<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=2006193252832260&amp;ev=PageView&amp;noscript=1">

26 Min Read

Watch Chris Penn's Keynote from MAICON: Become an AI-Enabled Company (Video)

Featured Image

Wondering how to get started with AI? Take our on-demand Piloting AI for Marketers Series.

Learn More

In the words of Christopher S. Penn (@cspenn) at this year’s Marketing AI Conference (MAICON), “Machines think like us, only better, faster and cheaper.”

Penn, Co-Founder and Chief Innovator at Trust Insights, joined us at the inaugural event to discuss how artificial intelligence (AI) will change marketing forever.  

He emphasized that while, yes, AI will transform our trade, we must also understand where it falls short to succeed. AI can’t accomodate for empathy (it can’t feel), judgement (knowing when rules don’t apply), general life experience (diving deep into a subject) or for relationships (that’s still a human-to-human strength). 

With that in mind, though, AI can be used to solve some of our biggest marketing problems. During his talk, Penn lists the following five practical applications.

  1. Untapped Data: Most marketers have a ton of data that is left untouched. With AI, you can quickly make use of data you have to identify trends and uncover patterns. 
  2. Unknown Influencers: It’s critical that we always have industry influencers on our radar. With AI, you can easily run an analysis to see who is talked about most (not just who talks the most) without the manual research. 
  3. Unclear Priorities: Even if you do know what data you have, do you know how to leverage it to take action? Take your SEO for example: You might have a ton of keyword data, but do you know which ones to use?  Use AI tools to identify what competitors are ranking for (or aren’t) and react to take market share away from them. 
  4. Unfocused Data: This one is also about taking the right action, but not knowing which data is key. You can focus your data by using it to sequence the customer journey and identifying which actions have the most impact on conversion. 
  5. Predict the Future: We have no way of knowing what will work best for consumers next. But, you can use AI to make a predictive forecast based on organized data. 

Bottom line from Penn: “Take the data, train a machine, and learn from it.”

In the second part of his talk, Penn overviewed a seven-step process to become an AI-enabled company—and solve for the problems above. For that full process and more, watch the video below, or read the full transcription. Please note that the transcription was compiled using AI with Otter.ai, so blame any typos on the machine :)



Session Transcription 


Someone had the coffee! They're going to be two kinds of jobs left in the future. Gonna get to those in just a second. Number one question, of course, we can get the slides for this talk and get them. Where can I get the slides calm? marketing is in an enormous amount of trouble. Right now, every single one of you in this room is being asked for the same three things by your customers, by your board by your cmo. Make your marketing better, faster and cheaper. How we doing? Let's try better first, how much better is our marketing getting? This is the amount of data that we as human beings in our civilization are going to be creating in the next few years this year and 2019 we're going to create about 40 zero bytes of data. Now you go to the digital camera store or shop on Amazon, there's nothing that's there's there's no one zettabyte How big is the zettabyte? If you were to start watching Netflix 55 million years ago, and you didn't stop to eat, sleep or use the restroom, you just get to one zettabyte. Today, we as a civilization are going to create this much 40 of these this year. That's an overwhelming amount of data, guess what none of us combined can handle processing that much data. We also have data quality problem. How many people have this person in your CRM test at test com? So we have lots of data, and a good chunk of it's bad. What's the impact of this? Well, in the most recent cmo survey from February 2019, CMOS when asked, How often do you use data to make decisions analytics to make decisions an all time high of 40 3.5% said they do, which means the other 56.5% of the time that is guessing or making stuff up. And yet, and yet when asked, okay, well what's your number one priority? Demonstrating the impact of marketing and financial outcomes?


So on the one hand, we want better outcomes. On the other hand, we're guessing so marketing, not doing such a hot job it better. How about faster? Are we as a marketing getting faster? Our marketing isn't by the world Sure is. This is one minute on the internet in 2018 and 2019 to 2018 266,000 hours of Netflix watch 38 million WhatsApp messages hundred 87 emails 1.1 million Tinder swipes. Fast forward only one year. We're at 41 million messages 694,000 hours of Netflix watch 1.4 million tennis webs. I'm guessing mostly right because that is a huge company amount of net clickstream watched.


