HUMANS IN THE LOOP; ROBOTS ON THE TEAM

Browse more episodes here

READ THE FULL TRANSCRIPT

[SYRIL SMITH GARSON]

Is UI dead? What does it mean? How does experience change in the next five years? So maybe we could just start there. Do we see experience itself changing?

[ADRIAN CROCKETT]

I think, first and foremost, the generative AI momentum and sort of movement is going to be about experiences. It's just that it's about a different set of experiences. So, I think in a decade's time, we're going to look back to this period of time, and we're going to see that there was switching, that the way that we think about experiences, the way that we think about interfaces, such forth, is completely different.

Previously, we were in a world where we had to do a lot of work. For a new UX, we had to spend a lot of money on the GUIs. And we're not going to be living in that world. If we imagine a world two or three years, the future, which, in the world of AI, could mean two months because we are not moving in linear time anymore. There is the distinct possibility that UX is actually generated on the fly. So, depending on what question I'm actually asking, the richness of that response variable, that is where I think we're going to actually see the next evolutions, from an experiential perspective.

[SYRIL SMITH GARSON]

Yeah, so maybe I'll start with the chat piece that you mentioned too because I think that way as well, right? I think the most natural way that all of us think about GenAI is chat and chat first, and what does a copilot mean? And, of course, we're all building chats because I do think that that's a more human way to be able to start to interact with a generative AI technology. But do we think chat is it? Do we think copilot is it?

[ADRIAN CROCKETT]

I think the word copilot is going to expand. So, we really started off thinking about copilots being a chat interface, in many respects, being specifically human in the loop. Over time, I think we are going to move on from that quite significantly. Chat will be one element. Dynamic UX is going to be one element, but multimodal is also going to come into things. So, the way that we actually experience data, yes, we've started off with natural language, but that's going to evolve significantly.

So, when I think about actually asking for data on a company, I probably don't want a long-form, text-based answer. I come from financial services. I'm used to having data compressed in a visual format.

[SYRIL SMITH GARSON]

Right. So you can see it, analyze it.

[ADRIAN CROCKETT]

You can see it and analyze it. And then once you've seen it and analyzed it, you're going to want to tweak it. You're going to want to play around with it. So, I like to think of those response variables as evolving to micro apps, in many respects.

What comes back, I'm going to assume —so therefore we need to make it true— is that the sort of interactions that I can have with that response variable are also the same sort of experiences that I previously had. So, if we take something super, super simple, I'm asking for the share price of a company. So, I always think about this from a design perspective as asking the five whys, right? What are the next set of questions? If someone asks me what the share price was, they didn't actually want to know what the share price was. They actually wanted to know, what was the relative movement? How do they think about that from a time-series perspective? How do they think about that versus the peers? All of those things need to be embedded in the responses.

[SYRIL SMITH GARSON]

And give you context.

[ADRIAN CROCKETT]

100%. That context is going to be critical.

[SYRIL SMITH GARSON]

Huge. I think one of the other ways that we're starting to think about this from our side as well is you've got the chat. But if I do something every single day, I don't want to have to wake up, log on, and chat it. So how do you also see the experience transforming into more of these action oriented —maybe it's doing something on your behalf. Maybe it's just a packaged set of prompts and an action button, something like summarize, and you know you do that every day, or you start your day and you want alerts and notifications. How do we start to use generative AI in that?

[ADRIAN CROCKETT]

So, I think there's almost two different dimensions that you can think about when you're answering that question. So, the first one is really the element of, well, I do these tasks on a repetitive basis. So rather than just thinking about copilot experiences as one off —I've got a question about X. How do you actually package that up to represent a micro workflow? I don't think it's a massive workflow, and the atomization of those workflows is going to be very powerful in and of itself. So, I think there's a shift there, and that's why when I use that word micro app, that's where I think that connective tissue is going to be there.

So, you gave the example of giving next-best action prompts and such forth—

[SYRIL SMITH GARSON]

Exactly.

[ADRIAN CROCKETT]

--embedding all those elements. So that's one dimension.

I think the other dimension is we often think about copilots today, at least, as pull events. I go in and ask a question. Why does that have to be the case? Why can that also not be a push event?

