Alter Everything

AlteryxMatt · ‎11-22-2023

Ready to take the dive into AI utilization? Whether you're a seasoned professional or just beginning to explore the potential of AI, tune in as we delve into the exciting possibilities that arise when organizations take their first steps into integrating AI technologies. We are joined by Mara Cairo, product owner of the advanced technology team at the Alberta Machine Intelligence Institute, and guest host Heather Ferguson, Sr editorial manager at Alteryx. Interested in sharing your feedback with the Alter Everything team? Take our feedback survey here!

Panelists

Mara Cairo - LinkedIn
Heather Ferguson - @heatherferguson, LinkedIn

Topics

Ep 146 (YT thumb).png

Transcript

Episode Transcription

Ep 146 Driving AI Adoption

[00:00:00] Megan: Hey Alter Everything listeners! We know that turning raw data into insights can be time consuming and challenging, but we've got good news for you. Alteryx can transform your analytics. Alteryx drives positive business outcomes by enabling fast, data driven decisions. And you can try out our products today.

Start your 30 day free trial of Alteryx desktop or the Analytics cloud platform at alteryx.com/alter everything.

[00:00:33] Heather: Welcome

to Alter Everything, a podcast about data science

and analytics culture. I'm Heather Ferguson, senior editorial manager at Alteryx. We've been conducting

surveys around

generative AI adoption with data leaders and board members. So I'm excited today to chat with

Mara Cairo. about her expertise in AI adoption and how companies can get past hurdles to realize the value this technology can bring.

Let's get started.

Mara, thank you so much for joining us today. I'm really excited to sit down with you, have this conversation about AI, hear about your background. So let's start. Why don't you tell us a little bit about yourself and what brought you into this space?

[00:01:16] Mara: Sure, sounds good. Thanks for having me, Heather. Uh, I'm Mara Cairo, product owner of the Advanced Technology team at AAMI or the Alberta Machine Intelligence Institute.

So I and my team work with companies who are a little bit further along on their AI adoption journey and really ready to get some hands on support to start building out machine learning models to solve whatever business problem they've come to us for support with. Amy has a history of about over 20 years at this point, starting really from the academic side of things.

Our Amy Fellows at the University of Alberta have been doing fundamental research in this space for a very long time, and the Alberta Machine Intelligence Institute was really stood up to make sure that that fundamental research is being translated into industry adoption. So, my work is heavily on the industry adoption side of things, making sure that we're translating this really powerful technology into industry applications that will hopefully make this world a better place.

[00:02:22] Heather: So, it seems like a really interesting structure for an organization, especially to come out of a university, but then be focused on the industry adoption. What was the driver behind that?

[00:02:33] Mara: We didn't want this technology and these breakthroughs to stay in the lab and my entire career has really been focused on moving industry along and helping to diversify our economy, make sure that companies have the right tools in place to solve their really tough problems.

And I think with machine learning and AI, there are so many kind of real world applications that can take advantage of this technology that there was that recognition that it's really imperative that we put some emphasis on taking all of that knowledge and expertise out of the lab and applying it to real world problems because it's such a useful tool.

[00:03:19] Heather: And you have a unique background yourself as well. You're coming to this from more of an engineering background versus like a data scientist background. How did that happen?

[00:03:28] Mara: So yeah, I spent about 10 years in more of the hardware space. So my background is in electrical engineering. And for about 10 years, I worked in a clean room building really, really small things.

And there was obviously a very heavy focus on building the hardware. So lots of sensors and optics and It was interesting to see over that period, there was more and more interest put on the data that these sensors were actually pulling out from their environment. So I was introduced to the space just organically because the companies we were working with were more and more interested with putting that data to use.

So that was my first introduction into data science. And then I also was making a career shift. From a very hands on technical person to more of a project manager. So that's how I got my foot in the door at Aimee, was through an opening in a project management role at Aimee. Which I learned was significantly different than the kind of traditional PM in a hardware space.

It's been a learning curve, but I've been at Amy for about three and a half years now and still learning every day, but yeah, that's how I made that transition.

[00:04:46] Heather: So Amy's in such an interesting space because you're working cross industry, cross business. I think it's a really interesting perspective to look at what's happening in the industry.

So I'm curious, what is the current state of AI in industry? Are you seeing any trends or are you seeing any, anything shifting or changing?

[00:05:06] Mara: Yeah, so definitely over the last year or so, there's been a ton more interest in generative AI. This time last year, most of the companies that we were working with and talking to were using the more traditional aspects of machine learning, computer vision, time series, bioinformatics.

But we have seen a shift in lots of the companies that we talk to now that are looking for support. are really looking to use generative AI. So it's awesome because I think it has put it on everyone's radar. It's kind of a tool at everyone's hands now and it's really easy to see the benefits of the technology.

So I think more and more people are understanding like, oh this is actually something that I can see being applied to my business problem. Whereas before You know, there was some hesitancy, like, what can this do for me, right? So, it's interesting that generative AI has got basically everyone interested in the fields.

It's also unlocked more off the shelf tools for people and companies to tinker with, just to, again, play with it, see what it can do, and then for them to be able to realize a direct value. So, that's the trend that I've seen. Over the last year, and I'm, I'm sure everyone could say the same thing, it is a bit of a hype train.

So right now there's companies using AI and then there's AI companies and we work with AI companies who really want the technology to be a core business driver. So with that comes a lot of kind of the customization of models and going above and beyond the off the shelf tools that are the lower hanging fruit.