For those of you who use Tinder... That was funny. The world is getting much, much faster, how much faster as of this year in terms of the number of just raw news stories on the internet. We're on track for 122 million news stories this year. That's 178 minute and hundred and 78 news stories per minute. Now, if you're a marketer, and you got a great piece of coverage, say in the New York Times, or The Wall Street Journal or on CNBC, guess what, it's gone in seconds, because the next story has come along. So we're not doing such a hot job and faster and keeping up. But we must be getting cheaper, right things costs are going down. Well, again, when asked only 36% of CMOS and roughly a third, were able to say we can quantitatively prove the impact of our marketing spending. 50% said we have a good


qualitative sense, which is code for we're guessing and 30 percent said effet. I'm not even gonna try. So two thirds of us have no idea. None. How our marketing money is being spent, and yet we're spending like drunken sailors, right? every category of spending is largely going up. So we're not better. We're not faster. We're not cheaper. how's this going to end for us? How do we fix this? How do we get ourselves out of this problem, unsurprisingly, and the reason we're all here today is looking at using artificial intelligence, which has three benefits, acceleration, accuracy, and automation, acceleration, can we make things go faster? Humans are so limited compared to machines in terms of the speed at which data can be processed. Can we make things better? And can we stop doing crap that frankly, we don't want to do on surprisingly better, faster, cheaper? Karen and Paul talks a lot about what is artificial intelligence. It's just getting machines to, to do human like tasks. Jim stern and I were chatting yesterday and jokes that AI, if it's if it's on a PowerPoint, it's AI if it's actual code, it's machine learning, right. So that's a fun descriptor. But we see this we understand the evolution, right? humans evolved from very simple inputs to language to higher cognitive function. That's how humans evolve. By the way, as a parent of a teenager, I appreciate the fact that cognitive function goes right down the trend. 13 machines are the same from algorithms to machine learning to deep learning to general purpose AI.


And where we're going. AI is math, not magic. It is math. That's all it is. At the end of day, it's stats and probability that you glued together into algorithms. Every single one of you is using algorithms in your marketing. And even in your day to day life. This up, I would bet you a small pastry retail value of $5 or less, that you probably put the same general article of clothing. Every single day in the same order, right? Some people put the top on first and put the bottom on first, a few weirdos put their socks on first. But you have an algorithm that gets you a clear outcome. As Karen talked about where this flips over is when we start talking about how machines learn, right? How do they learn? How do we flip over from write the software to do the thing to here's some day to do the thing for me. Imagine you had a table of blocks like a kid. What are the different ways that you could process this you might have some desired outcome, maybe you want to find for example, blocks of a certain color is a branch of machine learning, called supervised learning, teach the machine to look for a specific outcome. Find all the blocks that are colored red, so we train it on everything that's red, and give it data and hopefully we can find the color red. So all the blocks that are red, the most famous example of this was from 2013 in the University of Tokyo, actually the idea IBM team working with their oncology center had a person who was suffering from leukemia. And they were treating this person they weren't getting better, they weren't getting better. So they took this person's genome, sequenced it, fed it to Watson then took 233,000 oncology journals, fed that to Watson said, figure it out. What are we doing wrong? We were looking for this what what should we be looking for Watson did all that and found that they were treating the wrong type of leukemia.


They're treating the wrong time they changed the treatment, the person made a full recovery. Now what's cool about that a, the person recovered, and B Watson did it in 11 minutes. Right. So better and faster. But suppose we don't know what kind of blocks we have. Suppose we just have blocks, we need to categorize them. We get what's called unsupervised learning, instead of just colors. What are all the different ways we could classify these blocks? There are the different colors, the different shapes, different sizes. in marketing, we face enormous amounts of messy data did we have no idea what to do with? I was getting ready for a client meeting a little while ago, and had 71,000 paragraph 200 to 2600 articles that I needed to summarize for the client to get ready for a meeting, I had 10 minutes left. We'll talk about my lack of planning later. So I fed it to a text my topic model and said, okay, machine, tell me what's in the bag, tell me what's in this box of articles so that I can then go tell the client, what they were, what they what coverage they got, and the machine did it. less than five minutes I made the meeting on time you had time to get a cup of coffee. So unsupervised learning is a way of sorting through massive amounts of data. And then the deep learning just gluing stuff together like stacks of pancakes where instead of syrup going from one layer to the next, it's data. What you get a machine to think like us but better faster and cheaper from very specific tasks.



