A poverty of attention

Attention as capital in an economy of agents

In 1971 Herbert Simon wrote that "a wealth of information creates a poverty of attention". He was no technophobe. Fifteen years earlier, with Allen Newell, he had written one of the first artificial intelligence programs, and he went on to collect a Turing Award and a Nobel Prize in economics. His point was plain enough. Information consumes the attention of whoever receives it, so the more information there is, the scarcer attention becomes.

Fifty-five years later, the industry Simon helped to found has made information almost free to produce. You still get the same sixteen waking hours your grandparents had. Far more people are bidding for them, and a growing share of the bidders are software agents working on someone else's behalf.

I've spent twenty-five years building products, and a few of them did their share of the bidding. The best defence I know comes from investing: treat attention as capital, the way Benjamin Graham, Warren Buffett and Charlie Munger treated money.

The partner who rings every morning

Graham's best invention was Mr Market, a business partner who turns up daily offering to buy your share or sell you his, at a price set by his mood. You deal with him when his price suits you and otherwise leave him standing in the hall.

Mr Market now has a product team and an engagement target, and he lives in your pocket. Answering him has a measurable cost. In 2000 Brad Barber and Terrance Odean studied 66,465 American households trading through a discount broker. The ones that traded most earned 11.4 per cent a year while the market returned 17.9. The difference went in fees and bad timing. I can't put a number on the career equivalent, though in the careers I've watched, the people who move every time a recruiter rings tend to reach fifty with very little that has compounded.

Three investors, three stages

Graham, Buffett and Munger each added one idea to the one before. Read in order, they map onto a working life.

Graham: protect the downside

Graham lost heavily in the crash of 1929 and spent the rest of his career avoiding a repeat. He worked out what a thing was worth before looking at the price, and bought only at a deep discount to that figure, so his mistakes had somewhere soft to land. He called the discount a margin of safety.

In practice: write one sentence each for career, home, family and money, saying what each is for, and price every new offer against it. Leave a fifth of the week unbooked, because the slack is your margin.

Buffett: fewer, bigger bets

Buffett began on Graham's cheap stocks, then moved to holding a few good businesses for decades. He told students to imagine a punch card with twenty slots for every investment they would make in their lives. By his own account his worst purchase was Berkshire Hathaway itself, a textile mill he bought in the 1960s because it was cheap. Cloth was a commodity, and the business stayed cheap.

In practice: treat a job move, a house or a partnership as one punch on the card. Then ask which of your skills would survive becoming free. I built websites in Perl in the late 1990s, when people paid properly for the work. Code, copy and first-pass analysis are now going the way of cloth.

Munger: avoid stupidity

Munger talked Buffett out of cheap stocks and into quality. His method was inversion: instead of asking how to have a good life, ask what would guarantee a miserable one, and avoid it. He told the story of a rustic who said, "I wish I knew where I was going to die, and then I'd never go there." His 1986 list included envy, resentment, unreliability and learning only from your own experience.

In practice: write down what would wreck your marriage, your health or your reputation, and arrange the week so it doesn't happen. When someone offers you something, work out what they are paid to want.

What agents do to Simon's sentence

AI works on both halves of Simon's line. It adds to the wealth, because a model will write ten reports where there used to be one. It also offers to relieve the poverty, by summarising them into a paragraph. History suggests the first effect usually wins: in 1987 Robert Solow remarked that you could see the computer age everywhere except in the productivity statistics. I'd expect AI to repeat the pattern, with low confidence, because some of the narrow gains are real and measured.

Simon anticipated the fix in the same 1971 essay. A new information system earns its place only if it absorbs more information than it produces, listening and thinking more than it speaks. A model that drafts forty emails a day adds to everyone's poverty. The one that reads two hundred and tells you which three need you is the one worth having.

Agents raise the stakes. An agent sends the email, books the meeting and chases the reply without anyone deciding to, so the volume aimed at you is no longer limited by anyone's working day. The sensible defence is an agent of your own at the door, and choosing it is a capital decision. Ask Munger's question first: what is it paid to want? An assistant built by a company that earns from your engagement is Mr Market employed as your butler.

Cheap production moves value elsewhere. Rory Sutherland has long argued that people value what is costly to fake. A generated email costs nothing, so the phone call and the handwritten note are worth more than they were. At work, the expensive-to-fake material is judgement, trust, taste and a long memory for one domain, and all of it takes years to build.

The same habit at company scale

Many of the boards and founders I advise have spent the past two years day trading AI. A pilot starts, the next model arrives before it finishes, and the board pack ends up listing a great many experiments and no capability that compounds. Each pilot was cheap. Together they used up the attention of the few people able to judge whether any of it worked.

The method transfers with little adjustment. Write the sentence for each capability: what it is for, and what good looks like in three years. Keep a person holding the judgement on anything that can't be cheaply undone. Make a handful of agent bets and hold them long enough to compound. Build the advantage on your own data, processes and people, since any particular model is heading the way of the textile mill. And run the inversion. A failed AI programme reliably follows from buying because a competitor did, measuring activity instead of outcomes, and never asking what the vendor is paid to want.

At home

When nearly everything can be generated, undivided attention is the gift that can't be. A meal with the phones in another room is a costly signal in Sutherland's sense, and children read those signals more accurately than any client does. Keep doing some of your own thinking and writing, too, so you can still tell when the machine is wrong.

I should own up that my working life looks more like a seed fund than a Graham portfolio: several small bets, each with a cost cap and a date by which it earns its keep or gets shut. That suits ventures. Applied to a marriage it would be a catastrophe. The skill is knowing which parts of your life belong in the venture book and which in the value book.

The audit I'd suggest takes ten minutes on a Sunday evening. Make two lists: what you chose to give your attention to this week, and what took it without asking. In a reasonable week the first list is the longer one.