The Diminishing Returns on Innovation

Why every technological innovation is a trade with society

Would you rather give up your washing machine or your social media accounts? Your smartphone or antibiotics? I started asking myself these questions as a way to figure out which innovations actually matter - and realised I’d been thinking about things entirely the wrong way.

Humanity has been gifted by innovation. Today, we live more comfortably than kings did 200 years ago and have technology that increasingly feels like magic. And yet a 2025 Pew Research study found that “Americans are much more concerned than excited about the increased use of AI in daily life.” Europeans agree.

The increasingly heated debates about the benefits of technology circle the same question: are we gaining more than we risk losing?

The net social benefits of technological innovation

I have always been fiercely excited about technology, but over the past few years I have also noticed myself becoming more critical about its effect on society. Answering this question is, first and foremost, a selfish endeavour: a desire to earn back my own optimism through clearer thinking.

And so I conducted an admittedly imperfect analysis of the social costs and benefits of major technological innovations over time.

A few caveats before we dive in:

Firstly, the below analysis is based on my subjective qualitative assessment. It is neither claiming completeness nor perfect accuracy. Your assessment will likely differ, and I welcome any discussion about why that is.

Secondly, for each innovation, I qualitatively assessed whether benefits outweighed costs across three dimensions: economic (livelihoods, productivity), human (health, wellbeing), and environmental, defining net social benefit as total social benefit minus total social cost.

Lastly, whenever I found it difficult to assess which innovation had higher total benefit, I ranked them against each other by asking “what would I rather lose?” It’s a delightful exercise and makes for some great conversations over dinner.

The chart below lays out major innovations over time, with their social costs shown on the left and social benefits shown on the right.

A few innovations stand out given their exceptional cost-benefit ratio: writing, the printing press, smallpox vaccination, and penicillin all had major social benefits (e.g. the foundation of our modern world, dramatically reduced mortality), while incurring little social costs (limited to for example to side effects or allergies).

Industrial-era technologies (steam engine, railways, agricultural machinery, refrigeration, assembly lines) still score high on benefit, but their costs rise too. They cut the price of essentials, expanded mobility and markets, raised food supplies, reduced disease from spoilage, and lifted mass living standards, but their social costs rose through job displacement and deskilling, rapid urbanisation and family disruption, dangerous pollution and environmental damage.

Recent digital innovations cluster in more of a grey zone, with benefits that seem more incremental relative to their costs. Social media stands out as one of the few innovations where I would argue the total costs are roughly on par with the total benefits, if not higher.

With AI, we don’t yet know exactly where the cost and benefit bars of new innovations will lead us or what eventual form factor they will take. On the one hand, AI has the potential to create human benefit at the scale of previous massive breakthroughs, e.g. by saving hundreds of thousands of lives that are lost in traffic accidents each year through autonomous driving, or by significantly improving the health of each human by eradicating most diseases through AI-led medical research.

On the other hand, if AI models are deployed in ways that amplify addictive or harmful tendencies, such as AI-driven gambling, and if they trigger major labour disruption without credible ways to mitigate the downsides, we risk tipping the scales into the territory of questionable net social benefit.

The shape of progress

I have three main takeaways from this analysis:

First: The benefits outweigh the costs of almost all innovations. Given the reduction in poverty, disease and illiteracy that much of humanity has benefitted from, this is no surprise - in the past, innovation has indeed been great for humanity.

Second: The speed of innovation has increased over time. The gaps between major technological innovations collapsed from millennia to decades. Bibliometric work estimates modern scientific output is growing exponentially with an overall doubling time of ~17 years. Should the promise of AI-automated science become true, we may be able to speed up scientific progress even more, achieving in years what otherwise would have taken us decades.

Third: While the speed of innovation has increased, the net social benefit of new innovation has not increased over time. In fact, according to the above analysis I believe that it decreased, even though most innovations still have an overall net social benefit.

Making similar observations, the economist Tyler Cowen coined “The Great Stagnation” thesis in 2011: that fundamental progress, measured by economic growth, has slowed since the mid-20th century despite rapid advances in computing. Some research also suggests that while scientific output is growing, it is becoming less disruptive on average.

Why is that?

Why the social gains of innovation seem more marginal now

Some suggest that the aftermath of catastrophic innovations, such as the atomic bomb, and the growing awareness of technology’s role in climate change have fostered disillusionment. More people now question whether technology is still, on the whole, a force for good. The brightest minds may be less drawn to the most ambitious, high-impact technological work and as a result, innovation becomes more incremental.

Others argue that innovation may also have been slowed down by regulation. As a consequence, even if we wanted to, we can no longer aim for ambitious change, and instead need to optimise more incrementally. We have traded real progress away for safety.