The reason you need to know this stuff. These hierarchies and all the structures and the reason why we're here is not because any of us expect you to become a machine learning engineer or data scientist, right your marketers that is your profession that is your trade. Why you need to know this is so that you can develop a very finely tuned detector of what our Spanish friends would call experimental disorder. Because every single vendor on the floor here and it every marketing conference, you're going to go to From now on, every vendors gonna say, hey, our product uses AI, or even though 35% of them are lying. By the way, the Financial Times did a survey of hundred companies that said they used AI when they pull back the covers 35% had zero in there. So by familiarizing yourself with at least the basic terminology, you can help you can start to develop that bs detector and be able to ask Okay, what kind of machine learning to use when you're evaluating vendors, what kinds of machine to use, what kind of data We're going to need in order to make use of it. The state of the art is actually really interesting. It's a blend of what Karen called those symbolic systems and and the deep learning systems. IBM this past March did this thing called Project debater they took a human debate champion put them on stage, it took project debater, which is a series of 10 interlinked a eyes, put it on stage, no prep 15 minutes before they said you were going to debate the importance of subsidizing kindergarten and both of them had to craft arguments. They each spoke for four minutes, and then they do rebuttals to the machine had to listen to the debate.


The other debater form rebuttals, deliver rebuttals and then have closing statements all done in near real time. So that combination of symbolic and construction of systems is the reality today. Imagine being able to use this for things like customer service where you can actually talk to them. machine and have a deliver a reasonably good experience.


Now there are some things that AI really is bad at. Its bad at empathy, which we define as knowing what someone is feeling and acting on it. Machines cannot yet feel machines cannot do judgment. Judgment is knowing when the rules don't apply. When you go to a restaurant, and the server brings out an extra appetizer when you go to a hotel, and the closest upgrades you because they like they, they appreciate the fact that you are a polite human being to them. These exceptions cannot exist right now in today's machine learning because they literally are the rules. Machines are really bad at general life experience and domain expertise digging deep down into something. But especially across domains machines are terrible at it. And for the most part, humans are still better at human to human relationships than machines. So these are the things machines can't do so what can machine Do what are the practical applications? We've heard a lot of theory and a lot of good foundational work. What are some practical applications of machine learning at work in marketing today? There's five kinds of problems in marketing that machines are really good at untapped data, unknown influencers, unclear priorities, unfocused data, and unpreparedness. For the future. Let's dig through each of these. You have somewhere in your organization, a massive pile of data in your CRM, it's in your customer service inbox, it's in your call center, and you're doing nothing with it. You haven't your social media systems everywhere. How could you make use of this? As an example International Women's Day was not too long ago, on that one day alone. There are 1.6 million social media posts. When you boil down the amount of texts that was 8960 novels written in one day. I can barely read one novel a day, much less almost 9000 of these but if you As a brand wanted to participate meaningfully in that day's theme. Could you have read all that so that you knew what to talk about? The answer's no. But using a combination of text mining and clustering, can very quickly surface what are the themes that were happening as they were happening so that you could work with your teams to say, should we participate in this conversation? Or not? Do we have anything to offer a value or not?


By seeing it quickly? You can make decisions on the fly. Another example of untapped data, we worked with a truck driving recruiting company. It's a mouthful, and said we want to help you improve your recruiting. So we took all of their job postings on indeed.com. And again, using topic modeling and clustering, figured out what are the things that show up most importantly, and most frequently in the job descriptions and it's all about the truck driving company. We expect you have this. We want you to have this. We want you to have this you should be able to do this and then we took steps 17,000 phone calls from their call center used AI to transcribe them and then analyze those. What a difference instead of licenses and classes and decorations and hazmat. It's what's your paper miles starting paper I'll driving record all these conversations. So when we go back to the recruiting company, let's say on the one hand, your job ads all say this, on the other hand, zero what the candidates want to talk to you about on the phone, or in your job ads, and you wonder why applications are going down? Because the candidates literally could not care less about what's in your job ads. Now think about your marketing. What are your customers asking you about and talking to you about and then what's in your marketing? Are they dining even in the same restaurant?


