How AI is Changing Marketing: Christopher Penn Keynote at Spark.me 2022

Tobardan, so them say Christopher Penn, and that is the extent of my Montenegrin. The rest of this talk. The rest of this talk will be in English because I'm American and we only speak two languages in America: English and bad English. The number one question people ask is where can you get the slides? Go to where can I get the slides.com.

We're faced with three challenges in marketing. Our stakeholders are the people who want us to do things, want us to do things better, faster, and cheaper. How are we doing at these challenges? Are we doing better? Well, in a 2019 survey of uh CEOs, Price Waterhouse asked CEOs, what are the things you care most about?

And at the far left, 94% of CEOs said data about your customers and clients' preferences and needs. And then when asked, how are good are you as the data you're getting? How are people delivering on this? A staggering 15% of CEOs said they were getting what they needed. So from a data perspective, marketing not doing such a hot job.

In the 2021 CMO survey, which you can see at CMO Survey.org, when asked, can you prove the impact of your marketing? 48% said they could actually do it. And 13% said, who knows? We have no idea what we're getting when asked, how are you using data? The n percentage of CMOs who actually use data to make decisions has dropped to about one third.

So the other two are just two-thirds are just winging it. So we're not really getting better. Are we getting faster at least? Well, the internet's getting faster. In 2017, IDC came up with this uh survey saying what they think the actual size of data produced by human beings is going to be over the next few years.

And as of this year, we're supposed to hit about 60 zettabytes. Now, a zettabyte is an unimaginably large number. A zettabyte, if you were to start watching Netflix, and you didn't stop to sleep or eat or anything, and you started 65 million years ago, uh, you'd be getting to just one of those today. And we're gonna create 60 of these this year. That is a staggering amount of data.

And everything that you saw in previous presentations, like what Dean's presentation was showing about the metaverse and things, all that's generating even more data. Everything around crypto and NFTs creates exponentially large amounts of data. In the annual uh what happens in an internet minute, just the jumps that we see from year to year are gigantic. 1.3 to 1.4 million uh scrolls per minute on Facebook, uh, going from 190 million to 197.6 million emails sent every minute on the internet. 194,000 people tweeting to a 200,000 people tweeting every single minute.

So the velocity of data is getting faster and faster. Just this year alone, we've had uh on any given day about 178,000 news stories, and that's not counting all the the circuses uh and things. That's just regular news, right? That is 7,420 stories per hour, 124 stories per minute, two stories a second. So if your company gets a great placement in the news, it's gone in a second.

Two more stories have taken its place. So the internet's gotten faster, civilization's gotten faster, marketing has not. Well, at least we're doing things cheaper, right? We've we've managed to be cost efficient. Well, not really.

In the 2022 CMO survey, spending on marketing has increased as a percentage of revenues, again, uh, to around 11%. Think about that. 11% of your marketing is being of your marketing budget is a percentage of revenues. That's an astonishing amount. And the spending growth is above 10%, projected to be above 13%.

When CEOs were asked, what do you plan on spending money on? Where do you expect to see uh growth? They're talking about digital transformation. They want to spend more, but they want to see results. So if you're not better, you're not faster, and you're not cheaper, what's gonna happen to your marketing programs?

Not good things, right? The great resignation alone shows the impact of this. Uh 10% of a company has voluntarily left in the last year. Um you've probably felt this. You've probably felt this as a customer, going places and having struggles with service.

You felt this as a manager or as an employer when you struggle to retain people because we're flooded with all this data and we don't have a way of managing it. So what's the answer? Well, one answer, not all of them, one, is the use of artificial intelligence. Artificial intelligence is about three things. It's about acceleration, accuracy, and making things easier, right?

AI can make things faster. It's all just math at the end of the day, right? It makes things a lot more accurate because anyone who's ever typed in a bunch of numbers into a spreadsheet knows how quickly those numbers go awry. And let's face it, there's a good chunk of your job, whatever your job is, that you may not love doing, right? So some very repetitive tasks, you'd like would love to see them go away.

So why hasn't the world embraced this yet? Well, because like with everything else in marketing, we haven't figured out yet how to actually demonstrate the value of it. But about 37% of CMOs say they're gonna get at least give it a shot in the next couple of years. So what is this stuff? It sounds great.

