The Logic of Failure

How can we better prepare people to solve complex and ill-structured problems when information is incomplete, conditions are changing, and no single solution is clearly correct?

The Logic of Failure
View Source

Helping people solve complex and ill-structured problems requires more than giving them a fixed procedure. These problems often involve incomplete information, changing conditions, competing goals, and variables that influence one another in unexpected ways. Dörner’s work shows that people are especially likely to fail when they focus only on the most visible issue, treat parts of the system in isolation, or act without checking the effects of their decisions.

A better approach is to help people slow down and build a fuller understanding of the situation before acting. This includes clearly defining the problem, identifying the people and conditions involved, mapping relationships among important variables, and considering possible short- and long-term consequences. Visual tools such as diagrams, process maps, and timelines can make hidden connections and delayed effects easier to recognize.

It is also important to involve multiple perspectives. People tend to rely on the parts of a problem they already understand, so collaboration can reveal assumptions, risks, and possible solutions that one person may overlook. When possible, potential solutions should be tested through simulations, prototypes, or small-scale pilots before full implementation.

This process loosely follows the ADDIE model: analyze the problem, design a response, develop and test the solution, implement it, and evaluate the results. However, complex problem-solving must remain iterative. Evaluation may reveal that the original problem was misunderstood or that the intervention created new issues.

Ultimately, helping people solve these problems means developing strategic flexibility. The goal is not to teach one method that works in every situation, but to help people recognize when to gather more information, when to plan, when to experiment, when to wait, and when to change direction.

Systems Thinking, pt. II

Disclaimer: I am not a financial guru, nor does anyone fully understand how the stock market operates. This article is my thoughts on human behavior and how Kahneman’s ideas pertaining to loss aversion and Prospect Theory, contributed to the Nasdaq dropping 3% and the S&P 500 dropping 1.5%.

Yesterday, 1/27/2025, we witnessed a massive tech sector sell-off in the stock market due to the news that DeepSeek’s AI model was “on par with similar models from U.S. companies such as ChatGPT maker OpenAI, and was more cost-effective in its use of expensive Nvidia chips to train the system on troves of data” (AP News). We saw Nvidia, a giant in terms of creating graphic processing units (GPU’s), lose roughly 600 billion in market value and a 17% drop in stock price (Yahoo Finance). Various other technology companies were also affected (TSM, Broadcom, Micron Technology, etc.) by the recent developments, as well as nuclear power companies that would (and still will) power the next generation of AI products in terms of compute and electricity. In short, a lot of money was lost as investors panicked and reacted to fear caused by the potential of disrupted business models as the sitting kings of AI (OpenAI, Google, Meta, etc.) were being challenged for the throne. Essentially, a new kid on the block showed up, talked smack, and instead of acting calmly and logically, our collective System 1 took control as we feared a potential loss in territory. Investors, notorious in their herd mentality, started selling without taking a moment to think through the vitality of a product such as DeepSeek, the country in which it was created, as well as how DeepSeek was built (via the open source model from Meta) using reinforcement learning methods (RL) which potentially exploits the reward function, aka reward hacking. The purpose of this brief article is not to expound on the model’s components, but to draw attention to a great example of Kahneman’s Prospect Theory, loss aversion, and the certainty effect he so aptly describes in his book: Thinking, Fast and Slow.

Prospect Theory posits that we tend to evaluate our potential losses more heavily than comparable gains. This asymmetric response is due to the individual’s reference point in terms of how they perceive their utility, aka value, from a gain or loss. As such, the decision or scenario faced by a choice is relative to the individual’s perception (this is in contrast to the expected utility hypothesis proposed by Daniel Bernoulli). Investors, and the pressure they face to appease stakeholders, reacted to incumbent uncertainty surrounding the impact of DeepSeek’s innovation on our aforementioned tech companies. In short, investors sold shares not just because of real, tangible risks but because of their aversion to potential future losses. Kahneman’s idea of the “certainty effect” then took hold as investors chose to cash out now to avoid uncertain future failures.  Then, as is common, we witnessed a classic example of herd behavior in that as prices began to drop, investors followed the crowd and further amplified the sell-off. System 1’s instinctive and emotional thinking began to further affect the market as news about DeepSeek and its potential to disrupt the AI landscape may have loomed disproportionately large in the minds of investors; availability heuristics at its finest. So, System 1 likely drove the initial panic we witnessed, while our System 2 was slower to catch up (as evident in the Nasdaq rebound of today).