There is also the “bits and atoms” thesis: it argues that progress has concentrated in software and information (bits), while advances in physical infrastructure, energy, and materials (atoms) that more directly shape the physical world have lagged because they are harder to build and come with longer R&D cycles. In the past, venture capital, which often seeks to maximise returns over short time horizons, has favoured easier software businesses over harder hardware and industrial ones, contributing to fewer breakthroughs in foundational “atoms” innovation.

Lastly, and in my view most importantly, there is the simple argument that much of the low-hanging fruit of progress is gone. The easiest, foundational discoveries have been made, and new breakthroughs now require far more complex and costly efforts.

The earliest waves of innovation reduced poverty, ill health, and drudgery spectacularly; later waves did so more moderately; and today, progress often seems incremental. Where past innovations saved many of us from terrible diseases and suffering, many current innovations seem to push against the upper rungs of Maslow’s hierarchy of needs.

At the same time, as innovations worked and our comforts grew, so did our entitlements. Once we adopt innovative technology, we adapt our baseline very quickly. As soon as something magical works, it just becomes technology. It is therefore easy to forget how gifted we are. We take the innovations that power our daily comforts for granted.

Not a gift but a trade

As the marginal social returns of innovation decrease, I have come to believe that rather than a gift to society, every major technological innovation is more akin to a trade with society.

In the past, innovation was a pretty obvious steal - vaccines, antibiotics, or the printing press. There was simply so much improvement to be made that most costs were well worth the massive benefits they offered in return.

But increasingly, some innovations look cheap at first, while we eventually have to pay for them dearly. We welcomed social media with open arms given the initial promise of connecting the world. We could not yet perceive the difficult consequences of algorithmic feed optimisation and incessant screen time.

If innovations don’t offer outsized benefits to most people anymore, it is difficult to make us accept even relatively small costs. We just question if it’s all really worth it.

Some of that scepticism is healthy: higher standards are a form of progress. We should ask more of innovation over time.

But some of it is less healthy: we have come to expect miracles without trade-offs, as if they were a birthright. This is why the public mood has shifted from gratitude to scrutiny. But innovation without risk is hardly possible.

AI might now be either the biggest bargain or the biggest rip-off in history.

A fork in the road

AI puts us at a fork in the road. During a period in which many innovations seem to deliver diminishing social returns, it could reopen high-benefit territory through breakthroughs in medicine, science, safety, and overall wellbeing. But if we squander its potential by optimising for low-value ends, such as addictive content or labour disruption without credible ways to absorb the downsides, we will accelerate the very trend of falling social returns that has fuelled today’s scepticism.

Despite technological innovation leading to net social benefits in the past, we are not owed good outcomes in the future. Optimism about innovation’s direction is not a given, but an active fight to make it a trade worth taking.

“You do not get what you want. You get what you negotiate.”

Harvey Mackay

As the AI community, we need to make AI’s benefits more legible by clearly measuring gains in health, safety, time, and learning.

We further need to price externalities honestly and early: companies need to more clearly communicate safety risks, attention harms, misinformation risks, and addiction dynamics to the broader public.

For all the abstract discussion of Universal Basic Income in the face of potential labour disruption, shareholders of AI companies need to make much clearer commitments to how they plan to share the upside of the economic value more broadly with society.

And lastly, we need to make virtues cool again: As an industry, we need to collectively push AI into high-benefit domains: medicine, science, safety, education, and care.

I still believe that being alive right now is miraculous. It is our duty to take the miracle seriously and instead of being entitled pessimists that are too lazy to change and blindly reject all innovation for fear of the costs we could incur.

At the same time, I am critical of the techno-utopian derangement syndrome that has befallen large parts of the technology industry, in which every question about trade-offs immediately gets waved off as doom. It isn’t helping the cause of progress because the public isn’t buying it. Rather, it belittles the instincts and rightful worries that humans have.

Is the outcome of progress under our control?

Back to our original question: are we gaining more than we risk losing? Today, too many of us still give the comfortable answer: “historically, things have been fine, so this will also be true in the future. Technology has always netted out positive, let’s relax and let the curve do its thing”.

My argument is that this logic no longer works because the curve is changing. The pace of innovation is rising while its social return is not. If the bargain is getting worse, then “historically, things were fine” should no longer be a reassurance to any of us.

AI is now forcing all of us to choose a philosophical camp: are we steering technological progress, or are we being steered by it?

Technological determinism suggests that innovations follow an inevitable path and reshape society in ways that are beyond human control. Just sit back and watch things play out.

Its opponent, social constructivism, argues that society, through individual and collective human choices and interactions, shapes technology’s design and use, viewing technology as an outcome of social forces, not an autonomous driver.

I advocate for aspirational social constructivism. That is the belief that technology has certain deterministic effects on society, as well as choosing to believe that it serves humanity. We form it, as much as it in turn forms us. This is the only way I believe we can live up to human freedom, rather than feeling like mere passengers on the journey towards our own destiny.

Technological innovation is not a gift, it is a trade. It comes with costs and with benefits. If we take our humanity seriously, we must do our best to make it a trade well worth taking.