What is artificial intelligence? At its core, it's just teaching machines to do skills that human intelligence requires. If you can hear the sound of my voice and understand what I'm saying, right, you have you're using language recognition, which is something we call in computers natural language recognition. If you can see me distinct from this background and see that I'm an entity, you have vision, if you can hear, you have hearing, right? You have all these tasks that you perform with human intelligence that we're trying to teach machines to do the same thing.

All AI begins with stats, probability, and data. It is the media has really sort of, I guess, borrowed a lot of things from movies and TV about AI. At the end of the day, it really is just a bunch of math, which if you got into marketing because you wanted to avoid math, I have bad news for you. Sorry. Think of these these things as the raw ingredients, right?

Stats, probability, and just good old data. It's like a pile of grapes. Then you have algorithms. Algorithms are nothing more than a repeatable process to generate a predictable outcome. Algorithms are a repeatable process to develop generate a predictable outcome.

We know all sorts of different things that have algorithms, right? We talk about things like the Facebook algorithm and stuff, but you have it algorithms that run your life. You have algorithms that you created. You probably put the same general article of clothing on first in the morning, right? Some people put a top on first, some people put a bottom on first, a few weirdos put their socks on first.

But it's all the same thing. You have a pattern, right? That's what algorithms are. They're patterns. If you've done A B split tests or multivariate testing, you've used algorithms.

Think about algorithms as taking that data, crunching it down and processing it. That's all it is. Now, in the world of software, there's two kinds of software. There's software that you and I, as humans write, and there's software that machines write. When a machine is writing its own software based on the data it's given, that is called machine learning.

And that is when you cross into artificial intelligence, because a machine is now learning to do something, right? It's like when you're making wine, the grapes are there, they're turned into juice, and then something else is doing the actual processing, usually yeast. So it's it is no longer under our control. When you pull out a game on your phone or a word processor, someone wrote the software, a human wrote the software, and then you were creating data. Machine learning's the other way around.

We give it a bunch of data and have the machine write its own software. And there's two general flavors. There's what we're called supervised learning and unsupervised. Supervised is all about identifying a key outcome. Can you make a determination based on a whole bunch of data of what the key outcome is?

Bless you. The most famous example of this is probably IBM Watson Health. Back in 2014, I believe it was. IBM had this patient at the University of Tokyo who was had a form of cancer and was not responding to treatment. They kept trying treatment after treatment, she didn't get better.

So they took her genome, sequenced it, stuffed it into Watson, then they took 233,000 oncology journals, normalized them, stuffed that into Watson, and said, figure this out. Look for this woman's illness in all the data you have and identify it correctly. It did. They changed her treatment and she made a full recovery. Now what's really cool about that is it also did that in 11 minutes.

Unsupervised learning is when we have a bunch of data and we need to categorize it. We need to figure out what's in the box. You know, what kind of information are we dealing with? Because we don't know because there's so much of it. For example, I took all the followers of Spark.me on Twitter, and if you follow the account, uh you are in here, and took, I think it's about 75,000 tweets and digested them all down just to figure out what you were all talking about before today's event, because I had no idea.

And so we'd see things like uh well, I can't pronounce any of these words because I don't speak Serbian or Montenegrin or Bosnia or Croatian or any of these languages. So I'm hoping it makes sense to you. But that's unsupervised learning. We don't know what's in the data, but we can have machines sort it out for us. And then you get to what's called deep learning, which is when you start taking big, big piles of algorithms of these different machine learning types and stacking them up.

And the easiest way to explain that is you're essentially distilling down all the data into just the most important parts. Like when you take a wine and you distill it down to raquia, right? It's the same thing, it's just a much more compact package. That's what deep learning is. The outcome is machines that think like us, but better, faster, and cheaper.

Why do you need to know this? Why do you need to understand the hierarchy of AI? The reason is our Spanish friends would call it detection of excremento de toro. There are a thank you for those. There are a lot of vendors, a lot of software companies that say, hey, we've got AI in our product.

