{PODCAST} In-Ear Insights: Evaluating New Technologies

Based on a question from our analytics community, in this episode of In-Ear Insights, Katie and Chris tackle the challenges of new technology. When a new piece of technology makes a big splash, how do we evaluate it? How should we assess whether it’s right for us, whether it makes sense to pursue it? What are some of the things that can go wrong? Watch this episode to learn the answers, plus the specific AI technology that prompted this question, OpenAI’s GPT-3.

[podcastsponsor]

Watch the video here:

Listen to the audio here:

Download the MP3 audio here.

• Need help with your company’s data and analytics? Let us know!
• Join our free Slack group for marketers interested in analytics!

Machine-Generated Transcript

What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for listening to the episode.

Christopher Penn
This is In-Ear Insights the Trust Insights podcast. AI Academy for marketers is an online education platform designed to help marketers like you understand pilot and scale artificial intelligence. The AI Academy features deep dive certification courses of three to five hours, along with dozens of short courses 30 to 60 minutes each taught by leading AI and marketing experts. Jordan Katie Robbert, CEO of Trust Insights in me Christopher Penn chief data scientist of Trust Insights for three of our courses in the academy five use cases of AI for content marketing, intelligent attribution modeling for marketing, and detecting and mitigating bias in marketing AI. The Academy is designed for manager level and above marketers and largely caters to non technical audiences, meaning you don’t need a programming background or background in data science to understand and apply what you learn. One registration gives you unlimited access to all the courses and invitation to a members only slack instance, access to new courses every quarter. Join now and save $100 off registration when you go to Trust insights.ai slash AI Academy and use registration code pen 100 today that’s Trust insights.ai slash AI Academy and use registration code pen 100 today in this week’s In-Ear Insights, we’re going to start off with something totally different. A short poetry reading. The Rime of the Ancient Google, Google has gone to God you cannot conceive it or guess it all the sudden it has happened. Come Come whoever you are, oh, come now is pacing hasten. So you may be seeing what the world is going on what has happened to the data focus podcasts that we all love. The over the weekend, researchers at open AI have published the results of their the newest language model of music called GPT three Theirs was the original one a couple years ago. GPT two was last year, which is something that a lot of folks included, including TrustInsights.ai have used to generate credibly sounding natural language by machine and GPT three has come out and for all of the research and data published so far has is a substantial leap forward in terms of what computers can do to generate language. Here’s another example. This is the poem a future consensus forecasts by the Board of Governors of the Federal Reserve System. No one knows what will come forecast for teller rising power. That is probably not to be deficit of several lines. The Golden future has tourniquets, no one likes it. I love the I love the expression on your face, Katie. But more practically, this is a tool that given a prompt is able to generate more credible sounding data in natural language and so This follows on to a question that we were asked last week on Joseph Jaffe show by Bob Farnham saying, When something brand shiny new comes out, how do you evaluate it? How do you fit it into strategy? How do you know whether it’s it’s worth paying attention to or just another shiny object for the squirrel to chase after for the brain squirrels to go after? So when you think about a new technology, and obviously Katie, you’ve had to deal with me for more than five years now on this front. How do you differentiate between, like the technology being shiny and new and chasing after and Okay, well, where does this fit in?

Katie Robbert
Well, first, let me take a step back and say I was always terrible at deciphering poetry in high school in my English classes when that was like what we were working on. I was terrible at it. So it’s all abstract. nonsense words shoved together for me. So everything you just said. I definitely had Like the Macarena playing in my head while you were saying it, because I was like, I don’t get any of this. So that being said, when the first thing that I look for when new software or a new version or something comes out, and this might be the, you know, Product Manager in me is I’m looking for release notes, what’s new? And so if I can’t easily see what’s new in a new version, like there should be some sort of a bulleted list, like, what’s new, you know, fixed, you know, the widget bug, you know, changed from blue to green, like whatever the thing is, and also in this case, because it’s more complicated machine learning with natural language processing. I need to have an understanding at a glance of what does it do now that it didn’t do before? Why do I need this version versus the old version? So that’s one piece of it, the other piece of it and Chris, this is something that I always ask you is For the sake of the business, for our business for our clients, do they need whatever the shiny object thing is saying that it’s going to do? Give me a use case show me where we need to spend our time figuring out how this thing works, because I think the challenge with software is that there’s so much of it, and it updates so quickly, and it’s constantly changing. I think the thing that marketers and companies need to ask themselves is, what’s the problem that this thing is solving? And do I need it? Or is it just like cool and exciting and I want to play with it. If the ladder, do it on your own time, and former? Let’s figure out what problems that we currently have, that this thing will solve and then spend your time exploring it?

