We Were Never Alone

Recently, while looking through old adverts, I came across the Auntie Ayşe of my generation. There was a time when marketing’s biggest question was “Who is your target audience?” Seeing that advert, a different question came to mind: how did we use to make decisions?

When we were children, Auntie Ayşe’s opinion could sometimes be worth more than an expert’s. Which doctor to see, which plot of land to buy, which medicine works, who should marry whom, where to shop… Our decisions were surrounded by an invisible advisory board. We did not call it a “decision system”; we called it family, the neighbourhood, friends and relatives.

When I started this piece, the sentence in my head was “We are no longer alone when we make decisions.” As I wrote, I realised my opening sentence was probably wrong. Perhaps people have never wanted to make decisions alone.

One reason may be that we sense our own view is not enough. Someone else may have seen something we did not, heard a story we never heard, or know of a possibility we do not. But I don’t think that is the only reason. Sometimes we consult someone to see better, and sometimes so as not to carry the weight of a decision alone, to spread the responsibility a little. When we said “I asked her too”, “That’s what the doctor said” or “Everyone said that plot was a good one”, we needed not only information but also someone to share the weight of the decision.

I think these two desires have always walked side by side in the history of human decision-making: being able to seek counsel, and spreading the responsibility.

Machines have long helped us with the second. The thermostat is a fine symbol of this. A person sets the target: 22 degrees. The system monitors, compares and acts when needed; no one has to check the room every minute.

More than a hundred years ago, the mathematician and philosopher Alfred North Whitehead observed that civilisation advances by extending the number of important operations we can perform without thinking about them. And indeed, over time we have reduced the work we have to think about. But what happens if it is not the operations we perform without thinking that multiply, but the decisions we make without thinking?

The thermostat never asked us what we wanted. With algorithms, things began to change. They started suggesting what to watch, what to buy, which route to take and whom we might meet. The shift looked small, but it mattered: the machine was now stepping in not after the decision, but before the decision had even formed.

We were the ones who asked Auntie Ayşe. Algorithms, on the other hand, often began to show us things without waiting for us to ask. We knew Auntie Ayşe’s opinion was Auntie Ayşe’s. We more or less knew what she liked, what made her cross, who she was not speaking to and why she told the stories she told, and we could weigh what she said within that context. With the new AI systems, things are more complicated. In some systems we can clearly see the objectives, the constraints and the way the calculation is made. With language models, we cannot always see with the same clarity which patterns of information shaped an answer, or why certain possibilities came to the fore.

What is more, Auntie Ayşe only ever gave an opinion, and the thermostat only carries out the target we give it. AI blurs the line between the two. For the first time, the thing we consult and the thing we grant authority to may be meeting in the same system, and to me that shift is far bigger than the arrival of a new adviser.

The problems I have encountered in years of working on Arya often bring me back to this question. In a production planning problem, the table may hold sales forecasts, inventories, capacity constraints and dozens of other variables. Most of the time the problem is not a lack of data; there is already more data on the table than we need. The problem is that no one can see alone which paths all that data makes possible together.

What happens if we make this decision? What if we change the assumption? Is there another path we are not seeing? Our algorithms can multiply the options, calculate the outcomes and simulate scenarios; but which price are we willing to pay, what do we care about, which risk do we accept? These are questions for the person at the table.

We have never made decisions alone. But if we are going to give a new intelligence a seat at our decision table, we should do so knowing what we want from it and what we will not leave to it.