In 2018, the Financial Times did an investigation of a hundred companies that said they were using AI and found that 35% of them were completely lying. Not a stitch of AI in it. So having some of the understanding of the vocabulary makes you a better marketer, a better business person to understand when somebody might be pulling their wool over your eyes, when somebody might be trying to deceive you that their product does in fact contain artificial intelligence. What is the state of the art look like? These days, a lot of the conversation is about natural language generation, machines able to create content.

Here's an example. I took the first two paragraphs off the Spark.me website from this event, and I fed it to a language model called GPT Neo X20B, which is probably the most poorly named product ever. And I said, autocomplete the rest of this web page. And it did a credible job. Now it doesn't know anything about the event.

It doesn't know anything about uh, but it knows language. It knows the ability to what's common, and it translates and it creates new text. You do this with any kind of predictable text. I fed it a couple of paragraphs of press release, and it's actually written a better press release than the humans did because press releases are pretty stock. So this is what's happening now, what's available today in market.

There's some things AI can't do. Number one, AI, no machines can have empathy, because empathy is knowing what someone's experienced and then acting on it. Knowing what someone has experienced and then taking action on it. When somebody stubs their toe, you know what that feels like, and you can give them an ice pack, right? Machines do not understand, which means they cannot have empathy.

They have no judgment. If you've ever been to a restaurant and the server looks like they've been having a really rough day, and you greet them kindly and you talk to them to a little bit, every now and again they're like, you know, you're like the first nice customer I've had all day. Here's an extra free appetizer, right? They're exercising judgment. They're going outside of the rules for you because you're nice to them.

Machines can't do that because they literally are the rules. So they can't go outside the rules. Machines are terrible at general life experience. They are very good at narrow, focused tasks, but they can't go across tasks. And all things being equal, most people would rather deal with other people for relationships.

The exception to this is if your customer service is so appallingly bad that people would rather deal with a machine. In my country, in the United States, our uh Department of Motor Vehicles that issues driver's licenses is legendary for its terrible service. So are our mobile phone companies and our cable TV companies. In all those cases, we'd rather deal with machines because the humans who work there just are very unhappy people. So how are we using artificial intelligence today?

About 28% of marketers say they're using it for things like attribution, mix modeling, econometric models, and the other uh 70% are just guessing. It's just a bunch of tools. So, what are some examples of these? There's five sort of buckets of problems. Number one, uh unfocused problems.

You don't know what's working and what's not. So, attribution analysis is probably the most common use of a machine learning, where you feed a bunch of data to a machine and you say, hey, tell me what's working. So this is an example using Google Analytics, which pretty much everybody on the planet uses. You take all the data out of Google Analytics, you run it through a machine learning model, in this case, something called Markov chain modeling, and it says, here are the channels that are most impacting performance. And then you can say, okay, well, if I know what's impacting performance.

In this case, the big green bar is organic search. I know that 73% of my conversions are coming from organic search. I need to make sure that I'm doing a good job with it. I see that social media is at the very bottom is only 5%. And yet here I am posting all day on Facebook and Twitter and dancing on TikTok or whatever, and it's 5% of my conversions.

So attribution modeling helps you figure out okay, if I'm spending 25% of my time on social media, I'm getting 5% of the results. It's probably time to do something else. You can dig deep and figure out channel by channel. Twitter is working for you, Facebook is not. And it's all dependent on your data.

So these are tools that are available today. Second, you don't know what content is priority. You are you're unclear. When you do things like SEO, for example, there are a number of really great SEO tools out there that will tell you here's 10,000 articles that worked. What happens when you're given that much content?

Or even on your own website. Using the same type of machine learning, you can see all of the pages that somebody visited on your website and then see which pages nudged people towards conversion. We were talking earlier about people being nudged, right? This is this is identifying in your data which content works the best. In this case, it's a blog post on your public speaking career.

You can figure out which content is most efficient at converting people. Again, this is all taken out of Google Analytics, so it's a tool that you already have. You just got to get the data out and process it. But if you know what pages convert easiest, you can feature them in social posts or in emails and things or in advertisements because you know those pages are really effective at what they do. Third use case, you don't know what's valuable in your data.