Christopher Penn
Yeah, yeah, this is still from a technology perspective. It is still early, it is still in the actual model itself is not publicly available yet. It is researchers who granting access to the private beta or publishing their findings now, but it is not generally available and I’m hopeful that in the next year or so, it will become available to the general public in ways that we can use it. To your point about the release notes it’s interesting because this when a new model is released, that does something substantially different it’s kind of like the original was a bicycle version two was a car and now there’s an airplane sitting in the driveway. And so it’s not necessarily a release notes so much as there’s a brand new thing that does something different that essentially still choose the overall goal of transportation but instead of it being you know, a car you drive it’s a plane you fly. And so there’s obviously a bit more of a learning curve to using it. Where I think it’s really important and like you said, is is understand what is the

Unknown Speaker
thing do?

Christopher Penn
Like it’s like it’s a puts a new appliance in your kitchen. Can you broadly tell, like, does it fry things? Does it bake things? Does it blend things like what what exactly is the thing If it’s not clear, then perhaps the the UI needs a bit of work with this model, the thing that fascinated me most of all that I think is has potential incredible both benefits and hazards for our industry is something called neural style transfer. So Katie, if I took your blog posts or the our conversations and embed them into the model as a, as priming as training data, and then I took Harry Potter, and I said, I want you to rewrite Harry Potter in the way that Katie robear speaks and writes, it will rewrite Harry Potter in as if you had written it as opposed to the way that the original author had written. When you think about that, that drastically changes how we think about content. Because what is content at that point, content is the idea that that is created in a certain way, but if you were to use this technology, you would be indistinguishable Theoretically, from your writing, but it also be unique, it would be something that would not be a copyright violation per se, but a derivative work. And so there are massive implications for content marketers, if you were to take, for example, a paper that McKinsey published and you were to do a neural style transfer on that you could create a McKenzie paper that was factually as correct. But in your own words.

Katie Robbert
It’s it’s an interesting dilemma in the sense of, you know, we go back to the question of do you need this thing? So, for organizations that are churning out large volumes of content, it sounds like perhaps it might be a worthwhile investment. You know, when we speak about and talk about in teach about artificial intelligence, one of the segments of the talks are is always build or buy. Well, how do I know if this is the right investor? For me, and so, you know, you’re talking about hardcore machine learning natural language processing. Let’s say we’re talking about something that’s a little bit more straightforward, like a social listening platform, for example, you know, choose any one of the dozens that are out there. If you’re evaluating whether or not you need this thing, you need to first know, do I have someone on my team who can use the thing? Do I just want to pay a third party company to do the thing for me? And do we have the money? Either way, and I think that those are some of the questions, you know, because what you’re describing with the natural language processing and having it right in my style, so that I never have to sit at my keyboard again, looking at a blank page sounds really enticing. But do we have someone on our team who can first figure out the model into So all of the, you know, learning information content, and then test it and then maintain it. And then you know, do everything with it. So do we have that resource? Do we have the time to do that? To do the r&d? And do we have the money? Because perhaps as an open source model, but you still have to pay someone to do the thing?

Christopher Penn
Yeah, I would say of, you know, people processing platform, I think your biggest limitation for something like this for any new technology really is the people and their skills because that’s, that is that is either the limiter or the enabler that you have to to leap forward on something. If you don’t have the skills in house. It doesn’t really matter what the technology is, you cannot use it, it’s like, again, if you cannot fly a plane, putting a plane in your driveway looks cool. And you can be proud that you own a plane now, but you still can’t go anywhere with it because you can’t fly.

Katie Robbert
I will say though, if you’re keeping your plane in your driveway, you clearly You’re doing it wrong, you probably shouldn’t own a plane in the first place.

Christopher Penn
It depends if you live in the middle and the Alaskan Bush, then you actually do because there’s no roads. But

Katie Robbert
let’s go with that one.

Christopher Penn
But so too. I think the other thing that we have to keep our minds open about that a hint is hinted at and Bob’s question is, does this have the potential to have a transformative effect as opposed to an iterative effect on our business? Meaning does this open up new revenue lines? Does this open up new brand new capabilities that not only do we not have, but the industry as a whole does not have something that our old buddy and friend Todd Jefferson used to say is, you know, there’s three things you could do that a newsworthy first best only you can do the first something the best is something that the only one that does something at some something and if you have a technology Do you like this and you have the people and the processes in place, you could achieve all three very, very quickly. And as we all know, from other machine learning and AI projects, there is a substantial advantage to you if you are first mover because the sooner you get going, the sooner you’re collecting data that you can use, and the more of a disadvantage competitors are at. Because they have to not play catch up on the technology has to catch up on the data collection as well.

Katie Robbert
Mm hmm. Well, and so, you know, it’s tricky, because part of my brain is like, Yeah, absolutely. That makes sense. But the other part of my brain is, what is the likelihood that a marketing team has a skilled data scientist on their staff in order to be experimenting with these new models, that to your point, Chris aren’t even fully available to the public yet are still in their testing phase, in order to be first out of the gate with something that’s easy. A unique position for a team to be in. And so it’s, you know, I sort of struggled with, you know, having the conversation about these, you know, natural language processing models, when I know that a lot of teams are still struggling with some of the basics.