The second is identifying influencers influencer marketing, everybody's favorite. It's the tools that are out there right now are so primitive, and there's better solutions. This is from the Adobe summit recently. A good measure of influences who is talked about right now who's the loudest mouth? Who has the biggest foot number of followers, but who is the most talked about? And again, being able to look at conversations as they happen in using technology like network graphing, and clustering, you can boil down all those interactions into a very simple spreadsheet that says, If I want to talk to people who matter most, for example, at a con, who should I talk to run the analysis as the events happening and know who to talk to in the audience on stage, you have to talk to Paul Of course, but influencer analysis is can be much more than it currently is today, which is simply who's got the biggest mouth. Even when you do know what data you have? Can you make sense of it? Can you figure out ways to take action on it? That is one of the biggest problems with data, what action should we take? Suppose you are doing search engine optimization This is an example for an insurance company they are ranked for hundreds of thousands of search keywords? How do they know where to start? What should they optimize for? What should they try and process? You can't do that from from just the data itself. But again, using something like very straightforward k means clustering. You can identify what pages on their website, are they ranking? least well for? fix those up? What pages? Are they ranking, the best for? double down on the content that works on those pages so that you can protect the company? And then you go after their competitors? And say, what of the valid good data we know about their competitors? What are the terms, the words the phrases, the topics that their competitors are bad at, and take market share away from them? A fourth example, suppose you have a lot of data, suppose you have no idea how to prioritize the data to take action from it. A lot of people love to talk about customer journey mapping, but what it ends up typically is being a bunch of people in a room with posters yelling at each other. Right? What if there's a better way to deal with that data? You have have that data already, chances are if you use something like Google Analytics, or Adobe analytics, or matumbo analytics, you have that data, you just don't have a tools to sequence it. But again, using machine learning a technique called Markov models, you can take every single input and the way people find you, and sequence it out to map the customer journey, not only by when does somebody interact on a certain channel, but what is the impact of that channel on conversions so that you can say, okay, we need to double down on organic search, we need to double down on speaking we need to double down on newsletters. We maybe don't want to do Instagram, because it's just not delivering. Right. So again, using machine learning, we can figure out what is the most impactful things that we should be working on. And one of the things that maybe we should leave behind customer journey maps, by the way, if you've ever sat through one of those meetings, now is it the full eight hours of the day, but then it's two and a half months of meetings afterwards. agree on a map that is out of date by the time you're done. When you use machine learning can be done in about 10 minutes and get on with your day, right? So you use this type of modeling, bring it to the meeting, make decisions and get on with your day faster marketing. And you can use this with any data where you're not sure what outcome and what what contributes the outcomes you care about, for example, this this one customer, they want to know what got them more Instagram followers who might argue that I think that's a stupid KPI. They said we want more followers, okay, run it through, in this case, gradient boosting and figure out Yep, it's it's the addressable engagement rate your audiences make stuff that people like. And you can do it even with what people are doing on your website. As somebody goes from page to page to page on your website. software like Google Analytics is tracking that. They can watch from what page to page after page and then see if it convert when you pull that data out of Google Analytics and model it again. Markov chains simulations, you can identify the most pages that were most likely to help somebody convert. So now you have an SEO priority, you have a content marketing priority, the pages that are converting the best, hey, maybe it's time to put some video on that. Maybe it's time to record a podcast about that. Maybe it's time to write a sequel about it.