You've got a bunch of data and it's you're just not sure. You know, when you take a look at all the people who are following the Spark.me account, that's a lot of information to process. It's easier to look at what people put in their biographies, right? You see a lot of people have uh Sonagoa with a Montenegro in their biographies. Makes sense, this is a Montenegrin conference.

But these are all the different terms that people have in their biographies. So instead of having to scroll through 2700 followers, you can digest it down and figure out what are the topics. So if you are Spark.me's social media manager, what are the topics that might be of interest to people, right? You have web developer being one of the top terms. Well, if you're not sharing a decent amount of web development content, perhaps it's time that you should.

You can look, for example, at when you should be doing things. When you look at uh your followers and how active they are, again, looking at Spark.me's data, digesting it down and saying, okay, if Spark.me wanted to improve performance at 5 p.m. in the Eastern time zone, which is 11 a.m. here, that's on Mondays. That is typically when people are most active on their own Twitter accounts.

So the folks who run Spark.me should probably make sure they have content going out at that time. How active are your followers, right? A substantial number haven't posted in 30 days. So you can digest things like your audience and understand what's happening behind the scenes and how to provide better value to your audiences. This is another example.

This was a hiring example. A company hired us to figure out why people were not applying for the jobs they were posting. So we took 5,000 job posts they'd put up, use natural language processing, condense it down all to a series of words. And we find things like CDL driver, which is commercial driver's license. You know, what type of license do you have?

Uh, is it valid, etc.? And then we took 17,000 phone calls from interviewers talking to these candidates, and they're talking about what's my pay per mile? Am I home on the weekends? How do I get uh paid for holidays? And so on the one hand, this company had all these job postings, and on the other hand, you have all these interviews.

Well, the what people were saying in the interviews, not a word of that was in the job postings. So we said, change the language you're using in these job postings. They did, and they saw a 40% increase in applications immediately, because all they did was say, all right, we're gonna answer the questions that we know people have. So using natural language processing. Fourth example is being prepared, using forecasting and predictive analytics.

Here, for example, is searches from the UK about the blue line on top there is flights to Montenegro. In early to mid-July, that's when that search term goes completely nuts. And as I understand, getting from Port Goriza to here in July is like a five-hour drive because you have all these tourists. And then in the late mid-to-late August, that's those search terms drop off. This is not last year's data.

This is this year's data forecasted forward. So if you're trying to figure out how do I time my products, services, or promotions to the tourists, the week they arrive or the week they're thinking about coming here, this is the kind of technology you would use. And this is true for pretty much anything that is has a cyclical pattern to it. So predictive analytics means that you can stop being so reactive and instead focus on okay, what are the things that are likely to happen, and can we anticipate and get in front of them? You can break it down then by week and then build a marketing calendar from that.

Building a marketing calendar says, okay, I know the week that someone's gonna be searching for flights to Montenegro is gonna be July 14th, 2022. Let's make sure that we have promotions in market for that. And then obviously you can look for things like hotel in Montenegro, etc. And finally, the fifth problem is you don't know who's who. From today, from this morning, who was most talked about at Spark.me, right?

We have Dean, Victoria. Uh we have Vladimir, because he tweets a lot. We have Sir Paul and the Spark. Me conference, and a bunch of other folks on there, many of you. If I were not here, I could look at this information and say who are the people who are most talked about, and then could I pitch them on whatever product or service I was working with?

So it's a way of identifying influencers within networks of data. And this is using Twitter, but it can be anything where you have connections between people. So a good example is academic papers. Every paper contains citations to other authors, to other works, to previous works. If you identify them with natural language processing and then build this network graph, you can understand who is most talked about, who is most cited.

And if you're trying to get in front of somebody, a subject matter experts, it's an easy way to do it. The graph is cool, but the data is more useful, right? This is the Twitter list of people you might want to follow from this event because everyone here has been talking about them. So that's what you do with this stuff. So these are the major problems that marketing in marketing, that machine learning has the ability to help solve.

The trick is you gotta get started sooner rather than later. And here's why. Companies that use machine learning, because it's based on data, the sooner you start gathering that data, the more of an advantage you have. If you wait and competitors get ahead of you, at certain point, they achieve such critical mass that you can never gather enough data to outperform them. This is why companies like Google and Amazon and Facebook are so hard to dislodge.