But there's different kind of company profiles, and I think the more the merrier, the more people using and interested in this technology, the better, and it is a way to, to move your business forward.

[00:07:06] Heather: Have you found that generative AI has been a gateway to more advanced applications of AI?

[00:07:12] Mara: For sure. Even the capabilities that it's unlocked for technical people.

So it's, it's made it easier to say, generate data if data doesn't exist. The more kind of traditional fields of machine learning specifically really are data hungry. Generative AI has unlocked this potential to not necessarily have that need for data because we can rely on generative AI. So yeah, it's, it's certainly unlocked different gateways, especially I think for the technical people who really understand the strengths, but also the weaknesses of generative AI and what it can do.

I What you shouldn't rely on it to do.

[00:07:54] Heather: So what can it do and what shouldn't you rely on it to do?

[00:07:58] Mara: Yeah, so, um, the way I think of it as a tool that's great to brainstorm, that's good to get a train of thought going when you're just feeling a little bit stuck. It's maybe like a friend that you're bouncing ideas off of, but the friend doesn't necessarily really know all the facts.

I think you actually have to be better at knowing the facts to use the tools properly. And I'm speaking specifically about the large language models and chat GPT here. It really is up to the user to be able to properly fact check and not just take the results as truth, but being able to detect what is probably made up, where it's hallucinating.

But I do think that it's still really useful when you understand that kind of limitation in order to help generate thoughts and the creative side of things without just copy and pasting. Like there is still that inhuman touch that's definitely needed.

[00:09:04] Heather: So how, how mature are you finding that companies are with using generative AI?

[00:09:09] Mara: I think it depends. Again, it's like the companies that are building the tools internally and customizing it for themselves and like really understanding the application that it can be used for. But I do think there are still companies who don't necessarily know the limitations of it or even think that that's all that AI can do for their business.

Like there's so many other applications. And at AAMI, I've been exposed to, you know, many different industries and projects. And I think some companies, just with seeing the release of these language models, think that that is AI, and that's all there is to it. And so, you know, we work hard on helping companies see the other opportunities and other technologies that you can apply to business problems.

But again, like we've talked about, it's a, it's a really great gateway to get people thinking about the technology.

[00:10:05] Heather: What are the, some of the more advanced applications that you wish more people knew about?

[00:10:09] Mara: Ooh, that's a good one. So like I said, we, we work with companies from across the map, all different industries, and I'm really most interested in projects where you can see that kind of direct benefit to society.

So our mission at Amy is AI for good and for all, and so when it's clear that the projects that we're working on and the tools that we're building for our, our clients are being used That's what gets me really excited, you know, one example would be we worked with a research group out of the U of A who had 40 years of fire weather data, and there's been lots of fires in Canada this summer and for several summers past.

And working to take that data and better predict where these fires might pop up in the future before they actually do so that they can send the necessary resources and allocate the right resources to stop these wildfires. Projects like that really interest me and it's certainly an advanced application of AI.

They're, they're hard problems to work on. Also obviously extremely rewarding.

[00:11:19] Heather: Yeah, that's really interesting. Have you seen any gaps between where companies are with where they want to be with implementing AI and then where they are currently?

[00:11:29] Mara: Yeah, you bet. At the very early stages, when companies are just wanting to explore this area for the first time, there can be misconceptions about like what AI does well and what it doesn't do well.

And what decisions should ultimately still be left for humans. There's lots of problems that are really well suited to machine learning. Again, specifically, my team is focused on ML applications. But some problems aren't necessarily the best for that technology. And that's, that's fine. There's some rule based methods that work just fine.

So, I think there can be a misconception that machine learning is just this magical dust that you sprinkle on a problem to get a better result and be more competitive, and I think that can sometimes be a problem. And then sort of related to that, we always want to start with a business problem. And sometimes companies want to start with the solution, which is machine learning, but there actually isn't like a clear business problem to work on.

So, machine learning, it's, it has many different sub components, depends on the type of data you're working with, the type of problem, and the industry, and the domain. So there's all these sub fields, and if you don't really understand the business problem well enough, it's really hard to figure out what problem we're solving, what does the machine learning solution look like.

So again, we work with companies who are at that earlier stage and looking for support. to just like help that brainstorming process. We actually start with the business problems. Let's talk about problems you're facing and then let's like kind of dwindle down what machine learning applications there are from all of that.

I think further along the journey, obviously there's sometimes challenges with data, data readiness, access and just overall differing perceptions of data readiness. So, what does that actually mean? And I think that varies depending on who you're talking to. We always tell our clients, garbage in is garbage out.

So we, we want to make sure we're working with a really good, clean data set before we start building models based off of what's in the data. And that's a big component of machine learning projects in general, is that data readiness piece. If you're looking to predict an event, you probably need examples of that event occurring.

So making sure there's a sufficient quantity of what we're trying to predict because if you don't have historical examples of the event happening, it's really hard to predict it happening in the future. Lots of challenges come to mind, even from like a bigger perspective, a lack of commitment from all stakeholders.

Like if you are wanting to become an AI company or even just use AI as a tool in your company, commitment is key. These projects are very, very collaborative, sort of by nature because it's... Machine learning expertise, its data expertise, its domain expertise, and it really does take everyone collaborating together to get to a good product in the end.

So it does often require a multidisciplinary team and. There needs to be that acknowledgement, I think, from across the organization that these resources will be required for this type of work.

