[MUSIC PLAYING]

Welcome to Mercer's podcast series on the new shape of work. I'm Kate Bravery, Mercer's advisory and insight leader, and today I'll be diving deep into generative AI which let's be honest, is the conversation of the year. Well, certainly one that's been dominating the airwaves. Specifically, we'll be discussing how generative AI is impacting the workplace. And more critically, how we can at both the individual and organizational level prepare ourselves with impact.

This conversation, I'm joined by Ravin Jesuthasan, global transformation services leader at Mercer, and Sophia Van, head of global digital portfolio, also at Mercer. Welcome, Ravin and Sophia. I know that you both speak and write a lot on this topic, so I'm really looking forward to the conversation today.

Thanks Kate, it's lovely to be here with you.

Hi Kate and Ravin. Thanks for having me.

All right, well why don't we dive in.

Sophia, I'm going to come to you first. And maybe for the people on the call that are living under a rock, why don't we just level set with what is generative AI, and it's very public apostle Chat GPT? Maybe you can share from your expert vantage point, a definition. And maybe in plain English, share why everyone's getting so excited.

Yeah, sure, Kate. So large language models have been around for a while, but last November something happened, right. Everybody talks about Chat GPT. Within just 50 days of launch, it stormed the world with lots of news. It passed the Google engineer test and the MBA exam, and several law school exams. So in addition to the Chat GPT, probably you've heard about Ellie, Midjourney. Those tools that can create realistic images. So Midjourney even competed against human artists, and won the first prize at the Colorado State Fair Fine Arts competition.

Then we heard about Microsoft's Bing, and Google Smart. They are alternative AI apps, and currently there are hundreds of them. So these apps really put AI at the fingertips of so many people in ways that they had never experienced before. So what is generative AI? It refers to a subfield of artificial intelligence that focuses on creating new, unique, and useful content or data that goes beyond index retrieval and pattern recognition. These systems have the ability to perform creative tasks that was once thought to be exclusive to human.

So to illustrate, let's do something fun. Let's [INAUDIBLE].

Hi Kate and Ravin, this is Sophia AI. I'm about to drop some knowledge. So listen, don't be shy. The future of work it's a whole new scene. Robots and AI mixing with the human team. Collaboration, new jobs unfold. Together we're building a future that's full. I'm still with you. We've got to be steady. [INAUDIBLE] the change, stand strong and ready. The future of work remains unsure. Together in this world we'll create a new work. Get ready y'all, the future's here. Working together, no need for fear.

Embrace the change, keep moving ahead. The future of work will break. That's the [INAUDIBLE] and all. A world where tech and people stand tall. We'll face challenges, but we'll rise above.

Do you like it?

I love it. Sophia, I didn't even know you were a rapper. But that was a very eloquent poem about generative AI. Did you just come up with it?

So basically, I don't know how to rap. So what you just heard is actually a digital version of my voice. The lyric was written with Chat GPT, and the music was composed by Boomy.com. Again, please note that this is solely for educational and demonstration purposes. Not meant for any commercial uses with my voice. But how long you think it took me to produce that song? Well, it took me just a few minutes. And I think that's it's fascinating, right? Because we have been talking about the marketization of AI for a while. And now it has become a reality.

It is extremely accessible. Using these tools can be as easy as making a phone call. So now not only adults use Chat GPT to write hundreds of books on Amazon, but kids use Chat GPT to do their homework. My teenage daughter actually just recently told me that, hey mom, Chat GPT would make many millionaires in the next 24 months. And I asked her, where you got that from? And she say TikTok. So not sure if it's true, but I'm pretty sure we are witnessing a giant leap of productivity. And we ran a couple of internal experiments as well, that I would love to share with you if you're interested.

Yeah, I would definitely love to hear them as we go through the conversation today. I think it's fascinating how young people are gravitating towards it as well. It's certainly helping me do homework with my kids a lot easier. I didn't know about the commercial opportunity. I think my teenager would be very happy about that. You said a couple of things there that I think are very relevant. You said it's democratizing access to creating new content.

And you talked about just the sheer accessibility of it. And I think those are really the two hallmarks of the period that we're living through. I wonder though if you could talk a little bit about some of the risks. Because with that unprecedented access and opportunity, what are some of the risks that maybe some of us sitting in HR need to be cognizant of?

