Why I think ChatGPT will make audit more human

Two things that I've read on ChatGPT have stuck in my mind:

1. It took ChatGPT five days to reach one million users. That's insane. To contrast - Netflix took 3.5 years and even Facebook took 10 months. This has piqued interest across all industries and the general public like nothing else before it, so both auditors and auditees need to be ready for what that means.
2. That an article on The Verge argued how large language models like ChatGPT have a "propensity to generate fluent bullsh*t". That means we need to think carefully about how we use it.

Sidenote: I've probably broken the corporate style guide with a profanity inside my first 100 words. But I am only human, after all.

So let's take each of those points in turn, and unpack why I think AI like ChatGPT will only make audit more human.

What does ChatGPT mean for Audit?

My colleague Matt Campbell spoke recently about the progress we are making in applying AI to Audit (listen to the podcast here).

We already use cognitive tech to read documents and match information back to system records, and to scan social media and filter out the noise to identify emerging risks. We even had one of our MSc in Data Science graduates build a voice recognition demo - allowing an auditor to navigate our audit workflow system, KPMG Clara workflow, just by saying "Hey, Clara!".

So technology like ChatGPT isn't new to us. But we are getting to the point where it can be embedded into our day-to-day workflows without us even realising it is AI - just like Apple and Google already do in their products today. Because of this, I think we are about to enter the era where the efficiency benefits of automation that we've been talking about in Audit for some time can start to become a reality.

Imagine being able to ask Clara to read an accounting paper prepared by management, summarise the key points, and highlight any potential risks by comparing it to similar papers produced by other organisations. All at the click of a button.

But that is where the humans come in. A wise old soul (actually, it was John Bennett ) once explained to me how the word "auditor" has its origins in the Latin verb "audire", which means "to hear".

Machines can listen for a command and they can react. But they can't hear in the same way that a human can. The machines won't have that auditor instinct that can sniff out something that feels out of place, or the intuition and sense of context that tells them to ask just one more question because it doesn't seem like we've got the full picture.

Why is this important? Well the world is only getting more complex, businesses are operating in different ways and at greater scale than ever before, the data being created by businesses is exploding, and traditional risk footprints are changing as industries and stakeholder expectations evolve.

The only way I can see us navigating that complexity, whilst continuing to provide trusted assurance to a diverse range of stakeholders, is to pair expert auditors with effective AI.

Why we need to think carefully about how we use it?

Whilst AI is certainly getting smarter, it doesn't have all the answers.

If we ask ChatGPT (or any other similar model) a badly formed or vague question, we are likely to get a badly formed or vague answer. If we ask ChatGPT a question on a topic it hasn't been trained on (I've noticed it isn't great with large numbers), we are likely to get an incorrect answer. This is where the humans come in again.

In my view, the auditor of the future will have easy to use AI at their fingertips. But to translate that into a higher quality audit, we'll need to teach auditors to be excellent at the most human of skills - the ability to ask insightful questions, being able to clearly articulate a vision and end goal, curiosity, scepticism, challenging norms and thinking outside of the box. We'll also need massive diversity across these skills - but that's a whole other post!

If we can be clear on what our strengths are, we'll be able to use the right type of AI for the right sort of task – unlocking our productivity and allowing us to focus on the really challenging issues.

But we also need to think about what not to use AI for. Whilst we could use a model like ChatGPT to inform our judgements, provide suggestions, and validate completeness of our thoughts - clearly it wouldn't be appropriate to rely solely on AI to form a judgement on a complex accounting position.

A big focus for us is working with auditors to embed AI in the audit responsibly. We need to build trust in an AI-enabled audit - firstly with auditors, but ultimately with our regulators and the consumers of our audit opinions. We're putting the right guard rails in place - making our AI explainable and building challenger models to spot inaccuracies. But we'll need the guiding hands of skilled, and very human auditors to lead the way.

So if I were an Audit Committee Chair – I’d be asking my auditor not about what AI they are building, but about how they are building it, and how they are ensuring their auditors are ready to make the most out of it.