[00:14:44] Heather: Oh, I'm curious if we can go back a bit. Yeah. And you were saying that people are often coming to you with the solution to the technology, like, we want to use AI.

What's driving them to push? That way, instead of thinking of that business problem first, like why are they thinking of the technology before the solution?

[00:15:02] Mara: I think it's sort of the hype of it all. It's what we're seeing in the media. It's all about AI and machine learning and it isn't necessarily about the solving of problems.

You see your competitor advertising their AI tool, and so that immediately makes you want to use AI. And again, I think there is this perception that it's just this fairy dust that you can sprinkle on anything to be more competitive and make more money, and people probably don't spend enough time reading about the actual applications and implications of machine learning and AI.

And it's easier to just think of it as a solution to a problem instead of actually brainstorming what that problem is, because that takes some time and effort, for sure.

[00:15:48] Heather: And then, you know, this collaboration and lack of buy in from all stakeholders, what are some things, so say you're a technologist and you want to bring in more machine learning and AI into your company, who should they get on board?

[00:16:02] Mara: If you are just looking to get started. Our recommendation is to start with the low hanging fruit, like what's the quick win that doesn't require a ton of resources that will allow you to demonstrate, show off to other people in the organization and hopefully get their buy in. So, probably not to start with a really complex problem.

But rather something that maybe you can solve through an off the shelf tool just to demonstrate the potential impact without that huge upfront investment.

[00:16:36] Heather: And then the multidisciplinary aspect of it, what have you found are like the best ways to pull in the right stakeholders on top of, cause it's not just people who know how to use the technology, it's people who understand the use case or how do you educate them around AI?

[00:16:52] Mara: So, Amy does have some educational services. We do have a whole team really dedicated to putting out training material. And one of the programs that we offer is the ML Foundations course. That's really meant for the general public. It's non technical. And we, we highly recommend that the companies that we work with go through that ML Foundations training up front.

So that they have a clear understanding of the technology and we go over the use cases, how you apply it, what machine learning problems actually look like. So I do think that upfront education piece is really important and that can do a pretty good job at informing various stakeholders from across the org on the benefits of the technology but also the limitations because you don't want someone.

That has unrealistic expectations. We want to make sure we're all on the same page and are really grounded in the types of expectations that we can promise at the end of projects. That introductory knowledge of machine learning is important. And we work with even executives on the delivery of that content to get them on board.

And then from there, I, I really don't think it's too hard for people to see the benefits and the value of it, but it sometimes just takes some high level understanding to get everyone on the same page.

[00:18:15] Heather: Around generative AI and it being the gateway. Have you found that there's an education piece there as well, getting people to understand the difference between generative AI and maybe machine intelligence or machine learning?

[00:18:28] Mara: Absolutely, yeah. Understanding that generative AI is a sub component of AI. But it's not it. And again, like what we were talking about earlier, the risks of those types of tools and the limitations. And well, especially with generative AI, that like ethics piece I think is really important too. And across the board, when you're deploying AI and machine learning, ethics is really important.

But generative AI has opened up this whole other ethical discussion. Partly because of the data that it's been trained on and the inherent biases that are out there. So yes, definitely that education piece on generative AI is really important to, again, apply it appropriately.

[00:19:10] Heather: How many people in the organization should be educated on generative AI?

I mean, is it just people who are implementing it? Like, should it be all employees that could potentially be using it? Like, what would the benefits be?

[00:19:24] Mara: I think definitely all users should be educated, but even outside of that, if there's someone in the organization who's using generative AI for anything, the whole organization needs to know that that's happening and where it's happening and what is the strategy around generative AI usage within the org.

Just similar to other tools that you might use at your organization, there's likely some sort of. Sign off procedure that has to happen, and I think generative AI is, is certainly no different.

[00:19:58] Heather: I've heard of instances of what I, I think it's being called shadow AI. So when people are using gen AI and they don't have buy in like from their company, what are some of the big red flags there or the downsides of something like that happening in an organization?

[00:20:14] Mara: If you don't know where, where your prompts are going, like the information you're putting into the the generative AI tools, where is that information going? Is it going to another company's servers? And what are they doing with that information, right? Like, what are they collecting about you? Obviously, if you're working with other customers or clients and you're using their information and inputting it there, like there's huge risks associated with that.

So, it can certainly open up a whole can of very messy worms and I think that's one of the reasons why it's so important that everyone's on the same page and that if you're using it in the shadows, I wouldn't recommend that because of all of these associated risks that you probably don't think about.

You know, it really does just feel like a fairly innocent platform, but there are certainly implications.

[00:21:10] Heather: Have you found that there's a good balance between kind of the pressure to adopt this technology and then also the slow and steady approach to implementing it? Like what kinds of things are you recommending to organizations that you're talking to?

[00:21:24] Mara: Strategy. We want companies to adopt this technology and to use it because like I've said, there's so many use cases and so many problems that it can solve. But I do think that there, there needs to be a strategy. overall using AI, but certainly generative AI. And that includes things like the data pipeline.

Where is our data being stored from the very beginning? How are we accessing it? Who is labeling it? Compute, depending on the types of models you might be playing with, where are your models living? Are they on the cloud? Are they on prem? Strategy includes all of those things, and then who's supporting the models after they're deployed?

Because it's not really just like, deploy it and then... Walk away. That's resources that you need to consider if you do want to take the technology internally within your company, who's responsible for supporting and maintaining the model once it's deployed, because your data might drift, things might change, and your model might become outdated fairly quickly.