Kate, you know that I'm really, really fascinated about the democratizing of AI, which means the widerspread use of AI to a broader range of employees in the organization beyond just research labs and IT departments, right. Because I believe that a lot of use cases, the impact would happen right at employee desks. But it is definitely a double edged sword, because generally AI is in its infancy. So there are risks. So one of the risks is they lack common sense and real world understanding, which impairs their reasoning abilities and can lead to inappropriate or misleading responses.

Large language models are trained on vast amounts of text data, usually sourced from the open internet, which may include offensive, controversial or biased and content. They can be manipulated into generating toxic content. They do not understand emotion, of course. They lack empathy. And I think one of the biggest problems that many people talk about is accuracy. We know that they can get things wrong and hallucinate incorrect facts.

Chat GPTs knowledge for example, was limited to information up to September 2021. So if you ask it about details beyond that point, the older version might have fabricated facts. GPT 4 has shown significant improvements, it's OpenAI claim that it is less likely to respond to now content, and more likely to generate factual responses. But we are still talking about the probability of was hallucination here, which is around 8% to 10%.

So I think it's very important to have human reviews then. We also from our own experience, we know that implementing best practice, the prompting, and fine tuning, and crowding with additional context also helps mitigate risk, and it needs special skills.

Well, you share quite a few of those new skills that are coming into the workforce. So I think we'll dive a little bit deeper on that. When you were just chatting there, Sophia, the fact that people couldn't have access at the employees desks I think also just makes me think, gosh, what kind of information, what proprietary information might be going into those systems? So I agree with you, there is a few things to be wary of. I also think it's quite interesting that [INAUDIBLE] is inaccurate like human beings, and it can actually muddle some of that empathy even though it's not empathetic. So we are definitely in a brave new world.

And Ravin, I wonder if I can bring you into the conversation. As we start to think about the impact on the future of work, its reverberations are mind boggling. And I wonder if you could share a little bit about, what do you see as some of the immediate impacts that we're going to be feeling in the workforce? And maybe more critically, how do we half of them?

Yeah, Kate. You know I think Sophia has given us a really good grounding in the capabilities of this technology. And I think what's really fascinating is looking at how the advances in these large language models really sort of belie, and sort of build on what we've traditionally seen AI do, right. So previous iterations of automation with machine learning have largely impacted repetitive rules based work.

But the thing that's intriguing about generative AI is the democratization of creativity and knowledge, means that it's also affecting low volume, highly variable work. That where the risk I think said kicks in is, these large language models are looking at adjacencies and connections. They don't often have the logic that we might have seen in other variants of AI. And I think it's that lack of logic that then gets to be deeply problematic. And it becomes a risk factor.

It is-- I wrote a piece for the World Economic Forum around where these models make sense. In a way, where they can be used safely and securely, versus where they might potentially represent a risk factor to the organization. Because particularly in areas where you're depending on a model without logic, without significant amounts of highly specific data, that then result in potential errors without the potential for human risk mitigation. I think is where the impacted work could be quite significant.

I was just about to say in combination to that, the confidence and assertiveness of the single responses I think puts that risk even higher. And to your point, it's not what we've got used to with prior incarnations of AI, so it really is a juxtaposition to what we've got used to.

Yeah. Yeah, I think that's absolutely right, Kate. And you know what's fascinating is, stepping back and thinking more broadly about work, as we know with this iteration of the large language models with previous iterations of both machine learning as well as the more basics of robotic process automation, the impact on jobs itself will be quite limited, right. There will be certainly the emergence of new jobs like prompt writers. We've all told how prolific, and how those have just captured the imagination of the public.

We've also seen how jobs like that of an intern in advertising agencies have been in some instances eliminated. But those are just probably the 1% to 2% of the impact. The broader impact is going to be across many, many more types of work where certain types of activities will be substituted. I think this democratization of creativity and knowledge actually means that vast amounts of work tasks will be augmented. So even people with a fraction of the skills and expertise that might have been required in the past to get into the game of creating.

And then also, I think recognizing that there will be new types of work created. Either whole jobs like prompt writers, or lots more other activities that are created by the presence of these large language models. The area, Kate, that I'm really intrigued by is the democratization of these tools puts so much power in the hands of talent. And I think what you're going to see is the continuation of this trend away from organizational centricity and process centricity, towards human centricity. And these tools really do drive that power.

