How to successfully implement AI into your business

Garreth Hanley:
This is INTHEBLACK, a leadership strategy and business podcast brought to you by CPA Australia. Today you'll be hearing a discussion with Erik Brynjolfsson, who dialled into the show from Stanford University. Erik is an economist and an AI expert who'll be speaking at this year's CPA Congress. Congress 2024 is happening from October 15 to 18 in Canberra and is open to anyone. So if you like what you hear today, check out the show notes for links to register for Congress.

Aidan Ormond:
Welcome to INTHEBLACK. I'm Aidan Ormond. And in today's show, we're talking with Erik Brynjolfsson about human centred AI and the future of work. Erik is a senior fellow at the Stanford Institute for Human Centred AI and Director of the Stanford Digital Economy Lab. He also holds positions as the Ralph Landau Senior Fellow at the Stanford Institute for Economic Policy Research. Professor by courtesy at the Stanford Graduate School of Business and Department of Economics, and Research Associate at the National Bureau of Economic Research.

Now, Erik is a highly cited author on the economics of information and was among the first to measure IT’s productivity contributions and the complementary role of organisational capital and intangibles. And his pioneering research covers digital commerce, the long tail, bundling and pricing models, intangible assets, and ITs effects on business strategy, productivity, and performance. Erik has also authored nine books including the bestsellers, The Second Machine Age and Machine, Platform, Crowd, co-authored with Andrew McAfee, as well as over 100 academic articles and 5 patents. Erik holds a bachelor's and master's degree in Applied Mathematics and Decision Sciences from Harvard University and a PhD in Managerial Economics from MIT. Welcome to INTHEBLACK, Erik.

Erik Brynjolfsson:
Pleasure to be here.

Aidan Ormond:
Now Erik, there is so much that we want to chat with you about today, and your body of work emphasises the importance of a human centred approach to AI and finance, and you're being quoted as saying, "The CFO should be on the frontier of the AI revolution." So a question to you is, how can CFOs and financial leaders level up their organisations by integrating AI? Will this process displace their workforces, and what, if any, role does human augmentation play in this process?

Erik Brynjolfsson:
Well, that's a great question and one of the most important questions that I get asked frequently, artificial intelligence is revolutionising every function, every organisation, every industry, and none more than the chief financial officer's organisation. And one of the most common misconceptions, I think, is that AI is just taking over jobs. That AI can do what humans do. And while that is true in a few narrow cases, the more general lesson is that AI augments or helps workers including the CFO to do their job better.

As I wrote in the Harvard Business Review a few years ago in an article with Andy McAfee, "AI is not going to replace managers, but managers who use AI will replace managers who don't use AI." And that's certainly true for CFOs, and it's true they help the rest of the organisation understand it's true for most of the workforce. We take a human-centred AI at Stanford and also in my company, Workhelix, because we look for ways that we can have AI augment what people are doing.

Typically, AI can do parts of the job, but other parts humans do better. And the way to analyse that is what we call the task-based analysis. And the task-based approach breaks down any occupation into dozens or even hundreds of fine-grained tasks. And what we find is that AI can do some of those, but the others humans do better. We never found one occupation where AI just ran the table and was able to do everything that the person was able to do.

Because of this natural division of labour between humans and machines, an augmentation or human-centred approach works best. In fact, I would say not even just keeping humans in the loop, but keeping humans at the helm, humans should be in charge of the process and use AI to allow them to do new things, to do the existing things better, cheaper, faster, and higher quality than they otherwise would.

Aidan Ormond:
That's really interesting you say that, and we're going to go into the economics of AI as well. You've extensively researched the economics of information and AI. In what ways, Erik, do you see AI reshaping the economic landscape and how can businesses adapt their strategies to harness these changes for increased productivity and competitiveness?

Erik Brynjolfsson:
While the technology breakthroughs are very exciting, ultimately the big payoff is increasing productivity and competitiveness. And in many cases, that just isn't happening. There's a lot of hype out there. There's a lot of misdirection. There are a lot of shiny new projects and proof of concept proposals that have impressive technological capabilities but aren't really linked to the business. And one of my goals in my career and especially right now, is to help translate technology into productivity. What we've done is we've analysed lots of different, including AI, and we see patterns emerging.

The powerful technologies that really boost productivity are what we call general purpose technologies. The first amazing general purpose technology was all the way back with the steam engine, and electricity was also a general purpose technologies. These general purpose technologies, economists like me, used to just call them GPTs, but that has been stolen from us by the AI researchers. But these general purpose technologies have important properties. They affect almost all of the economy. They rapidly improve, and most importantly, they spawn complementary innovations.

