Reverse Information Paradox that Nadella flagged is real, but ‘back to analogue’ is no answer

Microsoft CEO Satya Nadella has identified something that deserves serious attention. His ‘Reverse Information Paradox’ names an anxiety that many enterprises have felt but have not yet articulated precisely: in the age of artificial intelligence, firms increasingly enrich the very systems they pay to use. Every prompt, correction, evaluation and workflow becomes part of a learning process from which the model provider may derive future value.

Nadella’s is an important observation, but it is also incomplete. If his thesis is accepted without qualification, it is easy to reach a seductive yet dangerous conclusion: that the safest world is one in which information flows are tightly restricted and organisations zealously guard every trace of knowledge they produce. Push that instinct far enough, however, and one ends up romanticising precisely the analogue economy humanity spent the better part of half a century escaping. Would any of us genuinely wish to return there? The answer is almost certainly no.

More from Opinion

Kenneth Arrow’s celebrated Information Paradox explains why information markets behave differently from ordinary markets. Before purchasing information, one cannot know its value. Once it has been disclosed, however, it has effectively been obtained. Patents, copyrights and trade secrets evolved to resolve that dilemma without destroying the exchange of ideas altogether.

Nadella argues that artificial intelligence reverses this relationship. The customer must disclose information merely to extract value from the product. Instead of sellers giving away knowledge, buyers do. The better an AI system performs, the more context, institutional memory and proprietary expertise it requires. The customer’s competitive advantage risks becoming the vendor’s training data. His diagnosis is persuasive. His prescription is largely sensible as well. Every enterprise should retain ownership of its evaluation frameworks, institutional memory, fine-tuning data and accumulated organisational intelligence. Firms should not become tenants on someone else’s intellectual estate. Nadella’s emphasis on sovereign AI infrastructure, private learning environments and decoupled orchestration deserves close attention from every boardroom.

Yet this is precisely why the debate should not be framed as a struggle against data flows themselves. The Reverse Information Paradox is not merely a vulnerability but also one of the principal reasons why artificial intelligence has progressed so rapidly. Every time millions of users correct an AI model, clarify a prompt, or reject an inaccurate answer, the system's collective intelligence improves. Every interaction becomes another observation in a giant experiment involving almost the entire connected world.

This learning flywheel is unprecedented in economic history. Unlike industrial machinery, which depreciates through use, modern AI often improves through use.

Intelligence compounds. That compounding is impossible without information exchange. Those who see only the risks overlook the extraordinary gains that continuous learning has already delivered.

Back to the analogue economy? Nope

How about an instructive comparison with the analogue economy? Consider the humble typewriter. Manufacturers had virtually no insight into how customers used their products after purchase. There were no usage statistics, telemetry, crash reports or software updates. New models appeared years apart, based largely on intuition, market surveys and educated guesswork. Innovation was necessarily slow.

Television worked similarly. Broadcasters relied on Nielsen-style sampling to estimate audience behaviour. They knew almost nothing about what individual households actually watched, when viewers lost interest, which scenes they skipped, or what programmes they wished existed.

Programming was designed for the statistical average. Discovery was poor. Choice was limited. Feedback arrived months later, if at all.

The price I pay for the content I prefer

Today’s digital platforms operate on an entirely different principle. Streaming services know where viewers pause, rewind or abandon a programme. Music platforms understand not only what listeners enjoy but also what they skip after eight seconds. Search engines learn from billions of queries every day. Online retailers continuously observe purchasing behaviour.

These systems certainly collect enormous amounts of behavioural information. They also produce vastly superior outcomes. Recommendation engines expose audiences to niche content that traditional broadcasters would never have commissioned.

Search costs have collapsed, personalisation has become routine, small creators find global audiences, and consumers enjoy levels of choice that would have seemed utopian in the 1980s.

None of this emerged despite data collection. It emerged because of it. The economist George Stigler demonstrated decades ago that search costs are an economic burden. George Akerlof showed how information asymmetries create dysfunctional markets. The digital economy reduced both on an astonishing scale.

Artificial intelligence extends this trajectory precisely, lowering not merely search costs but cognitive costs. That does not eliminate new asymmetries but rather creates them.

The middle path

The answer, however, is institutional innovation rather than technological nostalgia. Arrow’s paradox was not solved by abolishing markets for information but by inventing new legal and commercial institutions. The Reverse Information Paradox demands something similar.

Nadella himself gestures towards the solution:

• Private evaluation frameworks

• Enterprise-controlled memory

• Model portability

• Choice between providers

• Tenant-bound learning environments.

Far from arguing against AI’s learning flywheel, they stress who controls it, a distinction that matters enormously.

Whereas today’s AI economy is indeed a one-way extraction machine in which value inevitably migrates towards infrastructure owners, it ignores the reciprocal nature of the exchange. Enterprises do not merely surrender information; they receive capabilities that would have been economically impossible to build independently.

A medium-sized manufacturer, for example, can now deploy language models that incorporate knowledge distilled from billions of documents. A hospital gains diagnostic assistance informed by research produced across continents. A law firm accesses analytical tools that would previously have required armies of associates.

The bargain is not obviously exploitative. It resembles many previous technological revolutions, with railways being a case in point. It generated value for railway companies while creating entirely new industries. Internet search enriched Google while dramatically expanding access to human knowledge. Smartphones created extraordinary wealth for platform owners while transforming productivity for billions of users.

Economic history rarely presents zero-sum exchanges. Hayek’s insights remain indispensable here. Knowledge is dispersed, existing within frameworks of time, place and institution. No central planner can fully aggregate it. Nadella is right to insist that enterprises preserve this particular knowledge inside their own trust boundaries.

But Hayek also celebrated markets precisely because they allow dispersed knowledge to interact without requiring central control. Properly governed artificial intelligence can become another mechanism for coordinating dispersed knowledge rather than appropriating it.

The challenge is designing the rules, bearing in mind the competition policy, data portability, open-source models, privacy-enhancing technologies such as federated learning, confidential computing, differential privacy, contract law, and regulatory scrutiny where market power becomes excessive. These are difficult institutional questions, but preferable nevertheless to retreating towards an economy defined by information scarcity.

There is another irony in the current debate. Many people declare themselves deeply concerned about digital data collection. Their behaviour suggests otherwise. When given a genuine choice, consumers overwhelmingly prefer Netflix to linear television, Spotify to compact discs, Google Maps to paper atlases, online shopping to printed catalogues and AI assistants to static software.

Economists distinguish between stated preferences and revealed preferences; the latter usually tell the more reliable story. People willingly exchange certain forms of data because the benefits are immediate, tangible and often enormous. This neither implies that every form of surveillance is justified, nor does it diminish legitimate concerns about enterprise knowledge leakage. It simply reminds us that information sharing is not merely a cost. It is also an investment. The Reverse Information Paradox, therefore, deserves neither dismissal nor panic but engineering. Just as patents enabled industrial economies to flourish without undermining incentives for invention, the AI economy requires new mechanisms that preserve enterprise sovereignty while allowing collective intelligence to continue to improve. The balance will determine whether artificial intelligence becomes another productivity revolution or merely another concentration of economic power.

Nadella deserves credit for identifying the problem. The next stage is to ensure that, in solving it, we do not inadvertently dismantle the very learning systems that have made artificial intelligence revolutionary in the first place.

The future does not belong to the analogue past. It belongs to institutions capable of building an intelligence compound without forcing those who create it to surrender ownership of what makes them unique.

---------------------------

(The author is a senior journalist and writer. Views expressed in the above piece are personal and solely those of the author. They do not necessarily reflect Firstpost’s views.)