FOR decades, corporate advantage came from accumulated expertise, refined processes, institutional memory, and hard-won best practices. But in an AI-driven economy, where foundational models rewrite the playbook monthly, these once-prized assets can calcify into liabilities. Knowing when and how to unl
earn, to deliberately dismantle outdated mental models and workflows, has become a quiet but urgent imperative.

What It Means to Unlearn

Unlearning is not forgetfulness; it is strategic amnesia. It requires recognising that legacy thinking, while once functional, now distorts perception and impedes adaptation. This challenge is especially acute in industries where AI tools have upended long-standing assumptions about expertise, decision-making, and value creation. A firm that treats knowledge as a static asset rather than a dynamic, evolving system risks falling out of sync with reality.

Paradigm Shifts and Ruptures

Thomas Kuhn described paradigm shifts not as smooth upgrades but as epistemic ruptures, moments when the very criteria for truth and judgment are replaced. AI is catalysing such ruptures across nearly every sector. A legal firm that uses large language models (LLMs) to draft briefs must unlearn its hourly billing incentives. A bank deploying predictive models for credit risk must unlearn the heuristics of human underwriters. A marketing agency using generative design must unlearn its traditional creative cycles. In each case, success depends less on the performance of the model than on the willingness to relinquish yesterday’s logic.

Yet, unlearning is notoriously difficult to institutionalise. Organisational power often clings to legacy knowledge. Promotions and prestige reward continuity, not volatility. Metrics favour predictability, not reinvention. The result is corporate cognitive dissonance.

Building Cultures of Humility

The remedy lies in embedding humility into the fabric of corporate life. Assumptions should be treated as perishable goods, not sacred truths. Companies must develop the habit of questioning their own knowledge systems. This can include internal “red teams” rotating dissent roles, zero-based knowledge audits, and institutional spaces for counterfactual exploration. As AI makes it cheaper to test alternative hypotheses at scale, the cost of clinging to outdated world views only rises.

Unlearning should not be viewed as a response to failure but as a proactive strategy. The firms best prepared for the future will be those that regularly ask, “What do we believe that’s no longer true?” and “What knowledge now blinds us?” In a world of accelerating models and shifting assumptions, perhaps the most valuable capability is not what a company knows, but what it is willing to leave behind.

Unlearning, in this deeper sense, is not a sign of weakness or ignorance — it is a mark of institutional maturity, a recognition that growth in the AI era may depend less on what is learned next and more on what is courageously unlearned now.

Paul Thompson is CEO and founder of HelloScribe, an AI research and consulting lab.