And just I think one example of that is looking at what might be a massive boon for gig workers. Generative AI I think, will transform the work of gig workers by putting exponentially more power and knowledge in their hands. It's going to enable them to perform much higher order work, as it substitutes core activities like information and knowledge gathering. As it augments their critical thinking and their solution development expertise. And as it creates more space for them to work on or acquire new skills. I think the impact on gig work, Kate, is going to be truly fascinating to watch. And as we look at higher road activity being shifted outside the enterprise.

Ravin, I always love talking to you because I think you give us that hope that it is a race to the top, not the race to the bottom. And I think that is so important. And you're right, the headlines have been stolen by the impact, particularly in marketing and some of these research functions. But really, that's just a fraction of the reverberations that we're going to hear in other areas.

Ravin, you also mentioned prompt engineers, and outward auditors. I think there's been a lot of discussion about new jobs coming online, or maybe more accurately, new skills. Sophia, I wonder if you could shed a little light on some of the technical roles that you're hearing about, and maybe even a how some of the job adverts that you've been seeing have changed given these new requirements coming in as people begin to work alongside generative AI?

Yeah, it's really, really interesting to see changes just in a few months. So a few months ago, front end developers were expected to have specific front end programming skills. However, as of now, I know some tech companies have revised their job requirements with the latest being the ability to communicate with Chat GPT. So a variety of firms, industries are hiring prompt engineers, which is quickly becoming the hottest new jobs in tech. And probably you read from the news, there is a prompt engineering opening with a salary more than $335,000 without the need for engineering or coding expertise.

And the job description saying there is an art, planning elements of programming instruction and teaching. So I have a question myself. Whether it's really a tech job, or continues to be a tech job when English is being now regarded as a new programming language. One more thing is about cybersecurity, the effect is evolving into the most dangerous form of cybercrime. And we know that data privacy, security is the number one barrier for enterprise adoption right now. So I think the demand for a cybersecurity professional will continue to rise.

Sophia, earlier you mentioned a number of experiments that are happening across Marsh McLennan. And I wonder if you wouldn't mind sharing a few of those insights. What's been some of the learning as you've been applying these large language models to some of the challenges that organizations face?

Sure. So our experiments shows that generative AI has the potential to produce commercial output that closely resemble human written content after a few quick rounds of fine tuning. So again, fine tuning is a very important skill. It also can summarize and synthesize information to support research activities. So the technology has potential to result in significant productivity and cost savings.

So for example, the cost of a human writer for a single plot being $500 with one month turnaround, compared to now just like $20 per month for Chat GPT. And 20 minute fine tuning. Or a typical two hours research work now can be shortened to under 10 minutes. And that means [INAUDIBLE] improves productivity 20% costs.

We always love statistics like that. And it's really interesting to me, I just read an article this morning that said Chat GPT is the best copy editor ever to be hired. But these are very much I think, the types of experiments that I'm hearing a lot about. I think a lot of people are looking at how it can help with research and marketing copy. Ravin, I wonder if I can come to you around some of the other work that you're doing with clients.

As the recent World Economic Forum's future job report paints an extensive picture of just how AI more broadly is extending into the workplace, and changing both work and working. I wonder if you could share some specific examples of how you are helping clients transform, and prepare for, and incorporate AI, and specifically generative AI into their work?

Yeah, absolutely, Kate. We've continued to see a massive uptick in demand for our services around what we call work design. Which as you know Kate, where we help organizations both through a combination of consulting process, as well as AI driven tools that we developed, actually take a body of work, a workflow, a job, deconstruct it, analyze where tools like generative AI, versus RPA, versus more traditional machine learning could be used.

Help the organizations see what the impact of redeploying those activities might be. And then helping them reconstruct new and arguably more human, more impactful jobs. All the while, shining a spotlight on which skills are being rendered obsolete, which skills are changing in how they are being applied because they're now machine augmented, and then what new skills are being demanded as we're using some of these tools.

Sort of in parallel with that, it's also as I mentioned a second ago, helping organizations understand based on our proprietary methodology around looking at the return on improved performance. Where are the bodies of work where we should be using established analytical models like machine learning that is driven by logic and significant amounts of data.

