An Expert on Entrepreneurship Tackles Health Care Innovation

October 7, 2024

For 35 years, Amar Bhidé taught entrepreneurship—at Harvard, Chicago, Tufts, and Columbia, where he was Lawrence D. Glaubinger Professor of Business. He has written dozens of case studies, synopses of real-world scenarios crafted to spur vigorous classroom conversation, books on entrepreneurship, innovation, and the financial system, and op-eds on public policy issues for the Wall Street Journal, the Financial Times, and The New York Times.

In recent years, he’s increasingly delved into foundational questions about the complex, dynamic advances in productive knowledge. “It’s not just science” he quips. “The steam engine did far more for the laws of thermodynamics than laws of thermodynamics did for the steam engine.”

In January 2024, Bhidé accepted an appointment as a professor in Columbia Mailman’s Department of Health Policy and Management. He teaches the course “Lessons from Transformational Advances,” which digs into a series of case histories Bhidé developed to probe the complex, protracted processes that produced life-altering drugs, devices, and practices.

One case describes how despite a long history and contemporary clinical promise—the widespread use of fecal microbiota transplant to treat gastrointestinal disease has been stymied by regulatory hurdles and provider resistance. The case on tamoxifen shows how tamoxifen became a gold-standard treatment for breast cancer—after failing as a contraceptive.

The overarching goal is to inspire, not just inform students about how new treatments and practices evolve. “The cases show how contributing to progress offers great scope for personal flourishing, whatever your role and whatever your financial reward may turn out to be,” says Bhidé.

Is there a core theme in your work?

Bhidé: I’ve gone from looking at things principally from a businessperson’s, an entrepreneur’s point of view, to trying to understand the overall process of how productive knowledge advances. But the core theme has been the human striving for change and betterment that cannot be reduced to an algorithmic formula.

How did you make the pivot to advances in medicine?

Bhidé: As it happens my mother was a pioneering cancer researcher, and my sister is an oncologist. But, with my general interest in productive knowledge, I could have written about anything—advances in computer science. I didn’t. I wrote about medical innovations. This was lucky. Health care is a broad arena but nonetheless has some common features.

What do you hope your students take from the case studies?

Bhidé: The process of practical advances is complicated, protracted, and involves a large cast of characters. There is instrumental and humanistic value in appreciating these processes: We could do things better in the future if we understand how past advances come about. They also teach us what makes us human.

In December, Oxford University Press will publish your fifth sole-authored book, Uncertainty and Enterprise: Venturing Beyond the Known(link is external and opens in a new window). How did it come about?

Bhidé: The book represents the culmination and synthesis of much of my writing and research. There are also many points of overlap with the seminar on transformational advances I’m currently teaching. The case studies have informed the book and the ideas that I’ve tried to distill in the book have informed how I’m teaching the course.

What are the foundational principles of Uncertainty and Enterprise?

Bhidé: We cannot or should not be sure of anything. We cannot be sure of what is or what was, and even less what could be or what should be. We can have only conjectures, provisional hypotheses that combine imagination and evidence. And inevitably, our conjectures diverge while much of our actions are interactive. We can’t act unilaterally. Imaginative yet grounded discourse plays a crucial role in aligning our conjectures.

Who is your target audience?

Bhidé: I want to persuade mainstream economists that there’s a broader way of looking at the world that—if they adopted it—could be beneficial to themselves and to society. A second target is the intellectually curious, possibly “highbrow,” general reader about the rewards, challenges, and reasonable ways of dealing with uncertainty that is so central to our lives, yet are often ignored in economics and decision theory. I don’t however want to pick a fight with mainstream economics or provide cookbook recipes to general readers.

This fall, Project Syndicate published your op-ed calling large learning models “mendacious talking horses(link is external and opens in a new window).” Another for Barron’s(link is external and opens in a new window) calls out the current AI investment craze a mania. What sparked your ire?

