Back button

How to map human and machine roles

March 18, 2020
Workshop poster for human vs machine

Once you’ve figured out the problem you want to solve and that AI is the right way to go, the next step is to figure out how to apply AI. It’s important to understand whether AI should replace what a human does (which can reduce time spent on monotonous tasks and free people up) or whether AI should enhance what a person does by making it easier to do a task, making task completion more powerful or add to the skills of the individual.

This decision is important to get right. The wrong decision will mean the AI system doesn’t match what people want it to do.

When to use machines

People like to see a task when fully automated when a machine can do things easily that people find hard. Machines can search across entire databases of photos and, increasingly, text for example. People like machines to do things that make humans more efficient, reduce tedium or keep them safe. Increasingly, there are things that machines can do to enable new experiences for people.

As a general guideline, the best way to find out if a task is best left to a machine is to test whether people lack the ability to the task or whether the task is boring, repetitive or dangerous. Sometimes this assessment can vary based on how often people do things or whether they are time constrained. For example, some tasks are a chore during the week but can be done for recreation on the weekend (preparing a meal, for example). Understanding the desire for convenience and how people derive personal satisfaction in different circumstances is an important.

When to use humans

It’s easy to think that all easy, repetitive or low skill tasks can be automated. This is the heuristic that the news headlines has us thinking. It’s more often wrong than right. This is because there are many tasks that people do that they like doing. In these cases, AI can enhance their ability to do them, essentially giving them “superpowers.”

If an AI can make a task more enjoyable, give a user more choice and control, give them a greater sense of responsibility or fulfillment, allow them to serve more people or help them be more creative, then the challenge isn’t how to replace people, it’s to figure out how AI can augment human work.

The trick to spotting an augmentation opportunity is to look for where tasks are more complicated, more multi-faceted and personally valuable. Often times they are things that give people a buzz and allow play. This can make them even harder to spot because they don’t yet exist. This information asymmetry can be tricky and it helps to have a diverse team brainstorm the possibilities and unlock new opportunity for new things. On one of the spectrum, augmentation can unlock unlimited craziness (say, fun filters on photos), while, on the other, augmentation can help people work with ideas they otherwise couldn’t (say, software that generates millions of different building designs for designers to choose from).

Simple guidelines can help you figure this out. If people enjoy the task, look for ways to help them stretch the boundaries of what they can do – more designs, more pictures, more songs. If people want to, or have to, take responsibility, then it’s best to think about AI as an augmentation tool. This is especially true if the task generates social capital or is a favor to someone. People value the time, energy and expertise that someone brings to a task so if this is replaced by an inhuman AI, the value of the task can disappear.

When the stakes are high it’s likely to be an augmentation opportunity. This is due to a confluence of factors: humans want people to be accountable, humans can pick up when machines fail, machines can pick up when humans fail and humans accept a different level of error or lack of explainability from machines than they do from other humans. For example, while there are many AI tools to help lawyers read screeds of texts, there are no robot lawyers defending someone in a court room. Medical imaging is another example. In some diagnostics, AI is more reliable than doctors, but the combination of both is the most reliable of all.

The augmentation opportunities that are often most difficult to identity are where people can’t define what it is they want – what constitutes “done” or what constitutes “success.” The important concept in this situation is choice and control: people want to figure it out or course-correct as they go. This happens a lot in market research and analysis where the task isn’t just about data crunching, it simultaneously requires someone to enter data while also creatively displaying wisdom in data, for example.

Use the templates to tease out human versus machine roles and challenge yourself and your team with questions such as “if you were training someone to do this, what instructions would you give them?” “How often do you do that thing you just did?” “What would happen if it was done by someone else?”

Workshop poster showing human vs machine