They allow you to do new things, but it turns out that simply plugging a GPT technology into an organisation rarely translates into productivity. You have to do those complementary innovations, and that could include technological innovations like in the case of electricity, air conditioning, motors, refrigeration, many other things. But most interestingly, it often includes process innovations like the assembly line or new ways of doing work, new organisations or skills, new types of skills like how to code or how to drive a car. These new innovations in processes and skills is ultimately what unlocks the power. The thing is, that these process innovations are often intangible. They're not well measured even by the best CFOs, and they take time. And that means that sometimes while you're making the investments in the skills and the processes, it looks like the technology is not paying off.

What we see sometimes is what we call a productivity J-curve, where at first it goes down a bit and then it takes off, and that can be discouraging when you're in the trough of that J-curve. But ultimately, if you do it right and you make those complementary innovations, you start harvesting benefits from the intangible assets. This can take months or even years, but ultimately the payoff is much larger than it would've been without the complementary innovations. So the trick to capturing productivity from general purpose technologies like AI is to understand that they're embedded as part of a broader transformation process, and having the metrics to track that and make those investments correctly.

Aidan Ormond:
You touched there, Erik, on productivity and your research does link IT innovation with productivity growth. Can you just maybe further elaborate on how businesses can boost their productivity by adopting digital technologies and what common pitfalls they tend to encounter during digital transformational projects?

Erik Brynjolfsson:
Absolutely. This is my passion. This is what I've been doing research on at MIT and Stanford and what my company, Workhelix, focuses on. I'm excited by my smart technological friends and computer science and AI and the amazing technologies they're developing, but it's time to translate that into productivity. And to do that, I really think you have to understand the task-based approach that I mentioned earlier. Let me say a little bit more about it. What we do is we think about any organisation can be broken down into occupations, and those occupations can be broken down into tasks.

So a radiologist does about 26 distinct tasks, an economist over 20, truck driver, CFO, accountant, call centre operator, they all have a set of tasks that they do. It doesn't make sense in most cases to think of AI as automating an entire job or function, but it can take care of a particular task either automating it or augmenting it as I mentioned earlier. It can help you write a memo. It can help you analyse a spreadsheet.

And when AI helps with those particular tasks and you understand what they are, you can get significant payoff. Common pitfall is to not do this rigorous analysis, to not be data-based and instead listen to the hype, listen to the anecdotes. There are lots of consultants that will describe to you amazing projects that they've done in the past, but that kind of management by anecdote is not what we need anymore.

It's going to lead to a lot of wasted money. It's going to lead to a lot of disappointment. But if you're data-based and you do the task-based approach, you're much more likely to be able to separate hype from reality. I'm not saying it's guaranteed, you still need to use human judgement , but it puts all the different projects on a common basis. You may have lots of people in your organisation coming at you and saying, "Hey. I need 60% of the budget. No, I need 70% of the budget. No, I need 30% of the budget." And you can't give everybody all the budget that they want. But if you have a common basis like the task-based approach that compares everything on a data-driven foundation, you're much more likely to make intelligent trade-offs between those different projects.

Aidan Ormond:
Yeah. I love the fact that you're mentioning data-driven there. Erik, let's talk about work and skills for a little while.

Erik Brynjolfsson:
Sure.

Aidan Ormond:
With AI advancing so rapidly, what skills do you believe professionals in business and finance need to remain relevant, and what organisations can do to support their skills development?

Erik Brynjolfsson:
Well, one of the things we're seeing is a real revolution in the way finance professionals can understand their businesses. Of course, finance has always been a very data-driven organisation. It's part of the company that has numbers and tracks things mathematically and quantitatively, and that gives them a big edge in understanding what's going on in the business.

But now with large language models, LLMs, we're able to analyse not only quantitative data, financial data, but also lots of verbal text-based, visual and other kinds of data, and this is really the majority of the information in organisation. All the stuff in memos and reports, 10-K forms, a lot of that is text-based LLMs can help you understand that, and finance professionals need to broaden their toolkit to be able to understand how AI can help them analyse the business more broadly. Now, the set of skills is evolving pretty rapidly.

Coding is an important skill, but now we have LLMs, that is large language models, that are able to help people with coding just by giving English instructions and create often very complicated applications, analyses and summaries that weren't done before. The other thing I would say about skill development is that as these technologies are rolled out and are able to automate or augment certain tasks, that means that other tasks become less important and some become more important. Almost every function in the organisation is going to have people who become in short supply and other people who you have a surplus of with this churn, this task churn.

A great approach is to continue with the task-based analysis, applying it not only to the technologies to invest in, but also the people and skills to invest in. What you'll find is that as one job loses the need for certain skills, it can transition into a new job or new occupation that overlaps with that. So we can identify task adjacencies or skill adjacencies. A job that say, had 26 different tasks in it, maybe six of them disappear, but the remaining 20 overlap a lot with another new job, and that person can be re-skilled to do the new job.

This task-based analysis and the skill-based analysis ends up being a way not only to manage your technology, but also your human capital assets, which after all are the most valuable assets in every organisation.

Aidan Ormond:
And you're talking about humans as well here, but I guess with AI and ethical considerations, that's something that we should discuss.