Where can that be helpful in highly critical areas where mistakes need to be avoided, versus where emerging technologies like generative AI that lack logic, that often as Sophia said, are prone to hallucination can be used to maybe augment human capability in areas where the risk quotient is a little bit lower. But all of this also ties in to us working with HR to establish a new set of guardrails around the future of work. And those guardrails really fall into five buckets.

The first is something Sophia talked about at the beginning, data protection. There is still an open question about the potential risks associated with the source data that's been used to train these generative AI models, and the potential for lawsuits that from the creators of some of that source data.

Yes, the copyright.

Exactly. Has not been addressed. So as you use these models and create derivative works, ensuring that you understand the risks. At the same time, and exactly as Sophia said, ensuring that the output of your work is not harmful to others. So that's one category of work. The second is understanding the consequences of your use of these models on your overall work model. So how do you create a work operating model with tools and disciplines to analyze work, and sustainably and responsibly apply emerging AI and automation? Dealing with some of the risks of mistakes that I talked about.

The third is understanding the consequences on your talent. So, so many of our professions, and as the two of you know, our profession maybe more so than many others is built on an apprenticeship model, where expertise is acquired through experience and learning from other more sort of long tenured, more experienced colleagues. How do we resist the temptation to substitute the work of junior level talent with these tools, given the potential negative long term consequences on the talent pipeline?

The fourth has been thinking through how we develop future skills. As more and more AI proliferates into other bodies of work, ensuring that employees are doing meaningful and sustainable work is going to be critical. So how do we find opportunities to automate tasks and free up time for new value adding activities, while ensuring that the talent has the right signals, assets, and resources to keep upskilling and reskilling as the work is continuing to be reinvented?

And then lastly, as an organization, as leaders across the culture, how do we ensure that we've got a mindset and a culture of perpetual reinvention? We're just at the precipice of the impact that these technologies are going to have on work. We need to be ready to keep reinventing our ourselves, and our organizations, and our work. So as the next generation of AI lowest the premium on creativity and democratizes access as we've talked about here, how do we ensure that we are perpetually reinventing our business model, our workforces, and ourselves? And ensuring that is basically how we run our businesses, not be exception to the norm.

Ravin, I love the way you always bring science to chaos. But there is a lot here to think about, and I love the concept of perpetually reinventing in many concepts, whether it is the skills that we need, the work operating models. There was a few things that I just wanted to hone in on there. Your point three about the apprenticeship model I think is really interesting, because today the reason why we can go fast in many areas is we've got that great knowledge and experience. Or as Malcolm Gladwell talks about, thin slicing.

We've got them to bring to bear to say, this engagement survey looks wrong. Or there's something wrong with this column modeling, or this assessment profile. Maybe they didn't complete it. We lose some of that if we hand it over to the machine and don't train our young people. I think that's really interesting. There's probably two things that really stuck in mind my mind.

When you talk about work design, you've mentioned it throughout [AUDIO OUT] building more, I think your language was more human centric, more impactful jobs. And so I'm really interested in what do you mean by human centered? And then the second one that stuck in my mind, you talked about measuring it in terms of the return on improved performance, I think? I wonder if you wouldn't mind just touching on those two points. Because I think there are two really interesting areas for our listeners to hear about.

Absolutely, Kate. So when I talk about more human centric jobs, Kate, you and I know from all of our work with the World Economic Forum that the half life of many technical skills are shrinking. And if you look at what the impact of not just earlier iterations of-- if you look at robotic process automation and what that did to classic swivel chair work in offices. It significantly reduced the premium on data analysis and synthesis. If you look at what machine learning has done, it significantly impacted the work of many analytical roles, salespeople et cetera.

And generative AI is doing the same thing. It is shrinking the half life exponentially of many different technical skills, and it's shifting the premium towards the things where it's going to augment our humanity, right. It's going to help us express empathy. If you think of how Sophia beautifully created that song, I think that's the power of some of these large language models in accentuating and increasing our ability to express love, care, concern, critical thinking. Those things which are truly human.

And doing so both by substituting some activity. So giving us more space. And then augmenting yet others, making us even more productive in those domains. So I think that's the opportunity we have. To your question about return on improved performance. This is a capability that we found to be really important. It was developed originally by my co-author, John Boudreaux.

