PwC's Tech While You Trek: Artificial Intelligence + Human Intelligence

Tune into this episode of Tech While You Trek to hear PwC Tax Emerging Technology Director David Resseguie, PwC Tax Emerging Technology Manager Amber McKenzie and IFS PwC Intelligence Technology Director Brad Blanchard discuss how fusing Artificial Intelligence with Human Intelligence is expanding intelligence beyond what either can do on their own. 

Transcript

Adam 00:07):

Hello everyone, and welcome to another episode of Tech while you Trek, I am your host Adam, and today we are here to talk about artificial intelligence. I have with me, Brad Blanchard, Amber McKenzie, and David Resseguie. Please introduce yourselves, say hello, and tell us a little bit about your roles with the firm.

Brad Blanchard (00:23):

Hey, I'm Brad. I practice machine learning within an IFS organization that focuses on delivering actionable intelligence to firm leadership, and I've been with the firm for about four years.

Amber McKenzie (00:32):

I'm Amber McKenzie. I'm in the emerging tech group in tax. My background's in linguistics and computer science, so we work on interesting problems in those domains.

David Resseguie (00:42):

I'm David Resseguie. I'm the director of emerging technology in tax, and also part of the AI lab.

Adam (00:48):

We've been hearing a lot lately about automation, and we've talked in previous podcasts about how AI can help automate additional processes across the firm. But Brad, you think there's a larger role for AI to play when we combine it with our own human intelligence. How do you see the power of AI coming together with the expertise of our staff in that practice?

Brad Blanchard (01:07):

That's right. I mean, AI has been hugely capable of automating simple processes, being able to take words from documents, take images, and process them down to create insight. But we have very smart people at PwC, and for us, being able to scale out the competencies and the capabilities of those people is critical, and AI is really allowing us to take very complex problems of our clients, and be able to simplify, and really help us understand all the nuances of those problems, and help us narrow down the actions that we can take off that information.

David Resseguie (01:36):

The way I like to say it is that I think we should be letting the computer do what the computer does well, and let the human do what the human does well. It frees our people up, if you will, to actually focus on the things that make them special, like where their expertise is.

Amber McKenzie (01:49):

I think a good example of that, we've been doing some work with M&A on their due diligence process. We asked them to work with us and break down what are the pieces that go into a decision for a due diligence engagement, and then we're putting together a tool that essentially incorporates some NLP, some AI machine learning, that says, "Okay, for each of these steps, we're going to take in the data that goes into it." And some of it, we end up with a question that the human needs to answer to move that process along. But the whole pipeline essentially works to automate what they're already doing.

Brad Blanchard (02:25):

And it's about listening, Amber, to your stakeholders. I was meeting with a partner in tax up in New York about six months ago, and he was describing the opportunity ahead of them to say they want to be able to predict the opportunity value potentially for these service offerings. The way he described it was we've got this individual on our team who has a great pulse on these opportunity values, and if we could just find 10 or 15 of those individuals, we'll be all set.

Brad Blanchard (02:50):

To me, that's kind of queuing up some language around what if we could just take that person, and transfer their mind into the code, and be able to not make 15 or 20 people, but to make tens of thousands of opportunities available to us on the fly. So being able to understand the nuances of the language, listen to people, and then tease out where they maybe were already laying the foundation for an AI process.

Amber McKenzie (03:13):

But don't worry, AI is not to the point of us downloading your brain into the computer, yet.

Brad Blanchard (03:18):

Not yet.

Adam (03:19):

Amber, your background is in natural language processing, so talk to us a little bit about how NLP fits in with these AI solutions we've been discussing.

Amber McKenzie (03:28):

NLP is a really broad field, and I think people don't understand that sometimes, but it handles everything having to do with any natural language, so speaking, writing, forms. You think about, as an accounting firm, how many different forms we have, and all of the processing that we have to do on all those forms, and how many man hours go into that. We use natural language processing to essentially read the text, to pull it out, to make interesting connections between different parts of the text, and then to provide it back to the user in a really usable format.

Adam (04:01):

Could you guys talk to me a little bit about how all of these things fit together, and how do you decide when what type of AI is an appropriate solution for a particular problem?

David Resseguie (04:10):

The way I think about it is really AI is an umbrella term over a lot of these other things that we're talking about. AI really is about how do we teach computers to sense, and think, and act like we do. That's not so much that it's making a decision on what should I do next, so much as learning to look at data and recognize patterns within it to surface up, maybe, here's some decision making points, we want to act on that.