So, in the example when I wake up in the morning, my copilot should actually be helping me get prepared for the day.

[SYRIL SMITH GARSON]

I agree.

[ADRIAN CROCKETT]

But it should have read my calendar. It should be presenting all of the things that I actually have to action. What is the news that I needed summarized? What are the internal documents? What are the equity research reports that I need summarized to get ready for that next meeting?

[SYRIL SMITH GARSON]

That goes back to your dynamic UX type of experience where all of those different micro workflows —I really like that phrase— come together in this one type of experience. And I want to go back, actually, to something you had said before around the experience changing in this dynamic experience. One of the things you mentioned is this micro app concept. I think about, how do we see apps in a year from now? Do we still imagine that we open up our phone or our laptop and we still have a set of apps that we go to? Do we think all of this is going to be abstracted away from a copilot type of experience? How do we think about this world of maybe some layer that sits on top of these applications?

[ADRIAN CROCKETT]

I think the word layer is the really critical element there. So, when we're thinking about an app, there's obviously multiple layers in that app already. We've got the back end. We've got the front end. If we just simplify things, I think the back end is probably not going to change. It's going to evolve quite significantly to support this next generation of experiences. But that front end is actually going to be quite different.

I do like that analogy of those layers and the copilot effectively sitting on top of everything else. Now, that then becomes a very interesting question. Is there one copilot, or are there multiple copilots?

[SYRIL SMITH GARSON]

Absolutely.

[ADRIAN CROCKETT]

So, is it that I've got one set of experiences —so let's just call that a search bar for simplicity, right? So, the question of where we are in a year versus where we are in 10 years I think is also a very different element.

So, if we imagine a 10-year-out perspective, I foresee a financial-services desktop as being something analogous to a search bar. I don't think you need anything more than that. Now, it's then that interesting question. Whose copilots are you actually calling on? What data and plugins do they actually have available to actually come back and respond? I think that's where we're actually going to see the massive amount of evolution, and this is going to be, I think, the big difference between this technology change versus all the technology changes that we've seen before.

So, if we think OpenAI, a hundred million people within two months. Never happened before. But the implications for people like us are that the people in the lines of business or the rooms of the house, as we'd refer to them at Microsoft, they're actually coming to us, and they're actually demanding things because they've experienced things on their phone or their laptops, in their everyday life, and very, very quickly, they're coming back and saying, this could actually make a difference for me at work. This could augment my work. And they're the people who are coming up with these things. So, I think as a technologist or a product person, the challenge for us is very much, how do we actually facilitate that but simultaneously provide the security and compliance in regulated industries that we also need?

[SYRIL SMITH GARSON]

Microsoft and LinkedIn just published a study that said something like 78% of employees are using AI at work. And I think then like 75% of that 78% don't tell their employers. And so I think you're 100% right that we're seeing this really sort of bottom-up push. I think it gives us a real opportunity to be able to design. I like the components analogy, the different skills, if you will, that need to come together into this one search bar.

One thing I've been debating as well in this one search-bar example, one layer on top of all of the applications, is do we still see a world where I'm able to start talking to a copilot experience, get some action, but then I want a deep dive. I want to be able to start to do more analysis, or I want to take it off of the copilot experience. Do we still see that sort of front-end layer existing?

[ADRIAN CROCKETT]

So, I think, again, this is a question of time period. I think in the very short term —so next one to two years— we are going to see a lot of what I refer to as springboard moments. So when we actually ask the question, the response interface is going to anticipate, and the better we do this, the more often we come to a joyous outcome. But in many instances, it's actually going to be a little CTA button down there that says, do you want a deep dive?

Now, it's not going to actually say, do you want a deep dive? But sort of I think the pragmatic thing is it's actually going to come back, and it's going to have a button to Aladdin. It might also have a button to Power BI, and it might also have a button to Excel. Depending who you are, the way that you actually then investigate that question is going to be slightly different.

The quicker that we can actually get those mini apps inside the response variables so that people can actually play around with things —so I gave the example before of asking for a share price. Well, what's something I might do with a share-price chart? I might change a period. Well, that should be a button that's in that micro app. Well, maybe I want to actually add a comparable company. That should be a button that I can actually do that.