And then again, that like ethical AI strategy, really taking a look at the outcomes of the decisions that this model is making, what that could impact, who that could impact, but even like on the data side of things, how was that data collected, making sure that you've removed or you're aware of all of the inherent biases in the data.

There's a lot to it. So yeah, we're, we're super excited to work with companies to like adopt this technology, but there's, there's more to it than just building a cool predictive model and walking away.

[00:23:07] Heather: Out of curiosity, who like, who owns that strategy within the organization? I'm sure it's different, but have you seen any trends there?

[00:23:16] Mara: I think it's executive ownership and at least they put it in place and then it's cascaded down to the entire org to make sure that they're adhering to it.

[00:23:28] Heather: Are you finding any hurdles to get executive buy in around it, or are they really on board?

[00:23:34] Mara: Sometimes, we more so act as a sounding board for companies who are looking to implement that, but we really do put the onus on them, so we explain why it's so important.

But then it is really up to the companies because every strategy is going to be different. We also have an ethics workshop showing them why it is so important and the potential implications of not having a strategy in place is usually enough to Make them take it seriously.

[00:24:08] Heather: I think a lot of companies are feeling that pressure to say that they're using AI in some way, they're getting pressure from the board or their stakeholders or internal, external pressures.

Do you find that sometimes though they'll say that they're using AI and the reality is different? Like what are some of the hallmarks of a company that has adopted AI successfully or has it a complete pipeline there?

[00:24:31] Mara: We think of it on like the AI adoption spectrum and that is from companies just exploring it, just learning about it from the very kind of get go all the way to advancing and operationalizing the tools that they've used.

In between those two extremes, most companies do fall, right? So, if we're talking about companies that are saying that they've implemented AI, to, to me that means, hate to repeat myself, but they have an AI strategy in place, and there's at least some initial results achieved. I think the next steps are really about Deployment and operationalizing and realizing the value, but at the companies that say they're using it generally to me that means they've thought about that strategy piece, they understand the problems that they can solve with applying the technology.

And there are those initial results happening. I think oftentimes what we see is companies that there's a system in place already, like they're already solving a problem, and they're getting some level of accuracy. So one of the first things that we do is we benchmark that against what our models can achieve.

Usually the stage of company that we're working with is they've already maybe even developed a simple ML model and they're looking for us to come in and improve it, or they're just looking to beat their existing system. That's how I sort of think about what I would consider a company that is actually implementing AI versus those who are just kind of In the exploratory phase, but again, don't understand that business case for it.

[00:26:15] Heather: And then just as you've seen trend, like what about when it comes to employees, like what kinds of skills do you think people are going to need going forward when it comes to AI?

[00:26:26] Mara: Great question. My team is mostly composed of machine learning scientists who are building machine learning models and are very specialized in the field, but I really think that it does take a village to build a company around AI, and certainly at AIME there's lots of non technical skill sets that are required.

That education piece, everyone should know the limitations and the powers of the technology. Obviously, the highly technical people need to understand the different tools that are available to them and the ethical implications. I think that should be an org wide awareness. Skills like where to apply machine learning, where not to apply machine learning, how to use the tools, including generative AI.

I think those skills will be. In high demand going forward, the people that are most well informed in this space and like not even from like the technical side of things, but there's obviously a ton of like non technical implications as well. And I think it's those people that will become and already are in higher demand.

[00:27:37] Heather: And then I'm just curious, what about the future? I mean, in the past year, the landscape of AI has changed so dramatically. I can only imagine what's going to happen over the next year. Do you have any bold predictions around AI or machine learning that you have an inkling around?

[00:27:55] Mara: Ah, I don't have a crystal ball.

I wish I did. We just see more and more companies coming to the realization that this is something that they want to build or use. You know, you don't always have to build a technology you can buy or partner as well. I'm hoping that the awareness just continues, but also that people start seeing applications outside of just generative AI, because there are so many other really cool opportunities within this space, and I guess my hope would be that there's a little bit more awareness about what else it can do so that we can companies adopt the technology and, and solve these really challenging business problems.

[00:28:38] Heather: I really enjoyed your story about the, the fire prediction. Do you want to share any other cool use cases that you've run into?

[00:28:45] Mara: Yeah, so I have to be a little bit careful because our products are covered under NDA, but we're also working with the National Research Council of Canada, NRC, with their protein group and there we're trying to predict protein abundance in different plant species.

So it's a really cool, heavy biology type of project working with. plant genetics to be able to better predict protein abundance. And obviously that has lots of real world implications in terms of food supply, making sure we can continue to feed our growing population and produce really nourishing plants for them to eat.

So that one is very much more in kind of the research side of things right now. But the implications for that are huge because we're working to make sure that the crops that we're growing can nourish the growing population, which to me is just really cool. And that's fascinating.

[00:29:47] Heather: Well, I think that's all I have for you, Mara.

I really appreciate the conversation. Aimee sounds like a really fascinating organization. You're doing really compelling and futuristic work. Is there anything you want to tell our viewers before we wrap this up?

[00:30:04] Mara: Yeah, maybe if anyone's interested in learning what Aimee does, you can always... Go on our website, amii. ca, lots of information there. We're always happy to chat with folks who are looking to use this technology, no matter where you are on the adoption spectrum, we're, we're always happy to chat. Wonderful.

[00:30:23] Heather: Well, thank you so much for speaking with me today.

[00:30:25] Mara: Thanks Heather.