Bhidé: Writing my Uncertainty book has a lot to do with it. I tried to use LLMs to research, edit, and illustrate the book—it was a source of unending frustration, though “earlier” AI was invaluable. I also studied the evolution of AI for my book. AI grew out of a “fork” in the cognitive revolution of the 1950s and 1960s which conceived of the mind as a computer, often relying on statistical models to recognize patterns. A second fork treated the mind as a “meaning constructor” where meaning was highly contextual, historical, and cultural. Both forks have value.

Long before LLMs, statistical AI had proven its worth in many applications. But reducing all thought and speech to a mindless statistical model is absurd. Yet that’s what many LLMs try do. The LLM mania also ignores the protracted trial and error through which cost-effective AI applications have emerged over the last 70 years. The mania also shows how ignorance of how transformational technologies like AI evolve can become a social menace.

Suppose someone said that smartphones were on the cusp of generating widespread transformations.

You might reasonably ask, “Where have you been these last twenty years, Rip Van Winkle?”

Smartphone apps like Uber and Airbnb have revolutionized transport and travel. Mobile search and social media have crushed mainstream media and advertising.

Given how far we have already come, is it likely that smartphones are at an inflection point? Similarly, with AI. Its applications have already been transformational. Indeed, it is AI tools and techniques that make smartphones smart. Nearly every smartphone app – from texting to sexting, mapping to matchmaking, video editing to streaming, Uber ridesharing to Airbnb rentals – incorporates AI. When we speak to our phones asking for weather forecasts or driving directions, we engage AI’s Natural Language Processing capabilities.

Moreover, AI’s widespread use precedes and goes far beyond smartphones. A 1956 workshop at Dartmouth kicked off academic AI research. In the following decades, practical applications evolved. Starting in the 1970s, George Lucas’s Star Wars epics dazzled audiences with AI special effects and animations. ‘Fuzzy logic’ proposed by UC Berkeley’s AI guru, Lotfi Zadeh, in 1965, was used to control a Japanese subway in 1987. By 1990, Japanese consumer electronics companies were using fuzzy logic in camcorders, vacuum cleaners, room heaters, and air-conditioners.

In 2006 – a year before Apple’s iPhone – Oxford’s Nick Bostrom noted that cutting-edge AI had “filtered into general applications, often without being called AI because once something becomes useful enough and common enough it’s not labelled AI anymore.”

Sixteen years later, the claim that AI has just reached a take-off stage. is perplexing. Merely maintaining historical growth rates from a high base should be a challenge.

Looking more closely at how AI became mainstream is instructive.

Traditional pre-AI software applications performed deterministic calculations. Payroll processing and optimizing complex operations were archetypal applications.

More often than not, however, uncertainties frustrate demonstrably correct solutions. Ambiguous information or incomplete knowledge makes calculating what’s truly best impossible. We must make do with guesses and approximations. Likewise, we often don’t use numbers or algebraic symbols to specify problems or discuss solutions. From everyday speech to Supreme Court deliberations, our discourse relies on ambiguous language – including analogies and metaphors.

Lofti’s 1965 “fuzzy logic” and natural language programming thus epitomize the more realistic aspirations of AI.

But how to combine the digital computer’s capacity to flawlessly manipulate 1s and 0s with the incompleteness and imprecision of human knowledge and discourse?

One early approach incorporated specialized expertise. Medical rules of thumb were a popular basis for the early expert systems. But, this approach was limited to problems where experts had codifiable knowledge.

 Other AI applications that used statistical approximations. Humans merely specified the data – text, images, not just numbers — from which computers inferred statistical patterns. No understanding of the underlying process or consideration of contextual meaning was necessary. The dictum, repeated endlessly in elementary statistics classes, that “correlation is not cause” was brushed aside. AI programs did not even have to be told which variables mattered or to what degree. They used data mining to calculate variable weights that best fit the observations.

AI programs used statistical correlations to mimic natural language. Actual natural language often requires reading minds — contextual interpretation of intent. The meaning of a simple ‘what!’ depends on context and tone. Going back to MIT’s Eliza, a 1960s-era psychotherapeutic chatterbot, AI programs used correlations as substitutes for any mindreading.

