Earlier this year, we argued that humans are the “OG ChatGPT”—predictive engines that use input, filter it through algorithms designed by nature and nurture, and generate output, some good, some bad, and some ugly.
The brain writes headlines from fragments, often mistaking speed for accuracy. In this article, we extend that idea by looking at how both humans and AI misunderstand time itself. Whether it’s a fast prediction or a misinterpreted epoch, both errors come from evaluating moments without their full context. If the first article explored how we misread the present, this one examines how we misread the past—and how both distortions limit good judgment.
In our previous column “Why We Believe the Stories We Tell Ourselves,” we explored how internal narrative shapes belief. When rationalization causes us to mistake the stories, we tell ourselves for objective truth, we begin to make decisions based on distorted “epochs” of our own experience that are out of context. The word “epoch” means “a turning point or a specific period in time marked by a milestone.” In artificial intelligence, “epochs” are deliberate cycles of learning in which a model analyzes past data to revise its understanding. Yet, while methodical, the models lack an understanding of the data’s context. Humans and AI both operate in epochs, but each miss context in opposite ways. Here, we propose that, whether it is humans or AI, looking at situations in a vacuum can lead to rationalization, especially when we don’t consider the full picture.
While using an epoch as a frame of reference can be an effective practice for looking at situations, the end result would be more accurate if it also considered the context of past actions and how they relate to the present.
AI vs. human epochs
In machine learning, a model doesn’t learn once, attach meaning, and move on. It reviews the dataset repeatedly over a defined timespan. Each training epoch ends with error measurement and internal calibration, and then another cycle of review begins. AI improves its output with every pass because it doesn’t overvalue its initial interpretation of data; it continually revisits the data before deciding what are the signals and what is noise. But it only looks at the past with the data, it has and nothing else, of course.
READ: Filling in the blanks: How AI mirrors our habits, errors, and possibilities (
Epochs matter because accuracy improves as the learning loop repeats, allowing the model to recognize deeper patterns rather than simply memorizing isolated examples. Eventually, the model converges to the point where performance on the dataset stops improving meaningfully, signaling that further training may no longer help. Finding the correct number of epochs depends on the complexity of the data and the depth of the model.
But what’s missing here is that the epoch might still be missing the context of situations not captured in the data.
In the human context, we hit “decide” before “review,” and we focus on the noise rather than the signal. By doing so, we too don’t take the time to review the past with pause, context, and its relevance to the present moment. While AI tries to retrain its decisions mathematically, humans try to rationalize it. Both with limited context and data.
Recent research suggests that our own brains progress through five distinct “epochs” across the lifespan, not as steady linear growth but as a series of structural rewiring phases marked by major turning points around ages 9, 32, 66, and 83. In the earliest phase (birth to 9), the brain is pruning redundant synapses and consolidating networks; in the next phase (9–32), it is growing and refining connectivity between regions, making cognition more efficient; by around 32, the brain hits what researchers call “adult mode,” a long plateau of structural stability; and after 66 the brain begins gradual reorganization and decline in connectivity, culminating around 83 in more isolated and localized networks.
Because the brain itself transitions through distinct structural epochs, it suggests that our capacity to learn, adapt, reflect, and reinterpret experiences may also vary across life stages. In childhood and young adulthood, the brain is highly plastic and rewiring rapidly; decisions made, habits formed, and experiences lived during that time may leave deep impressions. In middle adulthood, the mind may function more stably but perhaps less flexibly. In older age, reorganization may slow, requiring different strategies for learning, adaptation, and reflection.
AI, in contrast, reviews the past without truly understanding it. It processes patterns, frequencies, and correlations, but it cannot interpret the emotional, social, or psychological meaning behind them. There is no reflection. Its retrospective gaze is wide but shallow, scanning everything yet grasping nothing beyond numerical relationships. When an AI model revisits its history during training epochs, it is rerunning data without any sense of why certain moments mattered, what consequences unfolded, or how intentions shaped outcomes. In that sense, AI sees the past, but it does not feel it. Its reflection is mechanical rather than experiential, which makes its conclusions precise but not fully human.
READ: A peek into the 22nd century: Life in the age of AI and abundance (
We, by contrast, often make decisions in the moment as if the past were irrelevant or too heavy to revisit. This is corroborated in science with the drop in elasticity of our brains as we get older. Our emotional urgency compresses the present into a narrow window where instinct dominates over evidence. In doing so, we create our own epochs defined more by immediacy than by accuracy. Our memories are selective, our interpretations are biased, and our cognitive distortions are activated. We act as if the present is a clean slate, even when it is quietly shaped by everything that came before.
Where AI looks back without context, humans look forward without grounding.
The opportunity lies in designing a new kind of epoch that corrects both blind spots. Such an epoch would ask AI to incorporate context rather than merely catalog information, and ask humans to integrate history rather than selectively ignore it. It would push machines to approximate meaning and push people to revisit evidence with humility.
Using this approach, we become better at working with AI to support creativity or to create a decision-making cycle where humans borrow AI’s discipline for review, and AI borrows humans’ capacity for context.
This blended epoch becomes a fuller process of decision-making that honors the emotional truth of the present while not ignoring the lessons of the past. It is based less on emotion and impulse, and more on a reflective loop that captures what both humans and AI miss when they operate in isolation.
The question is how we might rewire these loops of learning to make them more productive.
In Part 2, we explore the relevance of these feedback loops.
Sreedhar Potarazu, MD, MBA, is an ophthalmologist, healthcare entrepreneur and author with more than two decades of experience at the intersection of medicine, business, and technology.
Carin Isabel Knoop founded and leads the Case Research & Writing Group at Harvard Business School, an output-driven team of researchers who collaborate with faculty members and organizations worldwide to craft world-class curricular and pedagogical experiences on work and leadership. She is the co-author of Compassionate Management of Mental Health in the Modern Workplace (Springer).