When Intelligence Becomes Cheap

For most of economic history, we have worried about scarcity. Land was scarce. Capital was scarce. Oil was scarce. Skilled labour was scarce. Even information, until quite recently, was expensive to obtain.

Artificial intelligence presents us with a rather different economic question: what happens when a certain kind of intelligence itself becomes cheap?

I don’t mean human intelligence in its fullest sense. AI doesn’t wake in the morning worrying about its children, regretting yesterday’s decisions or wondering whether another glass of wine might improve matters. Human beings remain complicated creatures.

But much of what we are paid to do is considerably less mysterious. We read documents, analyse numbers, write reports, search for information, compare alternatives, produce computer code and make reasonably informed decisions.

AI is becoming remarkably good at many of these things.

The steam engine made physical power cheaper. Electricity distributed that power. Computers made calculation cheap. The internet made information cheap. AI may now make large parts of intellectual work cheap.

For investors, the interesting question is what happens next.

When the Price Falls, Demand May Rise

When something becomes cheaper, we don’t necessarily use less of it. Often we use much more.

William Stanley Jevons noticed this in the nineteenth century with coal. Improvements in the efficiency of steam engines didn’t reduce Britain’s consumption of coal. They made steam power economical for more purposes, and consumption increased.

I wonder whether we are about to see something similar with intelligence.

If a piece of analysis costs one-tenth of what it used to cost, a company may not simply produce the same ten reports more cheaply. It may produce a hundred reports that previously weren’t worth producing at all.

A small company that could never afford an economist, programmer, translator, research assistant and designer may increasingly have access to some approximation of all five. A doctor may compare a patient’s history against an enormous body of medical evidence. A fund manager can interrogate mountains of economic and market data before breakfast.

This makes the employment question more complicated than it first appears.

If one analyst can do what previously required three, there will plainly be pressure on some jobs. But if the cost of analysis collapses, businesses may also demand vastly more analysis. AI can remove labour from a particular task while making that task cheap enough for the economy to consume much more of it.

Computers did something similar. They made calculation extraordinarily cheap, yet we didn’t consequently decide that we had enough calculations. We put computers into cars, watches, telephones and almost everything else.

The same happened with spreadsheets. They eliminated an extraordinary amount of clerical work while helping to create quantities of financial analysis that would have been unimaginable when accountants worked with ledgers and pencils.

So the employment question isn’t simply how many people AI can replace. It is also what becomes economically possible when intelligence costs very little.

There is an important qualification, however.

Jevons tells us that cheaper intelligence may produce much more intellectual output. It doesn’t tell us whether that output is worth producing.

A hundred additional analyses that improve decisions may represent genuine productivity. A hundred reports that nobody needs are merely cheaper noise.

That distinction matters.

Where Does the Money Go?

For investors, this is the question I find more interesting.

A technology can transform the world without every investor in that technology becoming rich.

Railways transformed the nineteenth-century economy. Railway investors didn’t all make fortunes. The internet changed commerce, communication and media, but much of its eventual economic benefit flowed to businesses and consumers far beyond the companies that built the original infrastructure.

AI could follow the same pattern.

Perhaps the owners of the leading models capture most of the economics. Perhaps semiconductor manufacturers do. Perhaps data centres, power producers and other infrastructure providers continue to benefit from the enormous investment required to build the system.

Or perhaps competition steadily reduces the price of intelligence until much of the economic benefit migrates downstream, to companies that use AI rather than companies that make it.

This distinction matters enormously to investors.

The technology can be completely real and the investment thesis can still be wrong.

Markets have confused those two things before.

My own inclination is that much of the lasting economic value will eventually migrate downstream. Infrastructure providers may earn extraordinary returns during the build-out, but if intelligence becomes cheaper and widely available, the greater long-term prize may belong to businesses that learn to use it to do something genuinely valuable.

That doesn’t mean the infrastructure investment is wrong. It means we should distinguish between the economics of building AI and the economics of living with cheap AI.

They may produce quite different winners.

The Productivity Question

Then there is productivity.

It seems obvious that AI should increase it. If somebody produces a report in five minutes rather than five hours, something has plainly become more efficient.

But economic history gives us reason to be patient.

Robert Solow famously observed in 1987 that the computer age could be seen everywhere except in the productivity statistics. Computers were already transforming offices and businesses, yet the expected economy-wide productivity gains were proving strangely difficult to find.

AI may give us another version of the same problem.

Buying technology is easy. Reorganising a business around it is harder.

A company can give every employee access to AI tomorrow. That doesn’t mean it has changed how decisions are made, which jobs are necessary, how products are designed or where capital should be allocated.

And if AI merely allows an organisation to produce sixty reports nobody reads instead of six, we have certainly increased output. Whether we have increased economic value is another matter.

The real productivity gain will come when companies begin doing things they couldn’t economically do before, rather than simply doing old things faster.

That may take longer than markets expect.

What Remains Scarce?

If intelligence becomes abundant, the natural question is what remains scarce.

The fashionable answer is judgement. I’m not entirely persuaded.

Judgement is cheap too. Everyone has a view.

What isn’t cheap is having to stand behind it.

An AI can tell me to buy a stock. It can produce an impressive analysis explaining why, and it may eventually be better at parts of that analysis than I am.

But when the stock is down 30 per cent, the AI isn’t sitting across the table from my client explaining what went wrong.

I am.

The same problem will arise in medicine, law, engineering and business. Machines will increasingly give us advice, and sometimes their advice will probably be better than ours.

But somebody still makes the decision and somebody has to own it afterwards.

As intelligence becomes cheaper, accountability may become more valuable.

For investment management in particular, that distinction matters. Information is already abundant. Analysis is rapidly becoming abundant. Producing another beautifully written investment case will become progressively less impressive.

Knowing what matters, committing capital accordingly and accepting responsibility when the decision is wrong are rather different things.

Following the Profit

There is one final economic question.

If AI produces large productivity gains, somebody receives the surplus.

Workers may capture some through higher wages. Consumers may receive it through lower prices and better products. Companies using AI effectively may retain some as higher margins. And some will accrue to the owners of models, chips, data centres, energy infrastructure and capital.

The proportions are unknowable today. But the direction of travel matters enormously for asset allocation.

We are still early enough in this transformation that predictions are cheap too. AI will abolish work. AI will create unimaginable prosperity. AI will end poverty. AI will make inequality unbearable.

There seems to be a forecast available for every temperament.

For investors, I think we can start with something simpler.

Intelligence is becoming cheaper.

When the price of an important economic input falls dramatically, people use more of it. Businesses reorganise around it. Old profit pools shrink and new ones appear. Some jobs disappear while other activities suddenly become economical.

The investment opportunity therefore isn’t merely to identify the companies producing artificial intelligence.

It is to follow where the economic value goes after intelligence becomes cheap.

That, I suspect, is where some of the more interesting investments of the next decade will be found.