Statistical AI could also improve through trial and error. But again, this ‘machine learning’ did not require domain expertise, judgments about “lessons learned,” or understanding or consideration of context.

Nonetheless, the cost-effectiveness of statistical AI that did not require specialized expertise vastly broadened the scope of AI applications. Google’s search algorithm, which handily outperformed Yahoo’s human catalogers of the internet, was a striking example.

At the same time, AI hasn’t sailed smoothly in every sea. Belying dire predictions, AI did not dominate or displace human’ knowledge work.’ Knowledge-intensive jobs grew, and wages stayed high.

AI even failed to automate many tasks that don’t require much thinking or training. Going back to Apple’s much-ridiculed 1993 Newton, handwriting was supposed to replace typing. In 2001, Bill Gates predicted that pen-based tablets would become “the most popular form of PC sold in America” in five years. They didn’t come close. Now, finally, convertible PCs with pens and touch screens have found a market, but keyboards remain the dominant input device. AI-enabled handwriting and voice recognition remain frustratingly hit or miss. Similarly, we usually still prefer the precision and accuracy of clicking or tapping on a button to giving voice instructions to personal assistants (like Siri or Alexa).

Where has the accuracy of statistical AI been acceptable, and where has it not?

Accuracy often depends on the ambiguity of inputs and outputs. Printed words that use standard fonts are less ambiguous than idiosyncratically handwritten words. Unsurprisingly, Optical Character Recognition software scans printed books and documents far more accurately than handwriting recognition programs.

Ambiguous outputs similarly undermine machine learning. Unquestionably correct or wrong results have helped make face recognition highly accurate. In contrast, correctly deciphering spoken words (“there” or “their”?) requires knowing the speaker’s intent. But, statistical correlations cannot reliably discover intent just as they cannot establish cause.

Accuracy also depends on the stability and uniformity of the process that generates the data used by AI applications. Physical or physiological processes, governed by invariant laws of nature, are usually stable. In contrast, human behavior and choices are subject to the whimsical vagaries of social attitudes and the zeitgeist. Statistical predictions about creditworthiness or purchasing behavior can, therefore, be highly inaccurate.

Data produced by a uniform process provides a more reliable basis for statistical inference. For example, OCR algorithms scan text more accurately if trained with materials in the same language and script. Conversely, data shaped in diverse ways by different contextual factors – if the observations are likeunhappy families unhappy in their own way – can make statistical inferences practically useless.

Acceptable accuracy depends on the cost of mistakes — the stakes — and the price-performance of the alternatives. Nearly every ad that Google and Meta Platforms throw at me is utterly remote from my interests. But the stakes are low and even the wildly inaccurate targeting of algorithmic advertising beats the alternative of blind advertising.

In some creative applications of AI accuracy can be both unknowable and irrelevant. There are no correct special effects in Star Wars movies or animations in video games and cartoons. There is no objective benchmark for restoring old movie prints — who knows what the original looked like? But, automated AI restoration wins because it is much cheaper and faster than human restoration.

Turning to the current AI mania.

Ignorance of AI’s seven-decade history may explain some over-the-top predictions about its future. But even some savvy techies who are aware of what came before assert that Large Language Models – often now conflated with all of AI – are game changers. A veteran software entrepreneur believes AI is still in its “early infancy.” He argues that “earlier incarnations, such as protein folding and chess playing, were esoteric and of little relevance to the general public. The chat interface to LLMs has suddenly made AI accessible to the wider public. New ideas and applications are exploding. The real creativity is coming from people using it and suggesting new uses, rather than from the engineers creating it.”

 I believe it is fair to say that before LLMs, most people were passive consumers, often unaware of the AI in their mobile phones, search engines, and social media. Certainly, LLMs have an arresting capacity for seemingly intelligent, natural language conversations with non-technical users, and they offer to automate several analytical and creative tasks. Could these abilities make LLMs a “killer app” for AI to an even greater degree than the AI that has long been embedded in smartphones?