[00:30:28] Heather: Thanks for listening. To check out resources mentioned in this episode and an article with Alteryx's generative AI pulse survey findings, head over to our show notes on community.

alteryx. com slash podcast. See you next time.

Ep 146 Driving AI Adoption [00:00:00] Megan: Hey Alter Everything listeners! We know that turning raw data into insights can be time consuming and challenging, but we've got good news for you. Alteryx can transform your analytics. Alteryx drives positive business outcomes by enabling fast, data driven decisions. And you can try out our products today. Start your 30 day free trial of Alteryx desktop or the Analytics cloud platform at alteryx.com/alter everything. [00:00:33] Heather: Welcome to Alter Everything, a podcast about data science and analytics culture. I'm Heather Ferguson, senior editorial manager at Alteryx. We've been conducting surveys around generative AI adoption with data leaders and board members. So I'm excited today to chat with Mara Cairo. about her expertise in AI adoption and how companies can get past hurdles to realize the value this technology can bring. Let's get started. Mara, thank you so much for joining us today. I'm really excited to sit down with you, have this conversation about AI, hear about your background. So let's start. Why don't you tell us a little bit about yourself and what brought you into this space? [00:01:16] Mara: Sure, sounds good. Thanks for having me, Heather. Uh, I'm Mara Cairo, product owner of the Advanced Technology team at AAMI or the Alberta Machine Intelligence Institute. So I and my team work with companies who are a little bit further along on their AI adoption journey and really ready to get some hands on support to start building out machine learning models to solve whatever business problem they've come to us for support with. Amy has a history of about over 20 years at this point, starting really from the academic side of things. Our Amy Fellows at the University of Alberta have been doing fundamental research in this space for a very long time, and the Alberta Machine Intelligence Institute was really stood up to make sure that that fundamental research is being translated into industry adoption. So, my work is heavily on the industry adoption side of things, making sure that we're translating this really powerful technology into industry applications that will hopefully make this world a better place. [00:02:22] Heather: So, it seems like a really interesting structure for an organization, especially to come out of a university, but then be focused on the industry adoption. What was the driver behind that? [00:02:33] Mara: We didn't want this technology and these breakthroughs to stay in the lab and my entire career has really been focused on moving industry along and helping to diversify our economy, make sure that companies have the right tools in place to solve their really tough problems. And I think with machine learning and AI, there are so many kind of real world applications that can take advantage of this technology that there was that recognition that it's really imperative that we put some emphasis on taking all of that knowledge and expertise out of the lab and applying it to real world problems because it's such a useful tool. [00:03:19] Heather: And you have a unique background yourself as well. You're coming to this from more of an engineering background versus like a data scientist background. How did that happen? [00:03:28] Mara: So yeah, I spent about 10 years in more of the hardware space. So my background is in electrical engineering. And for about 10 years, I worked in a clean room building really, really small things. And there was obviously a very heavy focus on building the hardware. So lots of sensors and optics and It was interesting to see over that period, there was more and more interest put on the data that these sensors were actually pulling out from their environment. So I was introduced to the space just organically because the companies we were working with were more and more interested with putting that data to use. So that was my first introduction into data science. And then I also was making a career shift. From a very hands on technical person to more of a project manager. So that's how I got my foot in the door at Aimee, was through an opening in a project management role at Aimee. Which I learned was significantly different than the kind of traditional PM in a hardware space. It's been a learning curve, but I've been at Amy for about three and a half years now and still learning every day, but yeah, that's how I made that transition. [00:04:46] Heather: So Amy's in such an interesting space because you're working cross industry, cross business. I think it's a really interesting perspective to look at what's happening in the industry. So I'm curious, what is the current state of AI in industry? Are you seeing any trends or are you seeing any, anything shifting or changing? [00:05:06] Mara: Yeah, so definitely over the last year or so, there's been a ton more interest in generative AI. This time last year, most of the companies that we were working with and talking to were using the more traditional aspects of machine learning, computer vision, time series, bioinformatics. But we have seen a shift in lots of the companies that we talk to now that are looking for support. are really looking to use generative AI. So it's awesome because I think it has put it on everyone's radar. It's kind of a tool at everyone's hands now and it's really easy to see the benefits of the technology. So I think more and more people are understanding like, oh this is actually something that I can see being applied to my business problem. Whereas before You know, there was some hesitancy, like, what can this do for me, right? So, it's interesting that generative AI has got basically everyone interested in the fields. It's also unlocked more off the shelf tools for people and companies to tinker with, just to, again, play with it, see what it can do, and then for them to be able to realize a direct value. So, that's the trend that I've seen. Over the last year, and I'm, I'm sure everyone could say the same thing, it is a bit of a hype train. So right now there's companies using AI and then there's AI companies and we work with AI companies who really want the technology to be a core business driver. So with that comes a lot of kind of the customization of models and going above and beyond the off the shelf tools that are the lower hanging fruit. But there's different kind of company profiles, and I think the more the merrier, the more people using and interested in this technology, the better, and it is a way to, to move your business forward. [00:07:06] Heather: Have you found that generative AI has been a gateway to more advanced applications of AI? [00:07:12] Mara: For sure. Even the capabilities that it's unlocked for technical people. So it's, it's made it easier to say, generate data if data doesn't exist. The more kind of traditional fields of machine learning specifically really are data hungry. Generative AI has unlocked this potential to not necessarily have that need for data because we can rely on generative AI. So yeah, it's, it's certainly unlocked different gateways, especially I think for the technical people who really understand the strengths, but also the weaknesses of generative AI