Leaning into Leadership – Podcast Episode #3
We’ve all heard the expression “use it or lose it”. But what happens when we start relying on AI to do more and more of the thinking for us?
Leaning into Leadership is a podcast series exploring the forces reshaping leadership across Australian businesses and government. In this episode, we explore looking at AI and decision-making in the workplace.
Across Australia, organisations are rapidly adopting AI into everyday work to analyse information, solve problems, and increasingly to support decision-making. The benefits are easy to see. AI can make it faster, more efficient, and potentially more productive. But there is also another side to the story.
What happens to our own capabilities when we start handing more and more of the thinking over to AI?
Do we risk losing some of our ability to analyse information, exercise judgment, and make good decisions for ourselves?
When AI gets it wrong, who’s accountable?
- is it the employee who used the tool?
- the leader who approved the decision?
- or the organisation that introduced the tool?
- if performance suffers, where does the responsibility actually sit?
These are some of the questions we’ll explore in this episode.
Joining us for this podcast episode are Rory Gregg from SpencerMaurice, and Amy McWilliams from Second Curve Psychology.
You can watch the video on Youtube, or continue reading below.
Rory Gregg:
AI is becoming pervasive in the design of work. I’m really curious to explore and unpack with you a little bit more around the psychological implications and potential hazards associated with humans working alongside AI tools.
Amy McWilliam:
I think what’s really interesting about this is that we have been handing over parts of our thinking to tools and tech for a long time. In psychology, that’s called cognitive offloading.
For example, using a notepad to write down your shopping list, or using a calculator to add up bills – those are examples of cognitive offloading.
Another really interesting example is using GPS for navigation.
So GPS has really changed the way we navigate. Firstly, it is now obvious to most people that we don’t need to calculate a route ourselves. The tech gives us options, and we just choose one.
The research now shows that as a result of routinely using GPS, we engage differently with our environment, and we pay less attention to our surroundings.
We are now building less of a mental map of the areas we are driving in. Over time, research studies have shown that heavy GPS use has been linked to poorer navigation skills, and declines in spatial memory.
This doesn’t mean that GPS is evil, or that it is damaging our brains. In reality, we are just practicing different skills.
And so the question starts to become: Does that actually matter?
And I would argue probably that it probably doesn’t matter for most people.
This is because GPS is reliable most of the time, and people are pretty happy to trade independent navigation for the convenience of just following directions and a map on a screen.
Leaning on these tools as a convenience can clearly weaken a person’s capability, if the tool eliminates the practice that lead to the development of the capability in the first place.
But this only really matters if we still actually want or need that capability.
So it is not whether that cognitive offloading relating to AI is inherently good or bad. We just need to be really deliberate, and think through which capabilities we are comfortable allowing tech to replace, and which ones we think are too important to stop practicing.
Rory Gregg:
In your view, what are some of those, skills and capabilities that are too important to lose?
Amy McWilliam:
So one of the really interesting ones is, is judgment.
Using AI to summarise some information or handle the routing of a customer inquiry, that is one thing.
But increasingly, we are starting to ask AI to assess risk, or rank options, and recommend an outcome. Those tasks generally involve evaluation and the use of judgement.
So insurance is a really good example of a situation where AI might assess a claim and recommend that it should be rejected.
The assessor may still technically make the final decision, and override the recommendation. But the very fact that the AI tool is providing the recommendation can make that recommendation seem objective and authoritative, because it appears to be based on data rather than a person’s opinion.
So if I accept the AI decision, then I am just following the system.
If I disagree, I now have to back my own judgment over the system’s judgement, and be prepared to explain and defend why I think it’s wrong.
That creates a really different burden, and it can make going along with the recommendation much easier than challenging it.
Rory Gregg:
So what you are saying is that by introducing AI to do some of that cognitive offloading at the initial phase, we are actually creating an additional challenge for the worker. They now have to manage a different set of stakeholders in order to come to a decision.
So we are loading on a different point of view that the AI is providing. This could now put the worker in the situation of having to challenge that AI decision. Which is essentially creating risk for that individual to go against what the AI is proposing.
Amy McWilliam:
Yes. And I think what tends to happen, is that workers just don’t ever challenge the AI decision.