David Resseguie (04:34):

AI is an umbrella term across all of that, and then things like machine learning, NLP, computer vision, some of these other areas like that, are all individual areas of research that handle certain types of problems.

Amber McKenzie (04:48):

What we need from the firm is not so much people who completely understand AI, who understand machine learning, and all of that, but we need for our people to recognize that these things are out there, and to know enough from a technology basis to say, "Hey, I think this problem warrants something greater than business analytics."

Brad Blanchard (05:09):

I would say that AI doesn't solve AI problems. It solves business problems, right? So we, the people within the firm, understand our clients, understand our domains really well, and we need people to be able to break down the way they approach decisions.

David Resseguie (05:22):

The first step of that problem is just recognizing that you have a potential problem that you can apply AI and machine learning or one of these other techniques to. You recognize that we're taking client data, and having to map it to some internal standard that we use.

David Resseguie (05:36):

That's a classification problem that machine learning's very good at helping with, right? Recognizing those types of problems in your everyday work, and then raising those up, as Amber was saying before, let's bring that to a team that can help me think that through of what data do we have, which technique would apply, is it really a good problem?

Adam (05:53):

In your experience, is there resistance? Is any aspect of people not seeing potential AI solutions this idea that they don't want to basically make themselves redundant?

David Resseguie (06:04):

I mean, I'm working in tax directly with a lot of partners and directors and other staff pretty regularly here, and I don't see that fear of this is going to replace me, so much as people just not understanding how they best can apply it.

Brad Blanchard (06:17):

I think it's understanding how the outputs of your process are going to be used, right? Sometimes we think about using AI to answer the question, but sometimes understanding how AI navigated the decision-making process, and generated a result, is actually information that can be used to exploit the outcome. Understanding the patterns and the leading indicators, those are things that really inform our partners on how they go and have a conversation with a client. That's been a transition I think I've seen since I started here four years ago.

Brad Blanchard (06:44):

At the very beginning when I joined, people wanted to interpret. They wanted to be able to make sure that the machine learning process actually reinforced what they already knew. Now I think we're transitioning to where I think people have that trust that David mentioned. I think now it's about how can we scale, how can we make use of it?

Brad Blanchard (07:00):

There's a famous quote by George Box who says, "All models are wrong, but some are useful." I adhere to that in all that I approach because I want the things that I'm building to be useful. It's not about perfectly answering the questions, but finding some utility in the way that we've been able to meet that solution to that stakeholder's needs.

Amber McKenzie (07:17):

When I first came, and it was less than two years ago, I went through our initial training, and I asked a question about, you know, "Oh, well we're bringing in all these STEM people. How are we changing our culture to be able to work with the STEM and tech culture, which is different from the accounting culture?" The prevailing attitude there was they assumed that I was concerned about the STEM culture, and they were like, "Oh no, no, don't worry. It's way easier for accountants to learn tech than it is for tech to learn accountants, so your job is safe."

Amber McKenzie (07:52):

People are understanding that tech is not coming in to replace everybody. That would be ludicrous, right? But lots of people are out there looking for who do I go to to get help rather than, oh, no, no, no, I need to keep this secret or someone's going to take it from me.

Adam (08:16):

Well, this actually transitions very nicely into the next question. Talk to me about the concept of human intelligence. How is the firm taking advantage of human intelligence and building it into the various AI solutions that we're offering to our clients?

Brad Blanchard (08:29):

AI is dependent on a lot of historical data or being able to generate data to then fuel those models. Where there is an absence of data, or where the data that has been captured in the past is not reflective of what the future will look like around regulation, and tax, or things like that, those big fundamental shifts in the way that the data has been organized over time creates a world where you just can't necessarily generalize that historical data into the future.

Brad Blanchard (08:53):

So humans come in, and make a big impact by being able to... kind of like an autopilot, right? A pilot then will let the instruments manage most of the flight, but then at some critical point, landing the plane eventually, they'll take control and they'll land it where there needs to be more of a human process involved to account for all the different parameters that may be playing out.

David Resseguie (09:13):

At its most basic level, AI machine learning is about data. That is the way that AI senses the world around it. In our world, our firm, data comes from humans. We are creating that from our own expertise, our own experience. We're building that out from, like you said, both historical and also thinking through what changes today mean to the future. All of those things that we're capturing from our people is what we're using to train and build out these models.