and what it can do. I What you shouldn't rely on it to do. [00:07:54] Heather: So what can it do and what shouldn't you rely on it to do? [00:07:58] Mara: Yeah, so, um, the way I think of it as a tool that's great to brainstorm, that's good to get a train of thought going when you're just feeling a little bit stuck. It's maybe like a friend that you're bouncing ideas off of, but the friend doesn't necessarily really know all the facts. I think you actually have to be better at knowing the facts to use the tools properly. And I'm speaking specifically about the large language models and chat GPT here. It really is up to the user to be able to properly fact check and not just take the results as truth, but being able to detect what is probably made up, where it's hallucinating. But I do think that it's still really useful when you understand that kind of limitation in order to help generate thoughts and the creative side of things without just copy and pasting. Like there is still that inhuman touch that's definitely needed. [00:09:04] Heather: So how, how mature are you finding that companies are with using generative AI? [00:09:09] Mara: I think it depends. Again, it's like the companies that are building the tools internally and customizing it for themselves and like really understanding the application that it can be used for. But I do think there are still companies who don't necessarily know the limitations of it or even think that that's all that AI can do for their business. Like there's so many other applications. And at AAMI, I've been exposed to, you know, many different industries and projects. And I think some companies, just with seeing the release of these language models, think that that is AI, and that's all there is to it. And so, you know, we work hard on helping companies see the other opportunities and other technologies that you can apply to business problems. But again, like we've talked about, it's a, it's a really great gateway to get people thinking about the technology. [00:10:05] Heather: What are the, some of the more advanced applications that you wish more people knew about? [00:10:09] Mara: Ooh, that's a good one. So like I said, we, we work with companies from across the map, all different industries, and I'm really most interested in projects where you can see that kind of direct benefit to society. So our mission at Amy is AI for good and for all, and so when it's clear that the projects that we're working on and the tools that we're building for our, our clients are being used That's what gets me really excited, you know, one example would be we worked with a research group out of the U of A who had 40 years of fire weather data, and there's been lots of fires in Canada this summer and for several summers past. And working to take that data and better predict where these fires might pop up in the future before they actually do so that they can send the necessary resources and allocate the right resources to stop these wildfires. Projects like that really interest me and it's certainly an advanced application of AI. They're, they're hard problems to work on. Also obviously extremely rewarding. [00:11:19] Heather: Yeah, that's really interesting. Have you seen any gaps between where companies are with where they want to be with implementing AI and then where they are currently? [00:11:29] Mara: Yeah, you bet. At the very early stages, when companies are just wanting to explore this area for the first time, there can be misconceptions about like what AI does well and what it doesn't do well. And what decisions should ultimately still be left for humans. There's lots of problems that are really well suited to machine learning. Again, specifically, my team is focused on ML applications. But some problems aren't necessarily the best for that technology. And that's, that's fine. There's some rule based methods that work just fine. So, I think there can be a misconception that machine learning is just this magical dust that you sprinkle on a problem to get a better result and be more competitive, and I think that can sometimes be a problem. And then sort of related to that, we always want to start with a business problem. And sometimes companies want to start with the solution, which is machine learning, but there actually isn't like a clear business problem to work on. So, machine learning, it's, it has many different sub components, depends on the type of data you're working with, the type of problem, and the industry, and the domain. So there's all these sub fields, and if you don't really understand the business problem well enough, it's really hard to figure out what problem we're solving, what does the machine learning solution look like. So again, we work with companies who are at that earlier stage and looking for support. to just like help that brainstorming process. We actually start with the business problems. Let's talk about problems you're facing and then let's like kind of dwindle down what machine learning applications there are from all of that. I think further along the journey, obviously there's sometimes challenges with data, data readiness, access and just overall differing perceptions of data readiness. So, what does that actually mean? And I think that varies depending on who you're talking to. We always tell our clients, garbage in is garbage out. So we, we want to make sure we're working with a really good, clean data set before we start building models based off of what's in the data. And that's a big component of machine learning projects in general, is that data readiness piece. If you're looking to predict an event, you probably need examples of that event occurring. So making sure there's a sufficient quantity of what we're trying to predict because if you don't have historical examples of the event happening, it's really hard to predict it happening in the future. Lots of challenges come to mind, even from like a bigger perspective, a lack of commitment from all stakeholders. Like if you are wanting to become an AI company or even just use AI as a tool in your company, commitment is key. These projects are very, very collaborative, sort of by nature because it's... Machine learning expertise, its data expertise, its domain expertise, and it really does take everyone collaborating together to get to a good product in the end. So it does often require a multidisciplinary team and. There needs to be that acknowledgement, I think, from across the organization that these resources will be required for this type of work. [00:14:44] Heather: Oh, I'm curious if we can go back a bit. Yeah. And you were saying that people are often coming to you with the solution to the technology, like, we want to use AI. What's driving them to push? That way, instead of thinking of that business problem first, like why are they thinking of the technology before the solution? [00:15:02] Mara: I think it's sort of the hype of it all. It's what we're seeing in the media. It's all about AI and machine learning and it isn't necessarily about the solving of problems. You see your competitor advertising their AI tool, and so that immediately makes you want to use AI. And again, I think there is this perception that it's just this fairy dust that you can sprinkle on anything to be more competitive and make more money, and people probably don't spend enough time reading about the actual applications and