Whatever the case is, the data gets converted to analysis that can convert it to an insight that you can then use and turn into action. And finally, dealing with unpreparedness how many people just a show of hands feel like how do people do content marketing of some kind, just raise your hands, half hearted if you haven't done enough coffee, okay, most of the room. How many of you have in the last month or so, felt like you've been scrambling to just put something together to meet your deadline, but it was the best thing you've published in your career. You can escape that trap. Using time series forecasting, predictive analytics. Look at all the words and phrases of topics in your industry. And then forecasting for to come up with plans for what to do and when You can catch more information about this at the four o'clock session of room 2016 with my co founder Katie robear. She's going to talk for 45 minutes about how we how you can do this. So, these are examples of machine learning in use in marketing today, right? Not futurism, not guessing, not idle speculation in production today. If you have any of these problems, you can solve them today with machine learning, but the most important thing you need to do is get started. Remember, machine learning is all about take the data, train the machine and learn from it. The longer you wait, the less time you have to collect to collect data. And if a competitor has started and you have not your competitor has a head start in data collection. That is very very, very difficult to overcome. If you wonder why has no one beaten Google at search, why has no one beaten Facebook and social media it's because they have years of data that Essentially let them crush a competitor because they are machine learning algorithms have learned more.


So how do you get started?


There's a seven step process a journey to becoming an AI enabled company. The first step is a strong data foundation. If your foundation is weak, if you have no data, you don't know where your data is, you don't know what format it's in. You can't do anything else.


So you have to get this part straightened out first, where is your data? Who owns it? Who's responsible for it? What format is it in? Your next step is to become a data driven company. If in the halls and meeting rooms at your company, you hear the phrase, well, this is way we've always done it. I would update your LinkedIn profile right now. Because your company is doomed. If on the other hand, you're a data driven company, where you're making decisions with data where you have KPIs that are connected to reality, you are ready to move on. By the way KPI is a really simple thing to define. It's the number for which you will get a bonus or be fired for. So when you're looking at your data architecture, think of it in that lens, what data what analytics, will I be given a bonus for or fired for? Everything else is less important. Your third step on this ladder is qualitative research capabilities. As good as AI is getting, we still cannot crawl into people's heads and ask them why. Why did you do something? Why did you abandon the shopping cart? Why did you not make a purchase today? only asking people and then analyzing those answers will help us fill in those blanks and train our models better. So you need to have market research capabilities.


Fourth, machine learning and artificial intelligence, our professions, right and therefore they require a lot of time and a lot of effort to learn to deploy and things like that, which means that you need to make time in Your organization if you said Oh, I just don't have time, you have to make time. So things like basic process automation to free up your time. What can you automate with very simple tools today to give yourself time to learn time to grow time to identify the correct vendors. Now it's at this point in your journey where you're going to go off for one or two roads. If you will be applying AI to your core competency in the next three steps are directly something you will have to do. If on the other hand, you just want to make your marketing better, faster and cheaper. These are these things to look for in your vendors. Number five is data science data science capabilities, the ability to explore the unknowns to build statistical and mathematical capabilities, code and engineering. Data Science is a weird mix up of stats and programming. And it takes a lot of effort to skill up on both. Once you have data science capabilities, the natural transition is into machine learning, deploying those models in production using the different types of machine learning solve problems. Eventually you reach the dream, as Karen said, of being an AI first enterprise, there are very few of these companies. So if you feel like oh my god, I'm so far behind, don't worry. They're like five on the planet, that are companies that are truly AI first where they try to solve everything with AI first. So the question is, should you buy it or build it comes down to time, money and strategy. If you have time and you don't have money, you're building it. You have money, you don't have time you're buying it. But most important, again, if you're going to apply artificial intelligence machine learning to your core competency as a company, you're building it because you do not want your secret sauce in the hands of a vendor. Because that vendor can say, Well, good tomorrow SAS fee is going to be 10 times what they are doing and I know you can't do anything about it. So if you are going to be applying AI to your core competency as a company, you must build it. If on the other hand, you just want faster marketing. You have a selection of vendors. If you do choose To buy it, there are a huge landscape of tools. There's all so many of the vendors on the expo floor stop by throughout the day and say hi to them, ask them tough questions. But some that you can use right now today, there's an app that I use called otter, which is a transcription app can record meetings and phone calls subject to local laws, and then get transcripts that you can search through later on. This is how we did the call center project. But I use this literally for almost every meeting. I'm in just so that can remember what I said later, because I didn't have enough coffee that morning. And as Paul said, early on, you're getting AI whether you want it or not. apps like Google Analytics or already throwing it in that little insights icon in the upper right hand corner if you ever tap it. That is Google's anomaly detection. It will say to you Hey, I noticed something strange on your website. You want to take a look at it. You can ask Google with little natural language recognition. Tell me what happened on my website yesterday and it will give you some okay answers but not Great yet, but they'll get better if you're going to build it. If you want to create this core competency, the tool the system I recommend most because my company is an IBM Business Partner is IBM Watson studio. just build in the methods that you like best whether you like clicking around or writing code. If you go on to write code, the two languages that you need to do to learn are either R or Python. As a matter of personal preference, which one you learn I like to joke that if you like avocado toast you like Python.