Amber McKenzie (09:38):

But also, there we have an opportunity. The problem that you end up having with historical data is that it's historical data. Over time, history gets further and further away. If you've created a model on data from 2018 to 2020, what happens when you're at 2025? Is that data is still relevant? So we've actually got a number of frameworks and things in the works that are putting in some computer learning and is saying, "Okay, how do we make sure that we're capturing new data that's coming in, and rolling those into the model, so that the models are not just static and unchanging, but that we're molding them into something that'll work for us in the future as well."

David Resseguie (10:16):

That's one of the things, like the AI lab within PwC labs, for example, is focusing on how do we build out the frameworks and the tooling that Amber's talked about to make that easier to monitor and govern some of these models over time, allow people to give that feedback. We're looking at doing an AI-driven annotation tool for example.

David Resseguie (10:32):

So as you want to teach new models how to extract information from different types of documents, or you want to review the output of existing models, and correct them, and feed that back in so those models can continually retrain and get better over time. We are trying to build out tools and frameworks that help with that, and all of that requires people who understand that data from a deep domain knowledge to come in and actually do that, and make those models better.

Brad Blanchard (10:54):

Which is a really practical place for just people to begin to play with this process and be involved in AI.

David Resseguie (10:59):

Yeah, absolutely. Yeah. Yeah, you can learn a lot about how AI and machine learning works by being involved with things like training, testing, evaluation, even if you're not "a data scientist," right?

Brad Blanchard (11:09):

Right.

Adam (11:18):

Before I let you guys go, are you ready to answer some rapid fire questions?

David Resseguie (11:21):

Let's do it.

Adam (11:21):

What is your one bold prediction for technology in the year 2040? I'm going to start with you, Amber.

Amber McKenzie (11:26):

I think that we're going to see a lot more of the computer figuring out what the problem is as opposed to us giving it a problem-

Adam (11:34):

And it solving.

Amber McKenzie (11:35):

... yep.

David Resseguie (11:36):

A lot of our time is spent on collecting data to build the models, and I really think we'll have more AI training AI, where AI's learning to extract the data it needs from our processes to then build the models from.

Brad Blanchard (11:49):

Yeah, I believe that voice in your head will become audible and AI-enabled, so instead of asking Alexa or Siri, I'll just ask myself and be able to allow that to be connected to the Internet.

Adam (12:00):

I like that. What is your source for new technology information?

Amber McKenzie (12:05):

I tend to look at podcasts. There's a TWIML Talk that does AI and machine learning that I like a good bit.

David Resseguie (12:12):

Publications like MIT Technical Reviews is pretty good.

Brad Blanchard (12:31):

Yeah, I really like futurism.com for a lot of my just future-innovative things that are out there, but I also like to be able to listen to podcasts, but also a lot of the things that the AI labs puts out. It seems like I constantly see David liking articles, and that pops into my feed, and I can read those articles later.

Adam (12:47):

In your estimation, what makes someone a leader in the digital age?

Amber McKenzie (12:50):

I think for me, it's an open mind, so you need enough of a background to understand the possibilities in the technology space, but then also know who to go to, and also know when there's a gap in your knowledge.

David Resseguie (13:04):

People who can sit between the business and the technologists, and really bring those things together to solve real-world problems. To me, that's really where the value in leading in this new digital world is.

Brad Blanchard (13:13):

Yeah, I think that analytics translator role is really important for being able to take that business domain knowledge, and fuse it with that technical domain knowledge, and be able to just help us interpret opportunities or interpret business problems in ways that technology can be used to solve them.

Adam (14:25):

Well, Amber McKenzie, Brad Blanchard, David Resseguie, thank you all so much for stopping by.

Amber McKenzie (14:31):

Thanks for having us.

Brad Blanchard (14:31):

Yeah, thanks so much. It's been fun.

Adam (14:32):

This has been another episode of Tech While You Trek, I have been your host, Adam, and we will talk to you next time.

Speaker 5 (14:41):

This podcast is brought to you by PwC, all rights reserved. PwC refers to the US member firm or one of its subsidiaries or affiliates, and may sometimes to the PwC Network. Each member firm is a separate legal entity. Please see www.pwc.com/structure for further details. This podcast is for general information purposes only and should not be used as a substitute for consultation with professional advisors.