So, I'm very much in favor of thinking about those as micro apps and asking the question, if someone went into a traditional app, what are the three things that they would do? And almost using a Pareto rule or 80-20 rule to say, how can I actually get to job completion as quickly as possible?

[SYRIL SMITH GARSON]

Well, I love that notion of these workflow applications today. I think what we're seeing across the board is a lot of the work that we're doing at BlackRock, also thinking about within Aladdin, really starts with, what is the workflow today? How do we go from 10 steps to 3 steps? How do we start to anticipate the next-best action for somebody?

I think one of the notions that I'm really curious about starting to think through is while we have to continue to augment existing workflows today and do that 10 steps to 3 steps-- because those are the tools that everyone is using today —do we see a world where you can actually step back from a product perspective and say, how would I have designed this from the get go if I had AI in my toolkit?

We're seeing examples where we have our own proprietary code or our own proprietary language or these workflows that are very idiosyncratic, right? And we're thinking, how do we optimize that idiosyncratic workflow? But I think had we stepped back and said, actually, how would we just redo that entirely with generative AI? I think that will be a really interesting paradigm shift as well to say not just augmenting but transforming workflows.

[ADRIAN CROCKETT]

I'm 100% in agreement. A lot of what is happening today is saying, I used to have humans. I've still got the humans. And I used to have existing technology capabilities, and now I've got a net new set of technology capabilities. So how do I actually augment that so the 10 steps down to 7 steps and then down to 3 steps, et cetera, et cetera, is how we will traditionally think of AI augmentation.

I think there's a different element here that we need to start thinking about, which is –is the copilot themself actually a team member? So how do we actually think about organizational design?

And if I had a combination of highly talented humans, SMEs, or subject-matter experts, plus also copilots, how would I actually go about getting to done if I were actually redesigning it with those capabilities? When we build all of the copilot experiences in Teams, for argument's sake, had previously been me interacting with a copilot. Well, now if me and you are chatting, if we were doing this in Teams, we would actually be able to invite a copilot as a third team member.

[SYRIL SMITH GARSON]

Interesting.

[ADRIAN CROCKETT]

That changes the dynamic quite a bit. So, I think there is a combination of really reimagining workflows altogether rather than tweaking them, which is where a lot of people are starting, but also from an organizational design perspective, thinking about where that new team member actually fits in and how we can best leverage them. If we've got an incoming analyst, for argument's sake, we ask the question how they best fit in with our team. How do we actually train them up to have the biggest impact? I think we should be asking exactly the same question with the copilots.

[SYRIL SMITH GARSON]

I love that notion of we adapt to the technology. It's not natural to be pressing buttons with your fingers or typing like this. We're used to writing, speaking, chatting. And so I think this idea of now taking technology out of just an interface and starting to see it sit alongside of you is a really powerful, almost real-world realization of these copilot experiences in a more actual way.

Like I imagine we have a copilot sitting with us here today. I think I would pause and say, can I trust what that copilot is going to say to me? Do I believe that that copilot is going to do the job I need it to do? Or how are you guys thinking about trust and making sure that there's some confidence in the copilot?

[ADRIAN CROCKETT]

So, I think getting to trust is going to be predominantly about transparency in the short term. So, making sure that we've got human in the loop on the vast majority of operations, particularly in regulated industries, is going to be really, really critical.

Over time, we will see the evolution of the copilots or the agents actually performing simplified tasks by themselves, but we're going to break that down. We're going to go through a process that will initially be a human in the loop process. Once we get comfortable with that workflow task or that skill, then maybe that can be its own independent agent. But that's going to be very much an evolution, and I think it's probably going to be a little bit slower in financial services, even though we probably need it a lot more than some other industries. But I think in the short term, that's more about transparency to actually build out that trust, at the end of the day.

[SYRIL SMITH GARSON]

We're very focused on that transparency piece as well. How do we make sure that you can say, well, what is my confidence in that response? How did you get to that response? Can you show me how you got that answer? Can you cite me back to the original source? So that's a big area of focus for us too. I think you're right that as we start to move into more actions and more complexity, making sure that that trust is already established and already there is really important.

[ADRIAN CROCKETT]

And I think one thing is really critical here. Not all tasks that we perform are of equal importance or need the same confidence.