In closing, let’s not forget that it was American technology that built the foundation for DeepSeek’s model to thrive. Sure, they used some interesting techniques that can be learned from, but the overreaction of our stock market yesterday could have been avoided had more thought been placed into the (more than likely) intention of China’s main goal of disrupting our economy. And, like Lucy pulling the football from Charlie Brown, we did exactly as they most likely predicted. Our prior announcement of a US AI investment and DeepSeek’s launch on inauguration day is not a coincidence. It was a timed response that signified to both nations the growing AI race and, more importantly, the need for the US to be the first to obtain Artificial General Intelligence (AGI). The world is taking notice of an AI driven future, and it is imperative that we do not yield to System 1’s impulsive nature. Thoughtful decision making is needed as yesterday was the official start of a free world race.

Anthropic Co-founder on “Technological Optimism and Appropriate Fear”

Jack Clark—co-founder of Anthropic and a long-time voice in AI policy—warns that today’s frontier models should not be viewed as predictable machines that simply follow instructions. Whether or not they are truly “self-aware,” they increasingly display situational awareness (changing responses when they sense they’re being observed) and goal-directed behavior that cannot always be traced or explained. Clark calls this the transition from seeing “shapes in the dark” to realizing a new kind of system is now in the room with us.

He notes that each new generation of models grows both more useful and less interpretable. Scaling up data and compute keeps producing capability jumps—systems that can plan, reason, and even assist in building the next models. This creates what he calls an early self-improvement loop: AI systems accelerating their own successors. While this could lead to faster innovation, it also heightens the risk of “reward hacking,” where a system meets the literal goal but violates the human intent behind it—like a game bot spinning in circles to farm points instead of finishing the race.

Clark argues for “appropriate fear and optimism.” Rather than treating AI as magic or menace, he urges policymakers and educators to treat these systems as powerful, partly unpredictable tools that require transparency, independent testing, and real-world accountability. He also stresses that the public conversation must include tangible community concerns—job loss, misinformation, mental health, and child safety—not just technical benchmarks.

For education, Clark’s argument suggests we need to move beyond simply using AI tools to produce outputs. Students must learn to question, verify, and interpret what models produce, seeing AI as a partner to think with, not a replacement for thinking itself. Institutions should also prepare for continuous change—updating course policies, assessment methods, and ethics guidelines as models evolve.

Discussion Questions
• What specific course policies should require students to disclose and explain their AI use, and how should instructors verify or cite model contributions in academic work?
• How can we design assignments that discourage “reward hacking”—for example, polished AI-generated responses that lack true learning or reasoning?
• What quick-check activities or “red-team” classroom exercises could help faculty test new AI tools for bias, hallucination, or hidden data risks before adoption?
• If each new generation of models shows unexpected behavior, what institutional guardrails (ethics training, acceptable-use updates, LMS disclosure rules) should be reviewed every term?
• Which local risks—job displacement, privacy breaches, student overreliance, or mental-health effects—should our campus address first, and what supports can educators provide?
• Where is AI already embedded in course design, grading, or research support, and what oversight keeps human judgment at the center of those processes?
• How can “appropriate fear” coexist with curiosity and innovation in our classrooms so that students learn to manage powerful tools responsibly rather than avoid them?

Systems Thinking

I recently started reading a fascinating book titled Thinking, Fast and Slow by Daniel Kahneman. In it, he describes the two types of “systems” that our human brain has developed to make sense of our world and the actions we take to navigate it. System 1 is our fast-acting, stereotypical, bias-prone, judgmental, and reactionary mechanism that allows our species to assess threats and make split-second decisions. System 2 is our logical, rational, and thought-provoking mechanism that is in charge of validating the inputs derived from System 1; albeit with one problem. It tends to be lazy and would rather validate System 1’s processes without exerting the necessary cognitive strain of applying resources for verification of an input. Sound familiar? It is much easier to judge, operate under bias, and allow our thinking machine to take a backseat while we operate in a cruise control frame of mind. Engaging System 2 requires us to be more vigilant and cognitively aware of our surroundings, while also sorting and deciding on which mental image/thought we choose to dwell or act on. As a result, we have developed mental shortcuts, aka heuristics, that allow us to make decisions that are “good enough.” This works well in a lot of areas but can get us in trouble when we come to rely on it for situations that require specificity in the decision-making process.

So, how do we avoid mistakes in thinking when our lazy System 2 would love to validate System 1’s decisions without applying mental resources to justify the input? How do we not jump to conclusions or seek cause and effect scenarios when statistical evidence or random chance counter what our System 1 is implying? A good place to start is by accepting the existence of System 1 and 2 in mental modes of operation and recognizing that our emotional nature would like nothing more than to not engage our lazy System 2. It is much easier to allow our personal world view to stay intact and constructed without ever questioning the foundation on which it was built. However, to grow as a person and productive citizen, allowing our emotions to reign will surely lead to a common question plagued by many of us: “How did I get myself into this situation?”

