Microsoft over the past few years has released some of the most interesting research about how people work, including analysis of the explosion of meetings since the pandemic started and the “triple-peak” workday phenomenon.
Now, building on its tight partnership with OpenAI, Microsoft is adding features to the many tools used in workplaces to introduce generative AI functionality. These new capabilities, including Microsoft 365 Copilot, aim to further change the way that knowledge work is done.
To better understand the implications for work and management, we spoke with Jared Spataro, Microsoft’s corporate vice president for Modern Work & Business Applications. Here are excerpts from our conversation, edited for clarity:
How would you characterize where we're at in the evolution of knowledge work? And how do you see how we spend our work days evolving over the next two or three years with the rollout of AI tools?
We're in the early stages of it. In many ways, the last couple of decades have really essentially been the digitization of analog work patterns. The spreadsheet, as an example, was a digital representation of a ledger book. Word processing evolved from the typewriter. You see these physical analogs that we have digitized. And in many ways we were kind of reaching the diminishing returns of that approach and now we're starting to tip into a new era where we're truly digital. We're really thinking about what we can do differently when we're truly digital. And we're realizing, especially because of the pandemic and the massive transformation that we've had of all sorts of different types of communication that we can apply AI to many of the tasks where we have thought, well, this is just the way it's done. So it's exciting for me because we're kind of coming to the close of a first phase of digitization and really entering into a new and very exciting phase.
What are a few things that we do today at work that you think we will stop doing within a few years?
If I take a step back and look at what in particular AI is doing for me and for the folks who are just getting their hands on it, there are a couple of patterns that we're seeing. Number one, we definitely are seeing AI help us find the signal amid the noise. One of the really interesting facts that we uncovered in analysis of our telemetry was almost 60% of the average worker's day is spent communicating or coordinating with others. And that percentage continues to grow. People feel this, for sure. The way they express it often is, I have a whole job to get to my real job, to be able to do the thing I was hired to do. So I see a pattern emerging where AI is going to help us cut through all of that.
It's not like we'll communicate less. But we'll be able to find the most important aspects of the communication to help us get our jobs done. I also am really intrigued by how AI is helping us spark creativity. We'll see entirely new patterns of human-machine collaboration emerge. I don't think that means that we'll collaborate less with people, but it will improve our ability to be creative. That's a unique thought for me. For many years, I thought that machines just wouldn't be good at creative tasks. Then finally, meetings will be pretty transformed. It's not to say we won't meet any longer, but I'm finding, for instance, I attend fewer meetings because a meeting has become less a point in time and almost a knowledge object that I can query, that I can ask questions to. And that's a whole new way of thinking about human interaction. So those are three things that will be pretty transformed in the coming years.
How specifically do you see AI increasing worker creativity?
There's the superficial task of taking a germ of an idea and fleshing it out. That sometimes is a really difficult task for people to do. They have the spark and they want to go from spark to more of a skeleton with some flesh on it, so they can imagine what the idea looks like in a more full form. And I'm definitely seeing that starting to happen. The one I appreciate the most though—which goes back even further in the creative process—is coming up with the right sparks. I found that I used services, agencies, and people over the years professionally to help me generate ideas. It's almost like a funnel of ideas and I'd get down to a smaller and smaller set until I locked on something that I'd wanted to pursue in the work that I do. I'm finding that I can open that funnel broader up top. I can have turns more quickly. It really has fundamentally changed the way I'm thinking about that early process of creativity as well. And again, the most surprising thing for me is I didn't anticipate that a couple of years ago. I honestly thought AI would be better at automating routine tasks than it would be at creativity.
You described meetings as 'knowledge objects.' Is that because AI is able to easily transcribe and summarize them?
Yes, but there's one other aspect. Transcription technology is getting amazing. It's not error free, but neither are humans when they transcribe. And in my experience, it is far surpassing what a human can do. The summarization is incredibly valuable, but for me it's the next step. A summary is good. It gives you an outline of what happened, but oftentimes you want to know more about the human interaction. I'm amazed, for instance, in some of the tools that we're releasing right now that you can ask very specific questions. Was my name ever mentioned? Was this topic ever mentioned? What was the group's sentiment on this? Not only what decisions did they make, but how did they feel about those decisions? Were there any dissenting opinions? You can get such a fine-grain analysis of human interactions that it really opens your eyes to this idea of wow, it used to be that a meeting happened and it was over and people maybe took sketchy notes. These days, that very full interaction can be looked at from many different angles.
That's both live and retrospectively? The current technology allows that?