Erik Brynjolfsson:
Yes.

Aidan Ormond:
Erik, no doubt about it. We're seeing a rapid development of AI, and with all the technological breakthroughs that we're seeing, there are rewards, but also some risks. What ethical considerations should businesses keep in mind as they start to use AI, and how can they ensure that these technologies are used responsibly?

Erik Brynjolfsson:
This is an incredibly important question. AI needs to be used ethically and responsibly, and that has to be part of every project analysis. One of the dangers of AI is that they can amplify some of the biases and misconceptions in an organisation because after all, machine learning systems learn from the data that you already have. So if you have patterns of hiring or patterns of giving loans or patterns of making other decisions that have some element of bias in them, racial bias, gender bias, other kinds of biases, the AI system's going to pick up on those and could even replicate and amplify them. Of course, that is doubly bad because it's bad enough to have the biases initially, but then to have them amplified and spread through the organisation and codified in software, that's particularly dangerous.

The good news is that AI also is, in many cases, easier to de-bias than are. You can run analyses to see what kinds of patterns that AI is using to make decisions, and you can reduce or even eliminate many of the biases in those decisions, say the hiring decisions. And ultimately, I'm optimistic that AI-based systems will end up being less biased than the humans on which they were trained on because they can be more transparent and easier to fix. It's very hard to teach a person to stop being biassed, although we all try. But with an AI system, you can codify it and identify and eliminate the bias to a much greater extent. So that's one ethical consideration that's very important. I'll just briefly mention another one, which is responsible use of the data and data provenance.

These systems are trained on data often from many different places, and you have to make sure that you have the rights to use those kinds of data, both from inside the organisation and that the data doesn't leak out inappropriately, that you don't end up violating privacy. These are things that can be done with technological systems. You also have to do the appropriate licensing, but it's something that you want to pay attention to as you set up the systems. And even new laws in many countries are being written to make it clear who has the rights to use which kinds of data, and you want to make sure that you're up on those different legal responsibilities.

Aidan Ormond:
And it's good to have those guardrails, but I love also your optimism there, Erik. I think we have to talk about innovation and management practises as well, because you've highlighted the role of management practises in driving productivity, but how can businesses and leaders level up their management approaches to foster innovation while still sustaining long-term growth?

Erik Brynjolfsson:
This is, I think, one of the most exciting things about the latest wave of artificial intelligence. A lot of previous computer systems would basically codify all the things we knew about how to do a process. So anybody who's coded knows you have to be very precise when you write computer code. You have to understand a process very carefully and write it down step by step. If you don't understand it, well, you're not going to be able to write the code for that. AI is different. AI, a machine learning revolution is different in the sense that it can write its own code, and if you give it examples of inputs and outputs, it can figure out how to map one to the other.

Some people call this software 2.0, but basically it involves the AI system figuring out those relationships, and that allows it to codify things that we didn't even understand fully ourselves. I was working in a call centre or studying a call centre operation, and there was an AI system that would listen in to the transcripts and learn from the successful calls and the unsuccessful calls and figure out what were the patterns they had in common. No human had identified some of those patterns before, but the machine learning system found these patterns by looking at literally millions of transcripts and analysing them. That's the kind of thing that we can do. And in every organisation, there are hundreds of opportunities to have the AI system learn things that no human had ever written down before, and even to take it one step further, to foster innovation. To figure out how to combine building blocks in new ways to do something that's never been done before.

Some people say that AI systems can't innovate because they're building on what was already there, but that fundamentally misunderstands the nature of most innovation. Most innovation is the combining and recombining of previously existing building blocks, and AI systems are very good at that combining and recombining. So I think we're in the early stages of a new era of innovation where AI systems will help companies identify new product opportunities, new processes, new marketing campaigns, and other new ways of innovating that we never were able to do before. Still, as I said at the outset of this interview, keep the humans in the loop and humans at the helm where they're in charge, but AI can help brainstorm and identify opportunities for innovation that we've never seen before, and that makes me optimistic, not only about increasing the level of productivity, but also increasing the slope of improvement in productivity. That is the rate of growth of productivity. If we can increase that, then the future is very bright, indeed.

Aidan Ormond:
And it is an exciting time in the world, I think so. There's so much we could talk about on this topic, Erik, but thank you so much for joining us today. It's been absolutely fantastic talking to you.

Erik Brynjolfsson:
It's been a pleasure for me as well.

Aidan Ormond:
And thank you for listening to INTHEBLACK. Don't forget to check the show notes for links and resources from CPA Australia and INTHEBLACK, and a link to register for CPA Congress 2024. Until next time, thanks for listening.

Garreth Hanley:
If you've enjoyed this episode, help others discover INTHEBLACK by leaving us a review and sharing this episode with colleagues, clients, or anyone else interested in leadership, strategy, and business. To find out more about our other podcasts, check out the show notes for this episode, and we hope you can join us again next time for another episode of INTHEBLACK.