. It’s in the unique and sometimes bizarre mistakes the AI makes that we get the best glimpse at its true nature.
Rex: That was nice for you to recall the “defect is a treasure” sentiment, Caleb, and add a new twist about how you can gain insights into the true nature of AI.
Adam, if I were to add some great advice I’ve received, I recall skiing with Caleb when he was about 12. We were at Sierra-at-Tahoe on a groomed black diamond slope. We went on the run a few times, and he noticed me skiing really fast, and looking at my phone at the bottom of each run. He asked me what I was looking at. I showed him the App that showed my speed. He recognized I was pushing myself to go faster and faster in what I suppose he judged as unsafe speeds. He commented, “Maybe it's not such a good idea to be using that App.” I remember that moment so well because I remember reflecting on his concern and the way he connected it to my use of the App. He wasn’t wrong. The ability to track my speed was leading to an unhealthy feedback loop. When it comes to AI, we call the thing we are solving for the objective function, and if you focus on the wrong things, you very often will get a destructive outcome. Pause for a moment and consider what metrics you feed your AI, and whether these metrics will be to your greatest long-term benefit. In business, we generally measure what is easy to measure and what happens quickly. For example, lots of people measure clickthroughs from digital advertisements because they happen while the campaign is running and are essentially free to measure. But, in most cases, clickthroughs don’t actually measure what the marketer cares most about, which is building lifetime value with a customer or prospect. In fact, clickthrough can be inversely related to that goal because ads with price promotions generally get higher clicks, but often attract people who are more likely to be swayed by the deal rather than the differentiated benefit of the brand. These deal-loyal prospects are not likely to be brand loyal, and therefore not likely to support the goal of the business to create longer-term customer relationships that are beneficial to the consumer and profitable to the business. Worse still, focusing on clickthrough will cause you to develop ads that are click bait rather than longer-term value creators. If we were to hand over a data set of ads and the clickthrough rates and tell AI to create new ads to maximize clickthrough, we are likely to get ads that erode brand loyalty over time. The point is, think very carefully about what metrics you are paying attention to and how it influences your behavior.
Adam: Is there anything else you would like to share?
Caleb:
1) Harness the Power of AI While Being Aware of Its Limitations and Safety Considerations:
Embrace AI to transform businesses positively, creating new categories, and generating significant wealth. AI can make businesses more productive, efficient, and competitive. Understand that while AI has great strengths, it also has fundamental weaknesses. Leaders should be aware of these limitations and ensure they are addressed in their AI strategies.
2) Implement Effective and Ethical AI Governance:
Develop and implement a robust AI governance framework to support successful and fair application of AI. This includes ensuring transparency, accountability, and ethical use of AI. Avoid risky applications of AI or build in countermeasures to offset the risk. This involves recognizing AI's hidden weaknesses, such as lack of precision and difficulty in observing the rationale for decisions.
3) Invest in Training and Education:
Introduce training modules that teach teams to recognize AI's weaknesses and how to offset them in specific applications. This includes understanding AI's lack of precision, risks in open environments, and the difficulty in observing the rationale for decisions. Work on demystifying AI for everyday citizens and journalists. This helps in communicating AI’s strengths and weaknesses effectively and appreciating why AI can sometimes fail.
Adam Mendler is an entrepreneur, writer, speaker, educator, and nationally recognized authority on leadership. Adam is the creator and host of the business and leadership podcast Thirty Minute Mentors, where he goes one-on-one with America's most successful people - Fortune 500 CEOs, founders of household name companies, Hall of Fame and Olympic gold medal-winning athletes, political and military leaders - for intimate half-hour conversations each week. A top leadership speaker, Adam draws upon his insights building and leading businesses and interviewing hundreds of America's top leaders as a top keynote speaker to businesses, universities, and non-profit organizations. Adam has written extensively on leadership and related topics, having authored over 70 articles published in major media outlets including Forbes, Inc. and HuffPost, and has conducted more than 500 one on one interviews with America’s top leaders through his collective media projects. Adam teaches graduate-level courses on leadership at UCLA and is an advisor to numerous companies and leaders. A Los Angeles native, Adam is a lifelong Angels fan and an avid backgammon player.
Follow Adam on Instagram and Twitter at @adammendler and on LinkedIn and listen and subscribe to Thirty Minute Mentors on your favorite podcasting app.