Crossed Wires: What if the doomers are wrong and AI solves everything?

Here’s a thought. Let’s say all the AI doomers are wrong and the most optimistic optimists are right, even beyond their wildest imaginings.

Let’s say, as some very smart people do (like Alex Wissner-Gross and Peter Diamandis who host the Moonshots podcast), that just about everything is “solved” – maths, chemistry, biology, physics, energy, disease, education, even ageing.

Suppose we wake up in 2035 and a box of science fiction magic tools has been opened.

Fusion works. Cancer is cured. Materials are designed atom by atom. The physics is done. Food molecules can self-assemble into meals. The bugs in your genes can be fixed. Healthy lifespan can hit 120, maybe 150.

And let’s say all of this happens with nary a hiccup – no biological or computer viruses, no uncontrollable autonomous weapons, no environmental catastrophes, no societal breakdown.

What then?

That AI solves everything is, of course, an unreasonably heroic assumption, especially in the 10-year time frame. But the thought experiment reveals a much less discussed problem. An AI can discover a cancer drug. It cannot make the drug appear in a clinic in Soweto. In fact, the great coming AI wizardry cannot reach all of us with equal effort.

Why?

Because “us” is a species that is risk-averse, frequently lazy, reliably change-resistant, mostly not rich, often not technically literate and – without exception – sitting at the far end of a supply chain mired in the sludge of legacy systems, money, commerce and politics.

“Us” is on the couch watching the rugby, or cooking, or scrolling, or staring out of the window.

The AI may hand humanity the magic box of science fiction wonders. But it will not deliver it to our homes by free overnight courier, along with “how-to” instructions.

Instability diffusion

Economists have a dull word for this – diffusion – and a deep body of evidence. Diego Comin and Bart Hobijn tracked 15 technologies across 166 countries over two centuries, from railways and the telegraph to steel and the PC.

On average, countries adopted a technology 45 years after it was invented.

The good news is that the lag has collapsed. The bad news arrived in a follow-up by Comin and Martí Mestieri. Technologies now reach poor countries almost as quickly as rich ones (think cellphones). But once there, they penetrate far less deeply – fewer users, thinner use, less of the economy rewired around them.

Together, these two shifts – converging arrival, diverging penetration – account for some 80% of what they call the “Great Income Divergence” since 1820.

Proponents of the extreme optimism side of the narrative talk often about “abundance”, a utopian vision of everyone on the planet wanting for nothing and enjoying the benefits of all of this. So, let’s stay with the thought experiment and zero in on robots.

Suppose AI and robotics really do create astonishing productivity – automata that can perform any physical task, from the dexterous to the boring, from folding laundry to building ships. Who owns the models, data centres, robot fleets, factories, the batteries, the chips and the intellectual property producing it?

If much of it belongs to a relatively small group of companies and investors, AI could simultaneously make society less burdened by physical drudgery and its ownership vastly more concentrated. These are not contradictory outcomes. They are, in fact, natural companions.

Some things will also remain stubbornly scarce. There is only so much land in Manhattan, London or Cape Town. There are only so many houses overlooking the sea, places at desirable universities and hectares in pleasant climates. Only some people can run fast or learn piano well.

If AI makes manufactured goods, intelligence, entertainment and energy dramatically cheaper, the value of genuinely scarce assets may simply rise. Abundance in one part of the economy does not abolish scarcity elsewhere.

A question of time

There is another problem that receives even less attention – time.

Suppose an AI-designed longevity treatment arrives in 2032 and costs $500,000. The wealthy buy it. Ten years later, manufacturing improves and the price falls to $5,000. Eventually governments provide it free. This looks like technological diffusion working exactly as advertised.

Unfortunately, the wealthy customer bought something that cannot be redistributed later – another decade of healthy life. They got it early; others expired while waiting.

The same compounding effect could apply to education, cognitive enhancement, personalised medicine, entrepreneurship and wealth management. Someone with five powerful AI agents working continuously on his investments, business and health is not simply consuming a better product.

He is acquiring tools that help him accumulate further advantage. By the time those tools reach everyone else, the early adopters may have moved another considerable distance ahead.

AI inequality could therefore be different from the inequality created by previous technologies. It may be self-amplifying.

Even if the optimists are right that AI could solve the hard technical problems, they are wrong (or at least silent) about the harder political and economic problem of distribution.

If we do nothing, AI will not create abundance for all. It will favour those with the capital to own the models, the compute to run them, the money to buy them, the skills and inclination to use them, and the political power to shape the rules. The poor township dweller, the subsistence farmer, the garment worker – they will be at the end of a supply chain waiting for a box of magic tools that never arrives, or that they do not even know exists.

Ones and zeros

So is it as binary as it seems, the AI-haves and the AI-have-nots? Not quite.

The floor will rise, possibly dramatically. A woman in Soweto with a cheap Android phone and a decent model will have better medical triage than a Harley Street patient had a decade ago. That is not nothing. It may be the most egalitarian thing a technology has ever done.

But the ceiling will rise faster. Those with means will rush to the front of the queue and get the whole magic box. The township will get the chatbot 1.1 version. This is the Comin-Mestieri pattern exactly – convergent arrival, divergent penetration. Everyone gets some magic. Only some get more of it.

The optimists may be right about solving science, but the hard part is the last mile, and the last mile has always been paved with the messy reality of politics, human frailty and money.

That is a much harder problem to solve, if it is solvable at all. DM

Steven Boykey Sidley is a professor of practice (ex-JBS, University of Johannesburg), a partner at Bridge Capital and a columnist-at-large at Daily Maverick, where he writes the weekly Crossed Wires column. His new book, Checkmate: How Shoprite Checkers rewrote the rules and won, is published by Jonathan Ball.