It does, and both are really amazing. The retrospective approach is kind of one of those, 'Wow, I don't think I need to attend all the meetings I intended to or attended previously to make sure I just was up to speed.' So that changes my meeting attendance patterns. But the thing that the team told me that it took seeing to believe was what the technology would do in the midst of the meeting. In fact, I had a really nice friendly tussle with my product group on this. They were like, 'Jared, you've got to see this. The way it changes the dynamic of a meeting is really something to behold.' And I told them, 'Guys, the real value prop here is post meeting.' And then they said, 'No, just try it.' The first time I did it, I was amazed because the GPT models these days are so good at even determining sentiment based on a real-time transcription that's happening. It never fails to impress when I show that to customers.
Stepping back, what AI capability or feature do you think will have the greatest impact on knowledge workers over the next few years?
This one's pretty easy for me. People ask me this question as if, 'gosh, Jared, I bet you have a whole bunch and you're not quite sure.' Nope, I have one. It is the emergence essentially of a ChatGPT-like tool for your business. ChatGPT is amazing because if you ask it questions about the material that it was trained on, it gives you surprisingly well-reasoned answers, sometimes wrong. And we can talk about hallucination, but surprisingly well-reasoned answers. Answers that can help you if you're trying to do specific tasks. Today we have no such thing as the equivalent of a ChatGPT for your business where you could ask it everything from Q4 sales to current trends to its prediction of how the quarter will end for a certain product line.
There's no such thing. We have no competition for that. That technology is emerging right now and it will come here much faster than people think. It will reshape job design, roles, even dare I say the operating model of a company. Because so many roles are essentially fashioned to make the company move forward and pass information along so that good decisions get made on some of the key aspects of what it means to make a product and sell it. So I think it's going to be that. I think people in finance to marketing to sales are going to spend less time in the traditional applications that have emerged over the last couple of decades and more time simply querying their company, Hey, what about this? What about that? What if we did this? How should I think about this? Do you have any recommendations on that? And that's a brand new way to think about running a going concern.
When you say that will happen much faster, is that a year or two years away?
Much, much faster than that. We introduced these ideas in March of this year. You should expect people to have them in production easily this fall. By the end of this summer we'll be showing some of what customers are doing with it. The tech is there. At the end of the day, the basic idea here is that you take a GPT model, you do what we've introduced a couple of months ago called 'grounding.' So you ground it in your data. It is simply bringing together in a very performant way, your data and that GPT model. That's the architecture of a co-pilot and people are building them and we're working together to build our own right now.
With the new AI tools that hallucinate and make mistakes, people are having to decide whether computers can be wrong some of the time and still be useful. Will that be the case for the foreseeable future and should we be comfortable with that?
Yes. The term that we've coined as we work with our products and with our customers is 'usefully wrong.' We actually tell people when it is wrong, it's often usefully wrong. The same way that someone who's working with you might have an idea that's not quite perfect or a viewpoint that isn't well informed, but the back and forth is what drives value. And I would say here is a really important discovery or kind of insight that I've had over the last couple of months: Most people are used to working with a computer like they do a calculator. I'm going to enter in some sort of deterministic question. Give me the answer to this question. You should give me the right answer every time. With these new foundational large-language models, you essentially have a general purpose reasoning engine. You give it information and you give it a prompt or a question. It's going to reason in the best way it can across that information and try to come up with an answer to your question. It might be wrong. What we're trying now to teach people is that even when it's wrong, it's very interesting to trace why it got there. How did it put the pieces together and how does that advance your own thinking? That's just totally different than thinking about using a calculator where you're just asking a math question.
Is anything different required from managers to succeed as work evolves over the next few years?
Oh, we're at a real turning point for management. This has been predicted for a number of decades too, but the management philosophy, structures, theses—all the models we're taught in business school—are all based essentially on post-war industrial complex approaches to organizations. Command and control was the first, it's kind of one of these very big approaches. But beyond command and control, even with some of the smaller, more nimble organizations, they've largely been modeled off of traditional structures. AI I think is going to disrupt that. Certainly distributed and flexible work is already disrupting that. So I think we're going to look back in five and 10 years and say, wow, the manager that emerged from the confluence of those technologies is just an entirely different person. They need to know how to manage the time and energy of the people in their organization across time and space. They have to be able to recognize where augmenting human capacity with machine-based capacity is going to help them get the job done faster, better, higher quality. All of those factors are things that we've never really factored in. So while I can't predict exactly what that will look like, it's easy to predict with 100% certainty that the manager of the next even two or three years is going to look very different from the prototypical that's been trained by business schools over the last few decades.