Not because they're better companies, but because the algorithms and the and the machine learning they have under the hood has so much data that they have a nearly unassailable advantage. Think about Google. It has search, so it knows what you type into search engines. It has YouTube, so it knows what you watch. It has Gmail, so it knows who you email.

It has Chrome, the browser operating system, which does record and send back to data to Google all the things you're doing. It has the Android operating system. So this massive footprint that is gathering up all this data gives it an advantage that frankly no other company has the ability to overtake easily. If you want to start a competing search engine, good luck. So how do you get started?

There's seven steps in your journey towards artificial intelligence. The first is called just data foundation. Do you know where your data is? And is it in any good condition to be used? If you are one of those companies where a developer put up a new website for you and they forgot to put your Google Analytics tag on four months ago, that's a that's a data problem.

So and this is where about where 90% of companies start and stop their AI efforts because they don't have the data. So that's the first and most important step. Second is to become data-driven, to adopt a culture of metrics and analytics. And this is really difficult. It's really difficult because you have two problems.

Number one, sometimes the data says things that make people unhappy and people don't like bad news. And the second is what's called the HIPAA problem in the United States. Highest individually paid person's opinion. And what that means is that there's some executive who says, oh no, no, I think this is what's gonna happen with the market, even though the data says clearly, the person is wrong. So creating a culture of being data-driven means overcoming those obstacles.

The third step is developing insights and analytics capabilities. Insights capabilities, especially because all the data in the world doesn't tell you why people did something. We can count up the number of people who chose to attend this event. But unless I talk to you or someone talks to you, I don't know why you're here. I can judge how much you're tweeting, right?

I can see how long you've been sitting in your seat, I can see how much you've been working on your phone and not paying attention, but I can't know why. So having those market research capabilities is vital. The fourth step is automation. And this is not AI, this is just getting ready for it. One of the biggest challenges people run into is that there's not enough time or resources to implement machine learning because you have to create time by automating as much as possible.

Now, at most companies, you face a fork in the road here. If machine learning is going to be part of your core competency, you're going to do the next three steps. If machine learning is just something you're going to do to bring efficiency to your company, you're going to look for these next three steps in whatever partner company you choose to work with. You need data science capabilities. This is the ability to process data and then use it according to the scientific method.

Data science is the underpinning of machine learning. To look at a pile of data and say, what's in what's in the box, right? What is possible to build? The sixth is the actual implementation of machine learning itself. And this is a technology play.

This is where you are trying to build technologies to start doing the tasks themselves, using algorithms, using code that machines write. And the seventh step, which maybe like one half of 1% of companies have managed to get to, is where you are an AI-first company, where you start thinking, how can I use machines to solve a problem? How can I algorithmically solve problems so that instead of solving a problem and then having it on your to-do list next week, you solve it once and then you never you never see it again. Again, very few companies have gotten to the stage. So, should you buy or build?

It comes down, the answer to this comes down to three questions: time, money, and strategy. If you've got time and you don't have money, probably you can build it, because so much of in AI, particularly the cutting-edge stuff, is actually open source software. If you have money and you don't have time, you're better off buying it. But most important is that strategy question. Again, if it's going to be part of your core competency, you want to build as much as possible because that is your secret sauce, and you don't want your secret sauce in the hands of another company.

If it's just for efficiency, buy it. If you're gonna buy it, chances are you've probably already got it, right? Literally behind me right now, we have artificial intelligence translating in real time what I'm saying. By the way, if it said something hilariously inappropriate, please let me know. And this is just built into Microsoft PowerPoint.

But the ability of machines to do natural language recognition and real-time translation, it's a product you already probably own. Right? If you use Google Analytics, most of the current version, Google Analytics 4, which came out in October 2020, is powered by AI. All of its ability to do attribution analysis, if you've got it set up right, which is a whole other talk, um, it's all powered by AI. Even it's anomaly detection.

Like it will literally just tell you things like, hey, I saw this because I'm I know you're probably too busy to pay attention. I saw this thing, why don't you go pay attention to it? Um you can configure it to just do the analysis for you to some degree. If you're gonna build your own, the tool I generally recommend for companies is a tool called IBM Watson Studio. Disclosure, my company's an IBM partner.