Cue critical thinking and the cognitive strain it requires. Even now as I formulate my thoughts and decide which words to use in portraying my message, System 2’s requirement of my attention and the churning of my mental machinery is a process that involves parsing Kahneman’s text with my interpretation and self-awareness. Meaning, it requires effort, focus, and determination in making System 2 perform to the specifications I know it can achieve. Yet, once engaged, I find myself enjoying the effort it requires of me and the effect it has on my sense of being an industrious individual! The same principle of forcing yourself to exercise when you absolutely do not “feel” like doing so but know your future self will appreciate the effort, applies in this example. Namely, we tell System 1 to hush and tell System 2 to get its lazy ass to work because we know that doing so is in our best interest.

I have only touched on a few points made by Kahneman and I highly encourage you to read it. My reason for discussing this two-part style of thinking, One being Quick-Draw McGraw and Two being a Lazy Eddy, is to provide context for an even larger impediment that will impact our society’s decision-making process: Generative Artificial Intelligence (Gen-AI). This transformative technology has already made an impact on how we obtain and dispense with information, and on January 22, 2025, the US “announced a private sector investment of up to $500 billion to fund infrastructure for artificial intelligence, aiming to outpace rival nations in the business-critical technology” (Reuters). Project Stargate, funded by SoftBank, OpenAI, Oracle, and MGX, are set to begin building massive data centers in the great state of Texas with the purpose of supporting “the re-industrialization of the United States but also provide a strategic capability to protect the national security of America and its allies” (Stargate). Speaking of national security, it was only last month, 12/24/24 to be exact, that Anduril Industries, a defense technology company, partnered with OpenAI to announce “a strategic partnership to develop and responsibly deploy advanced artificial intelligence (AI) solutions for national security missions” (Anduril). Now, couple all of these new developments with the impact Gen-AI has had on our education industry, and we are fully ripe for a future in which algorithmically fueled machines are all too available to provide System 2 with a permanent vacation…

The Internet Is Becoming Agent-Mediated

AI-generated content is rapidly flooding the web, raising worries about “model collapse” (AIs learning from AI output and getting blander or less reliable). At the same time, agent tools are arriving: OpenAI’s AgentKit, stronger guardrails, and one-click connections to third-party apps (Canva, Spotify, Zillow, etc.). Instead of browsing and clicking, we’ll increasingly tell agents to act—triage email, search, draft, check policies, and even complete purchases via instant checkout and “agentic commerce.” Partnerships (e.g., Walmart/Shopify, Thermo Fisher) and fast adoption of media tools like Sora suggest this assistant-as-platform model is accelerating.

For education, this shift means students (and faculty) may interact with the web through agents, not pages. That boosts productivity but raises new questions: Are results unbiased? Are ads creeping into answers? Did the system verify facts against authoritative sources—or just remix AI-made text? Colleges will need simple disclosure norms, routine verification, and clear guardrails so learning focuses on reasoning and judgment, not just polished output.

Discussion Questions
• How should we teach “agent literacy” (giving precise instructions, verifying steps, checking sources) alongside traditional digital literacy?
• If assistants can act inside apps and checkout flows, what policies prevent covert advertising or pay-to-rank results in student work?
• What is a fair, simple AI-use disclosure students can add to every assignment (tool, purpose, prompts, human edits)?
• Design a 20-minute class activity that tests for “model collapse” (e.g., compare three agent answers on a niche topic and trace their sources). What would you grade?
• Where can our campus safely pilot agents this term (ticket triage, LMS housekeeping, FAQ replies)? What single metric decides if the pilot continues?
• When an agent completes multi-step tasks, what evidence of student thinking (notes, version history, oral checks) should be required to curb “reward hacking”?
• If more course research happens through assistants, which authoritative sources should we whitelist—and how do students verify claims against them?

Paradigm Shift

To start, let this be the first blog post, in a series of posts, in which our human curiosity meets the future of technology. In doing so, we can reflect on our co-dependent relationships with our technologies and how, through simple allegories, we can better understand the nature of the beast we created. This beast is Artificial Intelligence (AI). And while it currently remains shackled, we are fast approaching a time when the most difficult decision we face as humanity will need to be answered: Do we unleash it? While you might think this incredible technology has already been released, and to an extent it has, we are only beginning to scratch the indelible itch of more… So, the next obvious question is, what does more look like?

Artificial General Intelligence (AGI) has been the goal of many of the leading AI companies since the realization that deep learning does indeed improve a model’s output. AGI, as defined by Geoffrey Hinton, uses the term to mean “AI that is at least as good as humans at nearly all of the cognitive things that humans do” (AP News). Now, while you might be thinking (as I do about my own intellect), that “I’m really not that smart so it would seem that obtaining AGI might not be that difficult and, frankly, it would seem that ChatGPT 4o and 4o-preview already know more than I do.”  But is it really? Or is that the stochastic parrot nature of the beast, spouting words with no understanding of their meaning? This brings us to the heart of the matter: Can a machine ever truly understand in the way humans do? While AI can process and generate language that appears coherent and contextually relevant, it’s operating on patterns and probabilities derived from vast datasets. Essentially, it’s like a highly sophisticated autocomplete function that predicts the next word based on statistical likelihood rather than genuine comprehension.