Now all this sounds great, you're ready to go right?


There are things that are going to go wrong and your first projects expect these things to go wrong. Try to prevent against them. When you look at the overall lifecycle of a machine learning project. It's this large rainbow. And I can tell you there are areas in each stage of this where you will go wrong. In the business requirements section that is the part where AI goes the most wrong. You're seeing Or your board or whoever comes to you and says, Hey, we need to get some AI stuff going in our product. That is the recipe for disaster technology for technology's sake, what is the business problem that AI can solve? Right? If you don't have that clearly outlined, it's going to blow up. Because you can't then choose an analytical approach. If you're just being told, hey, use some AI, at that point may just make like a facial recognition app with a cat and just call it a day. And the next stage and the data stage of a machine learning project, where you will go most wrong is on data requirements. Because what tends to happen, what I screw up all the time, is I'll start writing something, I'll start doing something this can be so cool. And then I'll get to a sort of an MVP. And then I see ob, like, did you think about this, like, Well, no. Shit, and go back and have to restart from the beginning.


What data do I need, what format does it need to be in one of the requirements? What is some going to ask me about at some point. So doing those data requirements is where you will most often go wrong, followed by data collection, because so many companies don't have data, or they don't know where the data is located. I was on a call the other day. And somebody said, Okay, we're ready to do this, this project. As a great, we're going to use a Google Analytics data. And they said, We forgot to install Google Analytics on our website. Well, it's gonna be a fun project. And the next step, model selection and model evaluation, where this goes wrong most often, is when you have an inexperienced data scientist who either just make something up or just picks them up the only model that they know, there are hundreds of different models suited for every type of situation for every type of business application, so not have so having someone who did the, you know, six week Crash Course and machine learning, you're going to end up with a lot of wasted time in this space. So having an experienced data scientist who can move At the problem, say this models probably going to suit your needs and then be able to evaluate and choose even what measure to look at should you be looking at r squared rmse ai CR OC. Knowing all those pieces is essential to being able to make this part of model deployment work of model selection work and finally, we're a so many companies go wrong as they don't mind the store. Once a models in production once you're out there predicting who's the next best customer, or which tweet is likely to get interacted with best. People just assume the system is smart and does its thing. Now it's like a car if you take your hands off the wheel, you will get what's called drift and eventually you'll hit a tree because you have the road changes or the car changes. Microsoft had a fantastic public very public failure at this when they created a the Twitter bot that people could interact with no one minded the store and 24 hours later it was a racist born bought right We laugh, but holy crap, their model went off the rails really fast, because no one was minding the store. Of course, just a plug for our friends, the project managers. Please look at the successes of your projects and figure out what you learned. So you can figure out what to do better next time. So let's talk about how to prepare your company. For artificial intelligence, you're going to need three kinds of people. As Karen said, great quote from 2006 data is the new oil. I agree with this because if you've ever seen crude oil, it's nasty, smelly, useless stuff. It's black, it's Tari. It makes terrible smoke and does nothing except stain your clothes.


