Embracing uncertainty in the age of AIEmbracing uncertainty in the age of AI
An opinion piece from the CCO of Peak, a decision intelligence company
June 6, 2022
An opinion piece from the CCO of Peak, a decision intelligence company
What makes an ideal leader? Traditionally, we might say assertiveness, certainty, decisiveness. That’s because typically businesses have rewarded confidence.
As humans, we all understand that there is uncertainty in the world – every situation presents a probabilistic distribution of potential outcomes – but show anything approaching a misgiving in a business setting and we encourage more questions or, worse, questions about our capability. So we learn to project confidence and certainty around a specific outcome. But even the most authoritative leader is often guessing when making decisions.
As artificial intelligence continues to move into the mainstream, many leaders are finding that they need to learn a new way to lead. AI systems for the most part produce mathematical probabilities of certain outcomes or recommendations. For leaders, this means adjusting the way we communicate about the future and how decisions are made.
As companies adopt AI technologies, leaders must embrace uncertainty, recognize our limitations and think differently.
Probabilistic decision-making — that is, decisions that consider the possibility of alternative outcomes — is at the heart of this new approach. And those of us that adopt a style of leadership that complements this technology will gain a competitive edge in the AI age. Those that don’t might never capitalize on the millions of dollars’ worth of data flowing through their organization.
The deterministic mindset that currently governs most businesses attempts to do away with uncertainty. Decisions are unpacked as either right or wrong, who ultimately made that decision and whether they should be in that role.
AI and machine learning (ML) models require a different set of questions. They don’t simply weigh right or wrong. AI models are probabilistic and provide a view on the mathematical probability of uncertainty. Layer on top of it human intuition and experience, and you suddenly have a very interesting and informed debate, with actionable insight provided upfront – as opposed to the data in arrears we’re all used to dissecting.
It is no longer a case of asking who made the “wrong” decision and if they should be in that role. Instead, it’s this: If the model thinks it highly unlikely that we need that level of inventory, why aren’t you recommending it?
This gives us a chance to ask questions ahead of an event, understand issues more quickly and pinpoint any incorrect assumptions. More than that, it has a huge impact on empowering teams. For one, employees are no longer at the mercy of a boss who insists that “what I say goes.” Good probabilistic systems will also help leaders detect their own biases or those of their teams in assuming certain types of outcomes.
Embracing probabilistic thinking
The data needed for probabilistic decision-making is already moving through a business, but the majority aren’t harnessing it to its full potential. Very few companies are using a probabilistic decision process across every function, although we are seeing strides in areas such as forecasting, asset maintenance and failure prediction.
Weather forecasting is an example of probabilistic decisioning with which we’re all familiar. Weather forecasts are given in terms of the probability of an outcome. For example, when predicting the path of a hurricane, forecasters include a range of potential paths and outcomes. This allows for emergency preparedness, and means individuals and the emergency services all have a sense of the total range of possibilities and can plan accordingly. These paths are determined by mathematical models, and interpreted by human experts.
How do we as leaders take the leap from familiar examples of probabilism in our everyday lives to embedding it in our teams and organizations? It’s a deceptively simple answer – ask questions.
The first step towards any change in any area is to be curious about the inner workings of it. In this case, how the data is being collected, curated and modelled. If there isn’t any data, then that’s the first question to ask: ‘Why don’t we have any data on this?’
If there is data on hand, then the questions are pretty much limitless. Different companies – even different teams within those companies – are at different stages of data literacy; and where they sit on a maturity scale will determine the questions you need to ask.
For those still in the early stages of their AI journey, the question might be ‘Do we trust the data we have on this?’ Typically, multiple teams within a business record the same data points, and there are often discrepancies between the two reports. Finding a ‘single source of truth’ − i.e., data everyone agrees is accurate − is an important first step in understanding an output.
By contrast, more mature teams will be able to talk you through the models they’ve used. The questions to ask here are – ‘Why did we use this model?’ ‘Are there any biases we should be aware of, and could they be mitigated by using a different model or approach?’
A word of caution to less-technical users looking to ‘lean in’ and learn: Make sure you approach the situation by explicitly stating you are genuinely looking to learn, and with a frame of curiosity. If data folks suspect you are there to criticize, critique or judge their work, and you may have a mutiny on your hands. Instead, ask questions and engage with the learning experience – look up terms you don’t understand, ask what technique the team is using and read a paper on it.
Adopting AI in your business
Remember, you’re on the same team as those data folks – and that’s an important consideration. When it comes to AI, cross-functional teams are consistently more successful, more quickly. Getting great AI projects into production with commercial impact requires a multi-skilled team that can infuse the model-building process with business intuition, insight and experience throughout the build; and ultimately commercial leaders who can act on the outcomes of the model.
It takes a team with technical understanding to build the model and business expertise to understand its predictions and how to act on them. If any part of this process happens in silo, the results tend to be lower quality, delayed and often irrelevant. It’s one of the reasons why many ML projects don’t go to production.
A future where AI is embedded in our organizations is closer than we think, and it’s going to take visionary leaders to support that transition. The most successful leaders, the ones that thrive in an AI-powered business, will be those of us that invest a level of curiosity in the process and are on a journey to become quite literate on the topic, as opposed to outsourcing that literacy. You don’t need to be an expert, but you do need to be curious.
You May Also Like