implications of machine learning and AI. And it's easier to just think of it as a solution to a problem instead of actually brainstorming what that problem is, because that takes some time and effort, for sure. [00:15:48] Heather: And then, you know, this collaboration and lack of buy in from all stakeholders, what are some things, so say you're a technologist and you want to bring in more machine learning and AI into your company, who should they get on board? [00:16:02] Mara: If you are just looking to get started. Our recommendation is to start with the low hanging fruit, like what's the quick win that doesn't require a ton of resources that will allow you to demonstrate, show off to other people in the organization and hopefully get their buy in. So, probably not to start with a really complex problem. But rather something that maybe you can solve through an off the shelf tool just to demonstrate the potential impact without that huge upfront investment. [00:16:36] Heather: And then the multidisciplinary aspect of it, what have you found are like the best ways to pull in the right stakeholders on top of, cause it's not just people who know how to use the technology, it's people who understand the use case or how do you educate them around AI? [00:16:52] Mara: So, Amy does have some educational services. We do have a whole team really dedicated to putting out training material. And one of the programs that we offer is the ML Foundations course. That's really meant for the general public. It's non technical. And we, we highly recommend that the companies that we work with go through that ML Foundations training up front. So that they have a clear understanding of the technology and we go over the use cases, how you apply it, what machine learning problems actually look like. So I do think that upfront education piece is really important and that can do a pretty good job at informing various stakeholders from across the org on the benefits of the technology but also the limitations because you don't want someone. That has unrealistic expectations. We want to make sure we're all on the same page and are really grounded in the types of expectations that we can promise at the end of projects. That introductory knowledge of machine learning is important. And we work with even executives on the delivery of that content to get them on board. And then from there, I, I really don't think it's too hard for people to see the benefits and the value of it, but it sometimes just takes some high level understanding to get everyone on the same page. [00:18:15] Heather: Around generative AI and it being the gateway. Have you found that there's an education piece there as well, getting people to understand the difference between generative AI and maybe machine intelligence or machine learning? [00:18:28] Mara: Absolutely, yeah. Understanding that generative AI is a sub component of AI. But it's not it. And again, like what we were talking about earlier, the risks of those types of tools and the limitations. And well, especially with generative AI, that like ethics piece I think is really important too. And across the board, when you're deploying AI and machine learning, ethics is really important. But generative AI has opened up this whole other ethical discussion. Partly because of the data that it's been trained on and the inherent biases that are out there. So yes, definitely that education piece on generative AI is really important to, again, apply it appropriately. [00:19:10] Heather: How many people in the organization should be educated on generative AI? I mean, is it just people who are implementing it? Like, should it be all employees that could potentially be using it? Like, what would the benefits be? [00:19:24] Mara: I think definitely all users should be educated, but even outside of that, if there's someone in the organization who's using generative AI for anything, the whole organization needs to know that that's happening and where it's happening and what is the strategy around generative AI usage within the org. Just similar to other tools that you might use at your organization, there's likely some sort of. Sign off procedure that has to happen, and I think generative AI is, is certainly no different. [00:19:58] Heather: I've heard of instances of what I, I think it's being called shadow AI. So when people are using gen AI and they don't have buy in like from their company, what are some of the big red flags there or the downsides of something like that happening in an organization? [00:20:14] Mara: If you don't know where, where your prompts are going, like the information you're putting into the the generative AI tools, where is that information going? Is it going to another company's servers? And what are they doing with that information, right? Like, what are they collecting about you? Obviously, if you're working with other customers or clients and you're using their information and inputting it there, like there's huge risks associated with that. So, it can certainly open up a whole can of very messy worms and I think that's one of the reasons why it's so important that everyone's on the same page and that if you're using it in the shadows, I wouldn't recommend that because of all of these associated risks that you probably don't think about. You know, it really does just feel like a fairly innocent platform, but there are certainly implications. [00:21:10] Heather: Have you found that there's a good balance between kind of the pressure to adopt this technology and then also the slow and steady approach to implementing it? Like what kinds of things are you recommending to organizations that you're talking to? [00:21:24] Mara: Strategy. We want companies to adopt this technology and to use it because like I've said, there's so many use cases and so many problems that it can solve. But I do think that there, there needs to be a strategy. overall using AI, but certainly generative AI. And that includes things like the data pipeline. Where is our data being stored from the very beginning? How are we accessing it? Who is labeling it? Compute, depending on the types of models you might be playing with, where are your models living? Are they on the cloud? Are they on prem? Strategy includes all of those things, and then who's supporting the models after they're deployed? Because it's not really just like, deploy it and then... Walk away. That's resources that you need to consider if you do want to take the technology internally within your company, who's responsible for supporting and maintaining the model once it's deployed, because your data might drift, things might change, and your model might become outdated fairly quickly. And then again, that like ethical AI strategy, really taking a look at the outcomes of the decisions that this model is making, what that could impact, who that could impact, but even like on the data side of things, how was that data collected, making sure that you've removed or you're aware of all of the inherent biases in the data. There's a lot to it. So yeah, we're, we're super excited to work with companies to like adopt this technology, but there's, there's more to it than just building a cool predictive model and walking away. [00:23:07] Heather: Out of curiosity, who like, who owns that strategy within the organization? I'm sure it's different, but have you seen any trends there? [00:23:16] Mara: I think it's executive ownership and at least they put it in place and then it's cascaded down to the entire org to make sure that they're adhering to it. [00:23:28] Heather: Are you finding any hurdles to get executive buy in around it, or are they really on board? [00:23:34] Mara: Sometimes, we more so act as a sounding board for companies who are looking to implement that, but we really do put the onus on them, so we explain why it's so important. But then it is really up to the companies because every strategy is going to be different. We also have an ethics workshop showing them why it is so important and the potential implications of not having a strategy in place is usually enough to Make them take it seriously. [00:24:08] Heather: I think a lot of companies are feeling that pressure to say that they're using AI in some way, they're getting pressure from the board or their stakeholders or internal, external pressures. Do you find that sometimes though they'll say that they're using AI and the reality is different? Like what are some of the hallmarks of a company that has adopted AI successfully or has it a complete pipeline there? [00:24:31] Mara: We think of it on like the AI adoption spectrum and that is from companies just exploring it, just learning about it from the very kind of get go all the way to advancing and operationalizing the tools that they've used. In between those two extremes, most companies do fall, right? So, if we're talking about companies that are saying that they've implemented AI, to, to me that means, hate to repeat myself, but they have an AI strategy in place, and there's at least some initial results achieved. I think the next steps are really about Deployment and operationalizing and realizing the value, but at the companies that say they're using it generally to me that means they've thought about that strategy piece, they understand the problems that they can solve with applying the technology. And there are those initial results happening. I think oftentimes what we see is companies that there's a system in place already, like they're already solving a problem, and they're getting some level of accuracy. So one of the first things that we do is we benchmark that against what our models can achieve. Usually the stage of company that we're working with is they've already maybe even developed a simple ML model and they're looking for us to come in and improve it, or they're just looking to beat their existing system. That's how I sort of think about what I would consider a company that is actually implementing AI versus those who are just kind of In the exploratory phase, but again, don't understand that business case for it. [00:26:15] Heather: And then just as you've seen trend, like what about when it comes to employees, like what kinds of skills do you think people are going to need going forward when it comes to AI? [00:26:26] Mara: Great question. My team is mostly composed of machine learning scientists who are building machine learning models and are very specialized in the field, but I really think that it does take a village to build a company around AI, and certainly at AIME there's lots of non technical skill sets that are required. That education piece, everyone should know the limitations and the powers of the technology. Obviously, the highly technical people need to understand the different tools that are available to them and the ethical implications. I think that should be an org wide awareness. Skills like where to apply machine learning, where not to apply machine learning, how to use the tools, including generative AI. I think those skills will be. In high demand going forward, the people that are most well informed in this space and like not even from like the technical side of things, but there's obviously a ton of like non technical implications as well. And I think it's those people that will become and already are in higher demand. [00:27:37] Heather: And then I'm just curious, what about the future? I mean, in the past year, the landscape of AI has changed so dramatically. I can only imagine what's going to happen over the next year. Do you have any bold predictions around AI or machine learning that you have an inkling around? [00:27:55] Mara: Ah, I don't have a crystal ball. I wish I did. We just see more and more companies coming to the realization that this is something that they want to build or use. You know, you don't always have to build a technology you can buy or partner as well. I'm hoping that the awareness just continues, but also that people start seeing applications outside of just generative AI, because there are so many other really cool opportunities within this space, and I guess my hope would be that there's a little bit more awareness about what else it can do so that we can companies adopt the technology and, and solve these really challenging business problems. [00:28:38] Heather: I really enjoyed your story about the, the fire prediction. Do you want to share any other cool use cases that you've run into? [00:28:45] Mara: Yeah, so I have to be a little bit careful because our products are covered under NDA, but we're also working with the National Research Council of Canada, NRC, with their protein group and there we're trying to predict protein abundance in different plant species. So it's a really cool, heavy biology type of project working with. plant genetics to be able to better predict protein abundance. And obviously that has lots of real world implications in terms of food supply, making sure we can continue to feed our growing population and produce really nourishing plants for them to eat. So that one is very much more in kind of the research side of things right now. But the implications for that are huge because we're working to make sure that the crops that we're growing can nourish the growing population, which to me is just really cool. And that's fascinating. [00:29:47] Heather: Well, I think that's all I have for you, Mara. I really appreciate the conversation. Aimee sounds like a really fascinating organization. You're doing really compelling and futuristic work. Is there anything you want to tell our viewers before we wrap this up? [00:30:04] Mara: Yeah, maybe if anyone's interested in learning what Aimee does, you can always... Go on our website, amii. ca, lots of information there. We're always happy to chat with folks who are looking to use this technology, no matter where you are on the adoption spectrum, we're, we're always happy to chat. Wonderful. [00:30:23] Heather: Well, thank you so much for speaking with me today. [00:30:25] Mara: Thanks Heather. [00:30:28] Heather: Thanks for listening. To check out resources mentioned in this episode and an article with Alteryx's generative AI pulse survey findings, head over to our show notes on community. alteryx. com slash podcast. See you next time.

This episode was produced by Megan Dibble (@MeganDibble), Mike Cusic (@mikecusic), and Matt Rotundo (@AlteryxMatt). Special thanks to @andyuttley for the theme music track, and @mikecusic for our album artwork.

Alter Everything

Episode Guide

146: How to drive AI adoption in your organization

Panelists

Topics

This episode was produced by Megan Dibble (@MeganDibble), Mike Cusic (@mikecusic), and Matt Rotundo (@AlteryxMatt). Special thanks to @andyuttley for the theme music track, and @mikecusic for our album artwork.