The easier option is to just go along with the AI output.
Over time, this could mean that the organisation loses the richness and depth of decision-making capability. When workers just accept what they are given, they are no longer practicing their skills over time, and they will end up lacking the experience and the capability then down the track to keep making those decisions effectively.
And so it does just become easier to say, “Okay, well, I’ll just do what the system says, then I can’t be held accountable.”
But what we are actually seeing is that sometimes that individual person is still actually responsible and accountable for that decision, but they’re not necessarily capable of making the decision.
Rory Gregg:
So this, poses an interesting conundrum for leaders and workers within organisations that are adopting AI tools for decision making.
What are some of the steps and strategies organisations and leaders can take in order to avoid those unintended consequences?
Amy McWilliam:
So there are a couple of areas that I think organisations can get started on, without adding another layer of bureaucracy.
Success measures often look at things like usage rates, time saved, cost reduction, and outputs. While these are really useful, they only tell you whether people are using the technology. and if work is moving faster.
There was a recent study that showed that forty percent of desk workers in the US said that they had been handed AI outputs that were substandard, requiring rework. The research indicated that it took them on average two additional hours to review and correct those faulty AI outputs.
So if you put an hourly rate to those two hours, the cost can start to add up really quickly.
So we might be saving time in one part of the process, but the whole process chain may not be working effectively, and the problems just pop up down the track.
So thinking through some of those success measures in terms of the quality of the work, the whole workflow, become very critical.
In addition it is worth looking at the actual workload as well. If a task now takes twenty minutes when it used to previously take an hour, we don’t magically get a forty-minute break. People generally start to do something else, or juggle lots of things at once.
Mentally, that chopping and changing means things start to take longer. I think actually starting to look at those workloads, the workflow, as some of those success measures, is really important.
The other areas to consider are around your one-on-ones and team meetings, I think starting to really discuss AI effectively in those meetings.
A lot of leaders are uncomfortable having these discussions because they don’t feel confident. They don’t feel like they’ve got it figured out.
But I think it is okay to say that, because none of us have this all figured out yet.
So I think we can say we are working our way through it, and that the best way to do that is to get feedback from your team members.
So one-on-ones, asking someone to bring a piece of AI work into the meeting with them, and talking through
- how accurate and effective are the AI outputs
- how much rework is required
- how long is it taking to get the final result?
- how is it impacting their overall work?
- what is working well?
It is worth regularly discussing this one-on-one with team members. This can also become a regular part of team meetings.
People making noise about concerns relating to AI tools may not necessarily be change resistance. It simply be a sign that something is not working effectively.
For team meetings, it could be conversations like:
- how is the team incorporating AI agents into their work?
- is it consistent across teams, the organisation?
- are any teams doing something really effective that we could all leverage?
- should we be using AI tools in a more centralised, structured manner?
- should team members specialise in AI tool use?
- should processes be redesigned further?
So you can start to have these conversations as a group, and work out what the most effective way to manage it is.
When we talk about governance, organisations typically go to things like security and who has access. Those are obviously really important elements, but I think we need to look again at decision-making,
When we are mapping out those process flows, thinking through what decisions are being taken away from individuals.
Documenting the RACI – who is responsible, accountable, consulted, and informed. Where does AI fit into the process, and the RACI.
If we are taking decision-making away from an individual, who is accountable when things go wrong?
Because inevitably at some point, something will go wrong that we didn’t plan for.
In terms of accountability, it might come down to the organisation’s decision to use AI in the first place, or the individual who was operating the system. We need to be more clear upfront about how that actually works in practice.
Rory Gregg:
Do you see a scenario in the future where a worker’s performance and productivity is measured not only on their own individual contributions, but the contributions of those AI agents that they are essentially designing and using?
Amy McWilliam:
I think yes. People will have their own personal performance measured at least in part based on the AI outputs and AI performance.
Organisations are starting to mandate AI usage for some teams, and it is becoming a compulsory part of some work.
It will increasingly start to be a requirement for workers to ensure they are skilled in the use of AI tools, so that they can work effectively. People will almost certainly be measured in terms of their ability to get useful performance out of the AI tools.