What it is is it's an environment that allows you to use the top two languages in in machine learning, which is R and Python, and build software with it that you can use for whatever tasks you point to. All of the examples that I showed today, those are built in the R programming language because I'm old and I don't understand Python. So how do you get your company ready for AI? You're gonna need three kinds of people, right? If data is the new oil, which is a saying that's been around since 2006, you're first gonna need developers, people who can write code, right?

And these capabilities could be in one person, but it's probably more likely it's gonna be a team. So you need developers first. You need data scientists, people who can take the data that developers have extracted out and process it, turn it into something, and then they need marketing technologists to take those models and bring them to market. Those are the three roles that you need. Again, it's a lot of data as a new oil is actually a great example because crude oil is pretty much useless.

It's dirty, it stains everything and makes a lot of black smoke when it burns, and that's it. You have to get the oil out of the ground, refine it and bring it to market. The same is true of your data. You've got to get your data from wherever it is in your company, refine it, and then bring it to market. Now, let's talk about you for a moment.

This is the number one question people ask about AI. Am I going to lose my job to a machine? And the answer is maybe. The Brookings Institute said this really well. AI will take away tasks, not jobs.

And that's true. Individual tasks, narrow focus things, machines can do, broad things they can't do. So here's an example. Let's say you have three workers that all perform these different tasks throughout the workday. Over time, what happens is you start to find ways to automate and to use AI to solve some of those tasks.

And then you get to a point where you have this, right? Where machines are doing most, you know, about half of those tasks. Now, at a progressive company, a company that's growth focused and maybe human-focused, you will say all that time that employees have now, let's get them doing new better tasks that humans are really well suited for. So that's progressive companies. At regressive companies, they'll say, let's divide up the tasks, give them all to just a couple people, and get rid of that third position because we don't need that seat anymore.

So this is what will happen to companies that are more operations and profit focused, not necessarily growth focused. So depending on what kind of company you work for, will depend on whether AI will take your job or not. So how do you prevent this? A few different things. One, developing multidisciplinary skills.

When you look at the top skills in demand on LinkedIn's annual surveys, you'll notice that most of these skills have relatively narrow domains, right? Cloud computing and stuff. There's a lot of things in cloud computing you can automate. But if you are good at cloud computing and you're good at sales, you're much harder for a machine to replace. If you're good at cloud computing sales and video production, you are really hard to replace because now you're crossing multiple domains of expertise.

Second, learn how to think algorithmically. That doesn't mean you have to learn to code, but it does mean you have to learn to think like a machine. How do I solve this problem with a system so that I don't have to solve it again? If you're the kind of person who's good at writing out checklists and processes, this is you. Then you work with a technologist to automate those things and to find automated solutions.

Third, an entire profession unto itself is machine oversight. Because we put the data in the machines, the machines build the software. Garbage in, garbage out. This is a real major problem with AI right now. In 2016, ProPublica looked at an example in the United States of policing, trying to determine if police departments were testing the software to predict what was called recidivism, the likelihood that someone would go back to jail.

The software was 20% correct. You can flip a coin and do better. But it identified black Americans as being five times more probable of recidivism than they actually were. There was a bias in the data, a very strong racial bias, and they eventually had to throw this thing out. In 2018, Amazon created an algorithm to forecast for hiring, trying to predict who would be good candidates.

Unfortunately, the data they put in made the machine immediately stop hiring women, which was someone probably should have thought about. Because the majority of the candidates they identified as great candidates were men. And it used those characteristics to make its predictions. Build a predictive algorithm where your good customers are. And you'll see on this map up here, there's red dots and there's black dots.

The red dots are where all the they think the good customers are. You'll notice there's no black dots down here. There's no red dots down here. Now, this is for a coffee company called Dunkin' Donuts. In in Boston, in all of New England, but in Boston especially, the only people who don't drink Dunkin' Donuts coffee are dead.