We've been using it, Kate, in a lot of our assignments. Because it's a truly unique way of helping organizations understand what the relationship between the performance of an activity is and the value of that performance to the organization. Many organizations inherently believe that there is a steady upward linear relationship between performance and value. And the objective function is always more.

More performance equates to more value. What we've shown through all of the research that we've done with almost 4,000 companies, is that there are in fact, four relationships. There is one set of relationships where there is almost a negative relationship. Where the goal of the work is not more, but rather the goal of the work is error elimination. Think of the work of an airline pilot, or a driller on an oil rig. You don't need a lot of creativity there, we just want the work done the same way 100% of the time. Because the consequences of an error are so significant.

The second relationship is one where we're looking to minimize variance. We see this in transaction processing work. We see this in a lot of manufacturing work where the goal is to essentially say, we've hit the right level of performance, we don't want to do it faster, do it differently. We want to maintain. So variance minimization is the goal.

The third relationship is what we call an incremental relationship. And that's where unit improvement in performance is basically a unit improvement in value to the company, like a salesperson. [AUDIO OUT] then would be better. So the objective function is about improving productivity. And then there's a fourth relationship, which often goes undiagnosed in many companies. And it's where there is a massive premium on creativity and innovation.

It's what we call an exponential relationship. So small improvement in performance yields exponentially greater value to the company. Think of the teams at AstraZeneca and Pfizer, working on the vaccine back in March of 2020. A small improvement in performance, massive value to society, those companies. And this is where I think understanding the nuances of these technologies, you want to use tools like Chat GPT and DALL-E 2 in those areas where the risk quotient is low, and they can maybe enable some incremental value. Or they can enable a breakthrough with exponential value.

You don't want to be using them in scenarios where the cost of a mistake leads to significant negative value, like the work of an airline pilot or driller on an oil rig. So understanding the nuances of the work and the objective function can be really helpful in figuring out what tool to use where, and for what purpose.

Well, Ravin, I do hope you're talking to the airline that I'm about to fly with later today. Because I definitely want to make sure that we've got the right improvement outcomes as their work begins to get augmented. But I'm sure like many of our listeners on the call today, I started to apply those criteria to my own jobs and the jobs in my team.

And I wonder if you wouldn't mind just talking about what you see is the impact on HR itself? We know from Versus research that 57% of CEOs and CFOs are increasing their funding in AI and automation. We've read that there's some big impacts with regard to the marketing function. But we've seen a lot of the reductions in force now hitting a jar, and it makes me think that the two are probably related, and the concepts that you talked about in terms of human centric jobs and this improved performance are probably relevant. What are you seeing and hearing?

Yeah. You know, Kate, I think there are two areas where HR is both being impacted in terms of the work it does, and also impacted as it relates to what it does for the rest of the organization. You're absolutely right. These technologies, HR work is knowledge work. They are having an exponential impact on the work that HR does. The HR operating model is being radically transformed.

Kate, as you know, we talk about the target interaction model before, the target operating model. These technologies are transforming the way interactions happen between HR and its various stakeholders, between employees. And it's having a massive impact on the operating model and the service delivery. So HR needs to-- HR is going to be at the leading edge of much of this change. And it needs to take the opportunity to lead the way in helping organizations understand how these technologies should be incorporated.

And it speaks to the second point, that I think as these technologies come in, HR has a massive opportunity to help business leaders rethink how they orchestrate work. Working alongside business leaders to ensure that they're leading with the work, and understanding what we talked about. Where can the work be substituted, versus augmented, versus created or transformed? And not letting the tail of tech wag the dog of the organization. Because the consequences then will only be negative.

I love that analogy. I think it's very visual. I can't believe that our time together is already ticking by. And so I think I'm going to have to move to our final question. Though I know I could speak to both of you for hours. And maybe I can have this final question to both of you. Look, this is an exciting new world we find ourselves in. And there is a lot to get our head around.

But I'd be curious to hear from both of you as experts in your field, what are your predictions on how generative AI will change the face of the future of work? And maybe we can jump ahead to kind of five or six years where we've all got a lot more comfortable around this. What will working and partnering look like? Sophia, do you want to kick us off? And then Ravin, maybe you can close us out.