The analogy with spreadsheets is seductive. Spreadsheets had simple user interfaces that allowed people with limited technical expertise to build useful programs. Running on cheap personal computers, they offered compelling value in many applications that did not require the power of mainframes. Symbiotically, they helped expand the personal computer market, prompting investments in better computers.

LLMs have even simpler and more natural user interfaces than spreadsheets. Yet underneath their hoods, LLMs run statistical engines with the same statistical issues that delineated the practical scope of earlier AI applications. As with earlier AI, LLMs can shine in creative applications, such as image generation, where accuracy is irrelevant. Conversely, as with other statistical AI models, ambiguous inputs and outcomes derail their reliability and limit self-corrective learning. They can trip over data that is not generated by a stable process or is highly dependent on context.

Relying on statistical correlations rather than deductive logic or math, LLMs have offered bizarre solutions to reasoning problems, highlighting, for example, the risks of being attacked by a cabbage while rowing across a river. The Khan Academy’s AI tutor for kids, struggles with elementary math. (It miscalculated subtraction problems such as 343 minus 17, couldn’t consistently round answers or calculate square roots, and typically didn’t correct mistakes when asked to double-check its solutions.)

Throwing every possible kind of data into LLMs’ training pots does not improve accuracy and reliability. Medical data does not make responses to legal or engineering questions any better. Training on Swahili literature does not sharpen statistical summaries of Shakespeare’s plays. Bulking up LLMs with disparate data so that LLMs can answer every question under the sun may increase their propensity to fantasize or hallucinate.

Spreadsheets, in contrast, didn’t overpromise and underdeliver. They didn’t tell jokes or write essays, but for their more targeted functions, they followed the user’s instructions precisely and correctly.

The chatty user-friendliness of LLMs isn’t a free lunch. It may well be a significant limitation. Yes, users need less knowledge of input rules and conventions than in their interactions with a spreadsheet, traditional search engine, or photo editor. But free-form inputs are also more ambiguous. Natural language prompts are more likely to evoke inaccurate or useless responses than traditional keyword searches.

In low-risk uses people will tolerate LLM mistakes for convenience as they do with autocomplete howlers in their text messages. The multi-trillion-dollar question is whether the benefits from low-stakes uses can  cover the costs.

One important reason for the nearly immediate popularity of spreadsheets (besides their ease of use) was that they ran on personal computers and not expensive mainframes. Similarly, Uber and Airbnb apps provided cheap, reliable alternatives to taxis and hotels through smartphones that users already owned. In contrast, LLMs require users to purchase more expensive hardware. Moreover, user hardware accounts for a fraction of the costs of building, training, and operating LLMs. For now, and as in the 1999 internet bubble, manic investors are willing to subsidize uneconomic uses. What happens when the music stops?

At best, LLMs are akin to a new high-powered automobile engine that can win car races but makes too much noise and guzzles too much gas for street use. The hype notwithstanding, LLMs aren’t like Nikola Tesla’s alternating current inventions that drastically changed the economics of electrification. Why then gamble on the transformative acceleration of AI and ignore so many other possibilities for innovation and operational improvements the world offers?

After nearly two years of focusing on book writing, I returned to an oped, to eject a bee that had been buzzing in bonnet. Published in Project Syndicate, the text is below..

The Boring Truth About AI

To think that artificial intelligence is advancing at warp speed and creating existential risks to humanity is to confuse a mania with useful progress. The technology is less like nuclear weapons than like many other slowly evolving technologies that have come before, from telephony to vaccines.

Experts who warn that artificial intelligence poses catastrophic risks on par with nuclear annihilation ignore the gradual, diffused nature of technological development. As I argued in my 2008 book, The Venturesome Economy, transformative technologies – from steam engines, airplanes, computers, mobile telephony, and the internet to antibiotics and mRNA vaccines – evolve through a protracted, massively multiplayer game that defies top-down command and control.