Executives are starting to be called on to have views on the implications of AI for talent strategy, including in the context of budget and strategic planning for 2024. How should they respond?
The best pattern I see right now out there is to jump in and start using it, form your own opinions, be willing to experiment with the technology to understand what it can do and what its limitations are. If I get specific for a moment, my recommendations, as an example, are you shouldn't have a marketing department that isn't aggressively experimenting with AI to do everything from targeting (how you isolate in the data the people you would target) to content generation (certainly content can be generated very effectively) to creative. If you're looking at the sales function across organizations, I wouldn't have sellers who aren't using AI to help them be more effective in their customer interactions, to find the right information as they're moving through the sales process. Essentially, we can go through each function and that's what we see most of the forward-thinking companies are taking function by function and looking at the application of AI in a pilot sense.
Then rather than taking what they're reading in the press, they're finding the contours of the state of the art, and they're just saying, 'Great. I think we can reduce the resources we have, for instance, on copywriting by even just 10% to start and see where that takes us.' Or 'I think we can use generative AI in a image-producing sense to help us out in our graphic design. Let's embrace that rather than fight against it.' For me, that's generally the strategy that seems to be the most rational right now. Nobody knows exactly how this will play out, so you have to get your own feel for it.
Do you expect AI to lead to a net increase in the quality and quantity of jobs? And how can people and organizations ensure that is the case rather than AI being used primarily to automate employment way?
There's a thing in economics called 'the lump of labor fallacy,' which is there's only so much labor out there and if you replace some of it, then you're going to displace workers. And over the years, the good news is it's been proven to be a fallacy. That's not what has happened in economies across the world throughout history. When we have labor-saving devices, economies find ways to use them, and as they use them, they grow and as they grow, they produce employment. So over the mid-term and the long-term, I don't have any concerns about this technology in terms of what will it do to labor markets and what will it do to our economies. I feel great. In the short term, there certainly is displacement. So it would be disingenuous to say that it's not going to have, in some cases, an effect that will displace.
But from my perspective, as you look back in history, the patterns again are very clear. What do you do when a disruptive technology comes in? You embrace it. You decide that typically it's ushering in a very interesting growth-driven era and the best way to adapt to that is decide you're going to learn. I'm going to adapt to this era. And humans are incredibly adaptable. So even in the short-term, if you will take that approach as an individual or as an organization, you do very well because there's essentially a wave of energy sweeping into the economy. So I'm personally, we generally as a company are very tech optimistic. We really feel like you shouldn't kind of operate from a place of fear here. You should certainly operate from a place of respect and thinking about what are the limitations of the technology, but there's so much that we can do together if we'll embrace the technology.
Historically, underrepresented groups, including women, have been left behind during technological shifts. What do you think is needed to ensure that doesn't happen?
I'll just go back to a very powerful idea, which is you have the choice when a new technology becomes available to decide what you're going to do with it. The exciting thing about what we're seeing with AI is it's broadly available. ChatGPT was the fastest-growing consumer technology we've ever seen. Less than 90 days to over 100 million users; today, over 200 million users. So when you think about access, it's kind of the broadest access we've ever seen in the world. Of course, we want to do more for access, and that's why we have helped with funding OpenAI so that it's available for free for anyone all across the world to use for any use case that they'd like. But at the same time, in addition to availability, you have to teach people the right attitude. And I'll just go back to that it's a growth mindset. It's an embrace-and-extend mindset. If you see an opportunity in a technology shift, you'll be able to take advantage of that opportunity.
You mentioned in your latest research the idea of AI aptitude. How can AI aptitude best be taught, and how do you ensure that it stays sharp as a skill when AI itself is evolving so quickly?
It is one of the challenges of technology today. I referenced the rapid rise of ChatGPT. We've never seen anything like it. The good news is, what I'm finding in my interactions with the rising generation is that they're quick to embrace. I think that's what you do, you teach people to embrace. I'm very heartened when I see people in education who are working with their students and saying, rather than fight against this, let's embrace this. What can you do? Let's learn how you can use it to become a better writer or to become a better thinker. And that is what I'd encourage society to do. Don't run away from these technologies. Don't worry about kids cheating with them. Instead decide we can work together and learn how to use them effectively. Most importantly, don't operate from a place of fear. Operate from a place of optimism and a place of opportunity. If you do that, things not only will work out well, we're really going to take the surplus you can generate with these technologies and be able to use it across the globe.
The handbook for this new era of business doesn’t exist. We’re all drafting our own as we go along—and now we’d like to start doing so together. You can sign up here to receive briefings like this by email.