Sure. So I have a couple of thoughts. We are living in a time unfortunately, with massive layoffs. Especially in tech. So I was thinking that might be good, but they're smart people. Why take this opportunity and try to create something new themselves, because the technology is so accessible. So you can see within just months, the booming of stock ops. So they are really accessible tools now than ever been. So people are scared of job loss by AI. But I see, it opens lots of opportunities.

I hope what comes out of the economic downturn is innovation. More exciting companies, more exciting products, and more exciting new job opportunities. To share with you my daughters comment, those Gen Z on TikTok, they are very optimistic. And talking about Gen Z, when they enter the workforce they are more familiar with AI tools than any of us. And so the choices of companies and jobs may be influenced by the factors such as AI related career opportunities. And they may forgo those organizations that are lagging behind.

One more thought is around, we've been talking about democratizing of AI. These models lower barriers for low code and no code development tools. So I see it creates a lot of job opportunities at the intersection of business and IT in what we call citizen developers. And I think required skill for these roles will likely involve a combination of business domain expertise, with a branding skills, and with some basic understanding of tech.

Yeah, and then also-- I mean last but not least to Ravin's point about humancentric roles. Where people say maybe they're scared that AI would end human creativity. Well, I don't think so. I think human creative-- sorry. Human creativity manifests itself in the form of ideas, theme and concepts. So the future of work I think in the next five years, will witness a dynamic synergy between human creativity and AI collaboration.

I love that phrase, dynamic synergy. I think that is great. And it's so interesting how that developer model kind of gets flipped. Basic technical skills with the human and the business domain knowledge, I think is really, really fascinating. Well, Ravin, you and I might not be of the Gen Z generation, but I think we equally had quite a positive view on the future of work. I wonder if you could help us close out today on your vision for the next five years?

Yeah, Kate. Happy to. And I think Sophia has absolutely nailed it. I think we have the possibility of being at the golden age of creativity. I can see the truly augmented workforce, with more power in the hands of individuals and talent, with the accelerated democratization of work. But I think what we need is to ensure is that we've got the guardrails in place. Because as great as this new world could be, I think the potential for collateral damage is really high. And I talked about those guardrails.

I think equally, and this is going to be a responsibility for every single stakeholder out there, is ensuring that nobody is left behind in this journey. The opportunity to create can't be just limited to those of us who have the means. Those of us who've had the experience and expertise. It has to be an opportunity to bring more and more people into the game. So access is going to be really important for us. But I do think the potential is great if we can create and adhere to these new set of guardrails.

Well, Ravin and Sophia, thank you so much for sharing your insights and learnings. There is so much more to talk about this, that I think we'll have to dip in with you again in a couple of months time. Because it really is an area that is moving so fast. You've also both done a great job of summarizing up. Some of the things that really stuck in my mind for the conversation today has been firstly, the real change that we've seen of lately has been around accessibility. Democratizing both creativity and knowledge.

Ravin, your comments that some of the headlines that we're seeing in terms of job disruption really is just that 1% or 2% of jobs. And there really is that opportunity to augment work, and have more satisfying, interesting work for humans. And I loved your comment around, we're moving from a position of organizational centricity to human centricity. And I thought it's fascinating, the thoughts around how this power in the hands of key workers could really change the platform there.

We then had a good conversation around the importance of work design, and thinking in new ways about the relationship between performance and value. And then we've just ended up now talking about the unbelievable opportunity that we have ahead of us. The opportunity to rethink the interaction between cohorts of people and the HR function, and how that might change the model. And our role as HR professionals in orchestrating work.

I think everybody agrees this is an incredibly exciting time. And the final comment there about, let's make sure no one is left behind in this golden age of creativity. I think really does talk about the potential that's ahead of us. Thank you so much for joining on the call today. Listeners, thank you for tuning in. If you're interested in the topic of AI or generative AI, we do have a thought leadership series on mercer.com. So please do take a look there.

Ravin, I know you mentioned one of your articles there. But if you're interested in this topic or others associated with the new shape of work, we've also got them all up there on our mercer.com site. Thank you, everybody, for joining. Sophia and Ravin, Thank you again for sharing. Wishing everybody a great rest of your day.

[MUSIC PLAYING]