Joseph Schumpeter’s “gales of creative destruction” and more recent theories trumpeting disruptive breakthroughs are misleading. As economic historian Nathan Rosenberg and many others have shown, transformative technologies do not suddenly appear out of the blue. Instead, meaningful advances require discovering and gradually overcoming many unanticipated problems.

New technologies introduce new risks. Invariably, military applications develop alongside commercial and civilian uses. Airplanes and motorized ground vehicles have been deployed in conflicts since World War I, and personal computers and mobile communication are indispensable for modern warfare. Yet life goes on. Technologically advanced societies have developed legal, political, and law-enforcement mechanisms to contain the conflicts and criminality that technological advances enable. Case-by-case court judgments are crucial in the United States and other common-law countries. These mechanisms – like the technologies themselves – are evolutionary and adaptive. They produce pragmatic solutions, not visionary constructs.

The Manhattan Project, which developed the atomic bomb and helped end World War II, was an exception. It had a high-priority military mandate. With the Nazis seeking to develop a bomb of their own, speed and effective leadership were essential. And as all-out thermonuclear war became a real threat, statecraft and strategic deterrence helped avert doomsday. 

But nuclear weapons are a misleading analogy for AI, which has followed the typically diffused, halting pattern of most other technological transformations. AI spans disparate techniques – such as machine learning, pattern recognition, and natural language processing – and has wide-ranging applications. Their common feature is mainly aspirational – to go beyond mere calculation to more speculative yet useful inferences and interpretations.

Unlike the Manhattan Project, which proceeded at breakneck speed, AI developers have been at work for more than seven decades, quietly inserting AI into everything from digital cameras and scanners to smartphones, automatic-braking and fuel-injection systems in cars, special effects in movies, Google searches, digital communications, and social-media platforms. And, as with other technological advances, AI has long been put to military and criminal uses.

Yet AI advances have been gradual and uncertain. IBM’s Deep Blue famously beat world chess champion Garry Kasparov in 1997 – 40 years after an IBM researcher first wrote a chess-playing program. And though Deep Blue’s successor, Watson, won $1 million by beating the reigning Jeopardy! champions in 2011, it was a commercial failure. In 2022, IBM sold off Watson Health for a fraction of the billions it had invested. Microsoft’s intelligent assistant, Clippy, became an object of ridicule. And after years of development, autocompleted texts continue to produce embarrassing results.

Machine learning – essentially a souped-up statistical procedure that many AI programs depend on – requires reliable feedback. But good feedback demands unambiguous outcomes produced by a stable process. Ambiguous human intentions, impulsiveness, and creativity undermine statistical learning and thus limit the useful scope of AI. While AI software flawlessly recognizes my face at airports, it cannot accurately comprehend the nuances of my carefully and slowly spoken words. The inaccuracy of 16 generations of professional dictation software (I bought the first in 1997) has repeatedly frustrated me.

Large language models (LLMs), which have become the public face of AI, are not technological discontinuities that magically transcend the limitations of machine learning. Claims that AI is advancing at warp speed confuse a mania with useful progress. I became an enthusiastic user of AI-enabled search back in the 1990s. I thus had high hopes when I signed up for ChatGPT’s public beta in December 2022. But my hopes that it, or some other LLM, would help with a book I was writing were dashed. While the LLMs responded in comprehensible sentences to questions posed in natural language, their convincing-sounding answers were often make-believe.

Thus, whereas I found my 1990s Google searches to be invaluable timesavers, checking the accuracy of LLM responses made them productivity killers. Relying on them to help edit and illustrate my manuscript was also a waste of time. These experiences make me shudder to think about the buggy LLM-generated software being unleashed on the world. That said, LLM fantasies may be valuable adjuncts for storytelling and other entertainment products. Perhaps LLM chatbots can increase profits by providing cheap, if maddening, customer service. Someday, a breakthrough may dramatically increase the technology’s useful scope. For now, though, these oft-mendacious talking horses warrant neither euphoria nor panic about “existential risks to humanity.” Best keep calm and let the traditional decentralized evolution of technology, laws, and regulations carry on.