Human in the Loop

The Lord insists: “You are the one who chooses to live this way. I build cities where things are human. I make Tuscany and you go to live with rock and silence.” 

Jack Gilbert, Going Wrong

The most dangerous ideas are those ideas that are intuitive, reassuring, and wrong. These are the ideas that lead us down Tuscan garden paths away from our goals and they are also ideas that are stubbornly hard to erase from human conversations. Even when proven wrong these ideas revived because they are so intuitive and reassuring.

Many of the most pernicious of these ideas are those from the world of health or politics and many are malicious creations. Those that are influential in business and management are often just misguided efforts to simplify. A few examples are the focus on extrinsic motivations to drive effort, the machine as a metaphor for the company, and the computer as a metaphor for human intelligence.

We may have a newer candidate. I had blithely accepted a new idea without depth of thought and it is widely cited. When we have concerns about the risks and challenges of our current wave of Artificial Intelligence, it seems obviously a good idea to ‘have a human in the loop’. This has become so generalised as a mantra when it comes to AI, I never stopped to consider it. When I did this week, I saw the ‘human in the loop’ may be intuitive, reassuring and also ultimately wrong.

Already Human

Better by far to embrace the hard truth than a reassuring fable. Modern science has been a voyage into the unknown, with a lesson in humility waiting at every stop. Our common sense intuitions can be mistaken.

Carl Sagan

Why do we accept the idea we have to put a human in the loop? Isn’t that the point of all our working efforts. Whatever we do is for colleagues, for customers or for the community. Humans are inherently the beneficiaries of all our working efforts. They should be the beginning, middle and end of all work. We aren’t using AI for its own sake.

Even if you imagine an organisation comprised entirely of AI agents, won’t its customers and community be human. The human is not included in the loop. The human is the point of human endeavour

Critically humans design the technology and its deployment. Even the Open AI hype reel for their acqusition of IO, highlights that the values of the humans creating this technology shape its future. Shoving a human widget into the flow of output of AI after the fact is at best a stopgap and at worst useless.

Will it actually work?

This will never happen again.
This keeps happening.

Ricki Cummings, The Failure Experiment

We talk of putting humans in the loop of AI processes because we lack confidence in the quality of AI and its ability to deliver a relevant empathetic experience to consumers and employees. We expect humans to pick moments of flawed inference. We expect humans to bring context, exception handling, and aiding with the complexity of real human edge cases at the edge of the broad probabilities of inference.

The wish that one human in a loop is going to solve all of the issues places a robust confidence in human’s ability to fight the system. People adapt and rely on systems. Look at our phone dependence. We rely on technology, especially anthropomorphised back boxes with renowned intelligence. It will take a brave human to challenge that loop of AI process.

The human in the loop is the latest example of treating humans as widgets in mechanical processes. In this case an unqualified quality control and an emotion widget to soften the machine. Are these humans in an increasingly fast and sophisticated AI loop just a better way to explain ‘the computer says no’? If we don’t give the human in the loop power to overrule, ability to manage exceptions and the agency to redesign the AI and the process we have gained nothing.

Yet again the human in the loop is an expression of management’s fixation with the machine metaphor of all work. The first human in the loop may have the flexibility required to be productive. However in our push for efficiency and for the greater adoption of technology we are just as likely to reduce rhetoric human’s role and influence over time.

So the next time someone blithely indicates they need a human in the AI loop ask the questions:

• for what reason?
• for whose benefit?
• how will it actually work?

When things are moving fast, we must stop and reflect on the ideas that race up at us. We don’t want to be swept away by new ideas that are intuitive, reassuring and wrong.

But when I do the obligatory double take
And glance behind me into the dark green future
All I see stretching out are vast
Arizona republics of more

Tom Clark, Human Life