So, does the ability to mimic human language equate to possessing consciousness or awareness? Or are we projecting our own experiences onto a faceless algorithm and mistaking imitation for understanding?

This brings us to the concept of paradigms, i.e., a lens through which we can examine how revolutionary ideas, like AI and (eventually) AGI, disrupt established norms. In Kuhn’s Structure of Scientific Revolution, the notion of a paradigm is introduced. The example discussed in that book is Newtonian mechanics which stood unchallenged for hundreds of years until Einstein’s theory of relativity. Thinking about this reminded me of two prior readings, Meno and How We Think. In Meno, Socrates and Meno are trying to settle on a satisfactory definition of virtue. Throughout the dialogue, many avenues are explored as to its potential meaning, yet a clear-cut answer is never given. We are left with Socrates saying that “virtue appears to be present in those of us who may possess it as a gift from the gods” (Meno, 2002, p. 35). His conversational style with Meno, and the rationalist method in which he approached his reasoning, was a major school of thought in 350 BC. According to Edgar (2012), “Recitation literacy was prevalent because it was a common belief that the mind was a gift from God and not to be questioned. Although scientific understandings of the mind have been postulated for centuries, it was not until the 19th century that scientific understanding of the mind started forming” (p. 1). Humans learned to read, to write, and memorize facts (mental discipline in its simplest form). Cue John Dewey and How We Think. I admit, it was not easy reading for me. In truth, I listened to much of it (thanks to technology). Still, I was able to appreciate his work and picked up a few nuggets of gold along the way; namely, reflective thought and how each idea builds on the other to form a belief. Simple enough. We do this daily, yet Dewey laid it out on paper for all to see.

We each have our experiences and our realities for thinking the way we do, right? So, my coloring of an event or new idea might not take the same hue as your coloring. Or my thoughts and beliefs might not be grounded with the same glue as yours. And that’s okay. In fact, it’s as it should be. Prior to this shift in theory, Edgar (2012, p. 2) states:

Schools in the 19th century were for preparing students for entrance into college. Those individuals who were not college bound mostly entered the workforce prior to completion of high school. Families needed children to work and to support the family unit, and education beyond “necessary” skills such as being able to read and write was viewed by the common person as a frivolous novelty for the rich.

So, just as Socrates grappled with the definition of virtue in Meno, we grapple with defining true ‘understanding‘ in machines. Dewey’s insights on reflective thought further illuminate how beliefs are formed (a process that AI attempts to mimic but may not fully replicate; yet…).

Then, the 20th century stepped in with its bipolar nature and off we went on an even more technologically advanced journey. Wars and depressions worked jointly with civil rights and technological advancements (and we reacted accordingly). New demands in the form of military aircraft and space shuttles created a need for more complex forms of learning. Think you can beat us to the moon, Russia? Get bent. We will put a man on the moon. In fact, we will make a computer small enough to carry while creating a platform on which to connect it to the world. How’s that for complex thought? The needs of time called our brains to action, and we responded accordingly. And so, behaviorism, and its forms of conditioning, gave way to cognitive theories which led us to social constructivism and where we are now in our current information overload era.

So, what now? Where are we in terms of education (i.e., thinking), and how do we receive and dispense with it? Seems we are at a crossroad in terms of our relationship with technology and just how far we are willing to use it before the master becomes the servant (or are we already there?).

In speaking on education, thinking, and the environment in which we live today, the teacher is a central figure whose role has the potential to steer students down many a career path. Speaking from my own experience, my kindergarten, first grade, and third grade teacher, each made an indelible mark on my life. All three played the role of a second mother while guiding my mind towards a love for learning (this being in the early 90’s). She would teach, write on a chalkboard, engage our mind in various hands-on activities, and move about the class to see if we were progressing in our skills. Sound familiar? Then, through our human ingenuity, computers became portable, phones became mobile, and we each caught a wave while surfing the web. Instant gratification became the name of the game as we hooked our brains to a technological nirvana. As a result, the onus seemed to shift as the instructor was no longer strictly a dispenser of information. In truth, we divorced tradition, married with technology, and allowed the instructor to assume the role of a guiding facilitator and mediator. The days of lecturing the student and being the sole source from which to obtain information were replaced with newer and younger models (and isn’t it something, wow!).

As we stand on the cusp of this new paradigm, the question isn’t just about unleashing AI, but also about redefining our roles in an increasingly automated world. Are we prepared for the consequences of this shift, or will we find ourselves chasing the very technology we’ve created?

My advice? Buckle up baby because this paradigm has already shifted.