AI, ESG and are we kidding ourselves?

I recently had the pleasure of speaking on a panel at the CFO Magazine Symposium in Sydney. It was a fantastic event, with somewhere between 300 and 350 CFOs and finance leaders in the audience. Our panel was discussing what makes a great CFO: the relationship with the CEO and the board, earning trust, dealing with difficult issues and, ultimately, what boards will expect from CFOs by 2030.

Not surprisingly, AI was a recurring theme throughout the conference. There was enormous enthusiasm about what it can do for productivity, how it will change the finance function and, inevitably, what it will mean for jobs.

I share that enthusiasm. I use AI extensively and think it will fundamentally change the way we work.

But as I listened, I kept thinking about a book I had read several months earlier, Karen Hao’s Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI.

For anyone unfamiliar with it, this is not some abstract examination of artificial intelligence. It is substantially the story of Sam Altman, OpenAI and the extraordinary rise of ChatGPT, the very technology millions of us are now enthusiastically using.

It is a fascinating and at times disturbing book.

Hao takes us behind the wonderfully simple experience we have with ChatGPT. We type something into a box and, within seconds, receive an answer that would have seemed impossible only a few years ago. What we don’t see is the enormous infrastructure and resources required to make it possible.

That made me wonder whether we are having conversations in our boardrooms that simply don’t join up.

Take ESG and greenhouse gas emissions.

For larger companies, particularly those now caught by Australia’s mandatory climate reporting regime, measuring and reporting our environmental impact is becoming an increasingly important part of corporate life.

Those responsible for ESG, including boards, CEOs and CFOs, are increasingly focused on the carbon footprint of the buildings their companies occupy. Yet how many of us have ever stopped to think about the carbon footprint of the AI being used by everyone sitting inside them?

I suspect most CFOs haven’t the faintest idea. I certainly don’t.

Yet AI requires enormous computing power and therefore enormous amounts of electricity. There seems to me to be something inconsistent about carefully considering the environmental impact of the buildings we occupy without even thinking about the environmental impact of the AI we are increasingly encouraging everyone inside them to use.

I am not suggesting that we stop using AI. Far from it. Nor am I particularly interested in debating precisely how much electricity an individual ChatGPT query consumes. The technology is changing and different applications consume vastly different amounts.

The point is much simpler. If we are serious about understanding our environmental impact, surely we should at least be thinking about it.

And then there are the people.

This is where Hao’s account of OpenAI and ChatGPT becomes particularly uncomfortable.

The development of AI has not depended solely on highly paid engineers in Silicon Valley. It has also involved people around the world labelling data and helping the systems distinguish between acceptable and unacceptable content.

Hao writes about workers in Kenya undertaking work connected with OpenAI who were required to review extraordinarily disturbing material, including graphic violence and sexual content, to help make the systems safer for people like us to use.

They were a very long way from Silicon Valley, both geographically and economically.

That made me think about another document with which most large companies and their boards are now very familiar: the Modern Slavery Statement.

That does not mean they were slaves. Low-paid labour and modern slavery are not the same thing, and we should not pretend they are.

We spend considerable time examining modern slavery risks in our supply chains. There are policies, supplier questionnaires, risk assessments, training programs, codes of conduct and board approvals.

All very worthy.

But are we measuring the process rather than the result?

Take child labour.

Suppose we discover children working somewhere down our supply chain. Our instinct is obvious: stop dealing with the supplier.

Problem solved.

Except what happened to the child?

We would like to believe the child went to school the next morning. But if that child was working because the family desperately needed the income, terminating our relationship with the supplier has not made the poverty disappear.

Perhaps the child simply goes somewhere else to work, somewhere we can no longer see and perhaps in even worse conditions.

We have eliminated child labour from our supply chain. Have we improved the child’s life?

The same problem arises with very cheap labour.

Looking from Sydney at what somebody earns in Kenya or another developing country can be confronting. Sometimes the conditions may unquestionably be exploitative. But simply comparing their wage with ours tells us little about local living costs, working conditions and alternative employment opportunities.

If we decide the wage offends our standards and move the work elsewhere, our supply chain looks cleaner. The worker may simply become unemployed.

Have we helped them?

None of this is an argument for tolerating exploitation or child labour. It is an argument for thinking much harder about what we do when we find it.

Corporations have enormous economic power over suppliers. CFOs know that better than most. We are very good at using that power when we want a lower price, longer payment terms or better service. Why not use some of it to improve people’s lives?

If a workplace is unsafe, can we help make it safer rather than simply leave? If wages are genuinely exploitative, can our purchasing power help improve them? If children are working because their families need the income, can we contribute to solutions that actually help those children get into school?

And if people are spending their working day looking at horrific material so that ChatGPT is safer for us to use, shouldn’t we care about the conditions under which they do it?

There is, of course, an inconvenient part to all of this.

It may cost us money.

It costs very little to put a modern slavery policy on a website. It costs considerably more to improve someone’s working conditions. Perhaps that is one reason corporations have become so fond of policies.

The more I thought about Empire of AI, the more I realised that the environmental and human issues Hao raises have something important in common. The costs are largely invisible to those of us receiving the benefit. We see the answer on the screen. We don’t see much of what sits behind it.

And this seems particularly relevant to CFOs because one of the things we are supposed to understand is the true cost of something. Moving a cost somewhere else does not make it disappear.

I remain enormously enthusiastic about AI and ChatGPT. I think they will transform finance and create opportunities we are only beginning to understand. But perhaps the great CFO of 2030 will need to do more than work out how much AI can save. We should also understand something about what sits behind those savings.

There is a wonderful irony in all of this.

We can now ask ChatGPT to help calculate our greenhouse gas emissions, prepare our sustainability report and draft our Modern Slavery Statement.

Perhaps before we do, we should understand a little more about what went into creating ChatGPT itself.

Otherwise, we risk becoming very good at reporting on problems while moving some of their real costs somewhere we can no longer see.

And surely a great CFO should know the difference.

---

About the Author: Jon Brett

Jon Brett has spent more than three decades in senior executive and board roles as CEO, CFO, Chair, Audit and Risk Chair, non-executive director, founder, and adviser.

His career has included leadership roles with Techway Limited, a pioneer of internet banking in Australia; Investec Wentworth Private Equity and Investec Bank (Australia). He has also served on the boards of numerous ASX-listed companies.

He created and hosted The Taking of Vocus, a podcast series examining the rise of Vocus Group from start-up to a market capitalisation of more than $5 billion, and the turbulence that followed the acquisition of M2 Telecommunications. It was subsequently adapted into a book.

The Taking of Vocus is available on Kindle.

Follow or connect with John – LinkedIn Profile