Why AI won’t replace your manager

Although the headline that gets the clicks is “AI is taking our jobs,” the current reality is that “Automation is replacing some of our tasks.” I know, it’s nowhere near as catchy.

I say automation rather than AI because it doesn’t really matter what technology the underlying system uses, so long as it does the job well. For example, being a bank teller was a human job requiring intelligence, but there’s no AI in an ATM cash machine (when it comes to technology evolution, some of them don’t even seem to have caught up to Windows Vista yet).

I also prefer ‘tasks’ over ‘jobs’ because mostly AI can only do certain elements. MIT professor Erik Brynjolfsson, one of the authors who, with Andrew McAfee, first wrote about this trend in ‘Race Against the Machine’ has more recently been emphasising the need to redesign jobs around AI opportunities.

AI’s specific strengths

The biggest recent advances in AI tend to be around pattern recognition and rule-based reasoning. For example, machine-learning, image recognition and speech recognition are all largely pattern-based. They work well when there is a large and appropriate data-set for training, but they don’t ‘understand’ that a picture of a cat is a cat in the way that we do.

Chatbots, self-driving cars and IBM Watson then overlay pattern processing with a set of rules for deciding what to do. Hard coded-rules are easy to implement but tend to be brittle. Flexible rules take more work and can become unpredictable.

See also this business-oriented summary by Siw Grinaker, community manager at Enonic.

Can AI replace your manager?

Several articles about AI replacing managers have been published recently. I suspect much of this is wishful thinking by oppressed copywriters. Let’s break down a manger role into typical tasks.

In 1990, Henry Mintzberg broke down management into 10 roles. Let’s work our way through to see how amenable each is to automation:

1. Figurehead. This is about inspiring and being a figure of authority. It requires emotional intelligence and AI is only at the very early stages of being able to respond to emotion.
2. Leader. Mintzberg says this is about setting a high-level direction. It’s a very open-ended problem unsuited to AI.
3. Liaison is about building internal networks. Although there’s an emotional component to this too, we’re already seeing automation help by recommending connections in social network software, for example.
4. Monitor. Keeping track of progress and industry changes. Monitoring is hard for humans — often repetitive and limited by our information processing capacity. In part then, it is ideal as a pattern-recognition task for automation. For easily-quantified worker outputs, it’s already happening. Think of call-center monitoring or gig-economy workers managed via apps.
5. Disseminator. This is about sharing information with colleagues. Arguably we can automate some of this in the way that news services try to second-guess your interests. However, right now there’s still a way to go. Being able to think through the implication of new information has a creative element that most AI lacks.
6. Spokesperson. Mintzberg talked about representing an organisation externally. It’s a diverse and again open-ended task. A chatbot may cope with simple information requests, but it’s unlikely to win new customers over.
7. Entrepreneur. Solving problems and innovating. In general creativity is a big barrier for AI. Conversely, problem-solving has a long AI-history (Tower of Hanoi is a favorite AI student assignment). The issue is that AI is good at solving problems when the domain is well defined and the criteria for success can be articulated. Usually, this isn’t what we mean by “entrepreneurial” problem-solving.
8. Disturbance Handler. Stepping in when a roadblock is hit. This is the opposite of how most AI works in practical terms. For example, when self-driving cars hit a roadblock, they expect the human supervisor to take over, not the other way round (and I’m giving myself a bonus point for the pun).
9. Resource Allocator. Allocation of resources is about both tasks to people and funding to needs. Where the skills and needs are well-defined, there’s plenty of scope for automation on this one. Rule-based systems have been helping with shift-scheduling and timetabling for decades, for example. It becomes much harder when soft-skills come into play, however.
10. Negotiator. Participating in, and directing negotiations sounds like a very soft skill to me. There is actually an active AI research theme around this, and sometimes people respond favorably to the idea of a machine facilitator because they feel it will be less biased. Sadly, there is plenty of evidence that algorithms can be biased too, it all depends on the training set. For example, one study of machine learning found that male names were more strongly associated with “professional” and “salary” than female names.

So of the list above, we have a decent case for automation being influential on two of the 10 roles (Monitor and Resource Allocator) and helpful in two more (Liaison and Disseminator).

Even if we can, doesn’t mean we will

Note that just because a task can be automated doesn’t always mean that people will get replaced. In the 30 years after ATMs were introduced, the number of bank tellers increased slightly.

Removing workers is one response to automation, but improving quality is another option to pursue. The same thing has happened with word processing: reports used to be dull, monospaced affairs. In principle, the introduction of word processors could have led to us spending much less time producing them. In practice we spend the same time or more producing dull reports that now have sophisticated layouts and slick graphics.

So don’t wish away management just yet — but be prepared for AI to make them more sophisticated and slick than ever before. But just as dull.

This article was originally published over at CMSWire.

Photo credit: Michael Coghlan.