Everybody, every race, every age, every gender, they drink Dunkin' Donuts coffee because people like it. What this algorithm really did though was reinvent redlining. Redlining is a practice in the 1930s of taking a map of drawing large sections of the city and saying, I don't want to do business here because that's the black portion of the city or the poor portion of the city. No one thought to look for this, right? So this is a major, major problem.

It's probably the problem in artificial intelligence that we need to solve. Because so many folks who are building AI these days are not trained as data scientists, don't have any awareness of bias and kind of live in their own little worlds. And so from the movie Jurassic Park, you know, your scientists were so preoccupied whether they or not they could, they didn't stop to think if they should. And be outcome-focused, right? The machines can write their own software.

So our job is to figure out whether or not they're doing the right thing. It's kind of like a blender, right? How many people here know how a blender works, right? And could take, if I gave you a box of blender parts, you could reassemble a blender. Probably one or two people at best, right?

But you don't need to know that to make use of a blender, right? You don't need to know it. All you need to know is what it does and what's expected outputs are, right? If you put a bunch of fruit and some alcohol in a blender, you should get a drink. Right?

Now, if sand comes out, something's gone wrong. Probably you put the wrong ingredients in. But that's the way AI works. AI works like an appliance. You put something in, and what you get out is what you should expect.

So you have to be outcome-focused in your work. So, what is the future look like? How is this stuff gonna play out? What's happening now that is really fascinating is that machines are getting so good at writing that this is an example from Google, they're just starting to rewrite our content for us and display the answers that they think predictively would be a better match than what's on our website. So all the time and effort that we spend on things like content marketing and SEO may be somewhat moot because the machines will simply rewrite stuff to be better.

And I've seen a number of really interesting startups in this space where you feed it copy and it rewrites it to be either slightly better or to be more optimized. Um if you ever want to try one of these things, take some like fan fiction off the internet that's poorly written and see what the machines do. It actually can improve things. Machines right now are only as good as the data that they're given, and most of the data they're given is mediocre. But there have been big, big leaps in the last 10 years, the last three years especially.

It used to be, for example, machines could not write, right? You would say, generate some text for me, and it would spit out like word salad. Now it spits out coherent, sensible, mostly mediocre writing. Machines are really good at cranking out press releases because they're mediocre to begin with. And so if you're a great writer, your job is safe.

If you're a good writer, you need to become better. And if you're a mediocre writer, machines can do your job better than you are. Than you can. And if you're a terrible writer, sorry. This line keeps getting higher and higher every year of what machines can do.

So today they are mediocre. Three years ago, they were terrible. Three years from now, they'll probably be good. They will probably never get too great because, again, data is taken to train machines. There's so little content out there that's actually great that there's not enough for machines to train on.

But there's a decent amount of good. There's a whole bunch of mediocre, and there's an ocean of crap. So as time goes on, machines will get to good. So if you are only at good, you have an immediate imperative to get better, to improve your skills in anything that a machine does so that you're not left behind, so that a machine can't do substantial portions of your job. But this is the future.

It's already happening now, and your mission is to make sure that you stay ahead of the machines. How do you do that? Creativity, true creativity is the only way out. Because, again, machines are trained on what they know. If they've never seen it before, they can't create it.

So if you are truly being creative, you're truly being innovative, you can beat the machines. Because you're creating something you've never seen before and that they can't train on. Now, I will close on a positive note. One of the things that machines can do that is spectacular is enable abilities we don't have. So for anyone who English isn't as strong, there's at least some option for you.

There's a tool called Gaugan by NVIDIA that will take badly drawn hand drawings and turn them into photorealistic landscapes that look better than what's possible than what I could do. There's another one called Wambo on the App Store that does exactly the same thing. So even if I have the if I have the desire to create something, but I don't have the ability, a machine can augment it. There's a tool called Ava that generates music. And like most uh AI, the music it generates is between mediocre and good.

It's mostly on the mediocre side. But guess what? I can't play any instruments. I can't sing, I can't carry a tune, I can do none of those things. But with AI, I can do the sum of that for me.

I can give some direction and then it can create. So it allows me to exceed my natural limitations. And that's what these machines promise for us and for you. If you have the desire, but you don't have the talent in a certain in these certain areas, machines may allow you to realize all of your potential. So with that, thank you very much.