Christopher Penn
When you deal with a situation like that it like you were saying earlier, it goes back to build or buy, right? If you cannot build you have to buy, right, you know, simply is, is no alternative. And that may mean by an agency like Trust Insights to say like, Hey, we need to engage you as a, as a client to offer this capability because we simply cannot do it. It’s just not in the cards, and there’s not going to be a budget to do that. So for those folks who are on the agency side, the challenge then is are you willing to make a big enough bet in your own agency, to create this capability that you can then sell and potentially offer as a substantial multiplier to your clients? Something that, again, first best only if you’re if you cannot find this capability anywhere else, and it’s valuable, and you know, it’s and it’s worth paying for. Do you? Do you take that bet?

Katie Robbert
So, you know, it goes. So this is sort of all circling around the question of how do you evaluate something that’s brand new, you know, to the market to the industry, as to whether or not it’s a shiny object or something of value to you. Because for every, you know, 10 people that call something a shiny object, one person is going to say, but I need that. And so the question is really coming down to how do you evaluate, I need that versus that shiny object. And so Chris, a lot of what you’re talking about, is really evaluating. You know, especially if it’s a brand new technology like, you know, what are your business models? What are your clients or customers need? You know, is there something like this already in the industry that would make it a worthwhile investment to you to be first out of the gate, does it solve an existing problem? Or are you looking for a problem to solve with this solution? And and can you make the investment? So those are those are not small questions for someone to ask their company and their team when evaluating something like this.

Christopher Penn
It reminds me a lot actually have a clinical trial right now phase one, does it cause harm? phase two, does it do anything? And phase three, does it do better than the existing standard of care? And I think that’s not a bad framework for looking at a piece of technology, right does is it something that’s going to actively cause harm to our systems? Does it work at all like, is it hype and vaporware? Or is there they’re there and then three, if there’s a there there, is it better than what we can currently offer? Because to your point GPT three and its cousins. In this language transform a family requires a massive investment in technology. And in brainpower and skills, you, this is not a piece of software like Microsoft Word that you download off the internet and put on your computer. This is and it’s like, literally the engine of a car, like there’s no it doesn’t come with the rest of the car, you have to build the rest of the car. It just you put you’ve got a bigger, bigger, bigger, faster engine. And so that phase three is it may turn out that for your company. It’s not better than the existing standard, because you can’t use it well. And how, when, when in the pharma world, how do they decide what drugs to invest in, like, you know, knowing that something’s gonna be a big better potential dud. You know, I’m looking at the clinical trials process.

Katie Robbert
I mean, that’s a heavy question. For a Monday morning, you know, it’s well, so a clinical trial depending on the type of trial so I worked primarily in the SP cir world, so small business innovative research. And so those trials, the phase one was six months, the phase two was 18 to 18 to 24 months. And phase three was commercialization. And, but so that right there, you’re already talking about a three year commitment to testing this thing. And that’s for, you know, software, when you’re talking about pharmaceutical drugs, you know, you’re talking about a longer, you know, typically a multi year thing, when we’re talking about software that may or may not intervene with, you know, people’s, you know, medical decisions, that kind of thing. It’s a little less of a time investment, but you should be treating it just as seriously because this is your this is your business. And so the first thing that you should really be thinking about, you know, Chris, to your point is that testing that proof of concept and so the way that I would approach that is the first thing I would do is run numbers, I would find the data both financial and otherwise. Because what I would want to know is if we take the example, this natural language processing software is going to start to write blogs for me, in my voice in my style. Okay. First and foremost, how long does it take me typically to write a blog, and what’s my hourly rate? So that’s one piece of data. How many blogs do I write per week, per month per year? That’s my second piece of data. You know, what, kind of, you know, return on investment do I get from these blog posts that I put out there? That’s my third piece of data. So I now have three hard data points to compare against my investment in this natural language processing software, where I need to know what is it going to cost me to set it up? What is it going to cost me to train it and how long is it going to take before I can replace my own physical writing, with the machine doing my writing and then how much time do I then as the human need to spend editing and publishing this content. And so there’s a lot of different variables that I would want to explore. And that would help me at least in this specific example, understand, is it worth the investment to go down this road? And so perhaps the software can do it better than I can, but it may not be cost effective. So it may not be a better option for me.

Christopher Penn
Yep. So that fits the it may not exceed the existing standard because using standard is the right balance of efficiency and effectiveness. Right. Okay, so if you’d like to chat about the model itself and things like that, and it’s of interest to you pop on over to our free slack group TrustInsights.ai dot AI slash analytics for marketers will be a topic of discussion. I’m sure that will come up multiple times over the next year. So as tools like this, or models like this or have tools built around them is going to be fascinating and it is absolutely a great water cooler for your other ordinary Regular non crazy far out there analytics questions as well so we have over 1200 folks who are enjoying the occasional conversation on marketing analytics and your follow up questions on this episode please drop a comment over the wherever it is that you’re consuming this or over on our website TrustInsights.ai dot AI and we will talk to you soon will help solving your company’s data analytics and digital marketing problems. Visit Trust insights.ai today and let us know how we can help you