To use crude oil, you have to extract it from the ground, refine and turn it into something useful, like you know, cereal bowls or gasoline. It's no different with machine learning. You need developers who can find the Delphi energy and extract it by writing code to extract it from all the systems throughout your company wherever they are public data sources those developers can help you with code you need to interact with all these systems. You need data scientists who can take that raw data and transform it, refine it, model it. By the way, if anyone's interested, this is the algorithm for figuring out influence on social media. And then you need marketing technologists, people who can take the model and do something with it, put it in production, create business impact. If you are a large company, do it in these in this order. Have your developers start building they collecting the data, get data scientists, and they get marketing technologists to deploy it. If you are a small company do it in reverse, because there will be enough small wins from that a marketing technologists can bring to the table to create incremental revenue that will then let you afford these other roles. Because developers and data scientists are really really, really expensive out in Silicon Valley. Your average qualified data scientist is going to run you anywhere from $300,000 a year and up, right? Someone who's highly qualified, who has your multiple letters after their name, you're talking a minimum half a million dollars. And you'll keep that person for 14 months, and then they'll move on to Google probably. What about you? What about your career? The question everyone wants to know the answer to who's going to lose their jobs?


Depends on your job. The Brookings Institute had this great quote, AI will take away tasks, not jobs, individual tasks will get suborn to machines and the tasks you don't want to do. Right? I wrote some code to build social content, because got tired of spending two to eight hours a week curating content for what I want to share on social media. So I wrote code to do it, and have a very simple machine learning algorithm to score what should be shared, and now it takes me five minutes, I don't want to do that job anymore. I don't wanna do that task anymore. Let's outsource that to the machines. Now where you will run into trouble is if your Job is one task.

And that task is I don't know, like stocking shelves or cleaning up the supermarket floor, then yes, your job is at greater risk because there's not a lot of additional value that you offer.


So how do you prevent against that? A few different strategies. Number one, you need multidisciplinary skills need to embrace all of what you can do, instead of being a hyper specialist in one thing, if you look at the top 10 job skills that LinkedIn says are in demand right now, there are tasks within each one of these that I absolutely will do and should do. But if you're good at multiple different skills, you are very, very difficult to replace. If you are good at sales and you're good at people management and you're good at analytical reasoning, you are very difficult to replace, because you have expertise across domains. Second, you need to think algorithmically think like a machine, instead of saying how can I solve this problem right now and get it off my plate so I can move on to the rest of my to do list. If you learn to think, how can I solve this problem for ever? How can I build a system to solve this problem, so I never have to. To solve it again, I can just tune the machine. That's an important skill. Third, you need to be able to oversee the machines to tell them what to do. From when does a model go wrong? For example, facial recognition is relatively easily confused by things like Insane Clown Posse makeup, it turns out that that screws up the edge detection. Karen and I both have the example of face mapping. Right? The number one job in the next five years in AI is going to be this one. Applying ethics is a great quote from the movie Jurassic Park. Your scientists were so preoccupied whether they could that they never stopped to think about whether they should when you are building or even buying machine learning an AI technology. Ask yourself, How can this be misused? How can this go wrong? Amazon gotten an enormous amount of trouble last year, because they built an HR hiring system that when turned on only hired men, because they only trained it on their past history, which of course, predicted mostly men as hires. So being able to oversee the machine and say, what's likely to go wrong, and how can we prevent it is going to be an essential skill. If you want to get a great free book on this, by the way, his book by Dr. Hilary Mason, called ethics and data science as her and Michael kitas, it's on Amazon is for free. If you want to do anything in machine learning, please read this book. And you have to become outcome focused, do not worry or stress about becoming a programmer. I used to have this cheat sheet hanging up my cubicle I was trying to learn keras which is a deep learning framework. I was slowly taking off functions and things as I learned, you know what it was why it was important how it worked, right? writing code and it was really bad at it. And then went to IBM think in 2018. And one of the people there looked at me funny. So why are you doing that? Just we have a drag and drop thing that does that now. So you need to know what the pieces do. But you don't need to like figure out where the semi colon goes. old shit. It was 18 months on that, getting back this year. They said don't even bother doing that now just drop the data and it'll produce the model then you inspect the model, you decide how to tune it. But for the most part, even coming up with the modeling is being automated AI is being used to generate AI. So we have to be outcome focused, what outcomes do we care about? You have to have the job title of Chief questions officer, what question should we be asking of our software of our modeling of our statistics of our outcomes? So where's this going? Probably the thing is going to impact your marketing in the next six to 12 months. That will just totally change how you do marketing. is now in the field of natural language generation. In the last nine months or so, the field overall has had some incredible, incredible advancements in getting machines to write coherently. Because writing is the foundation of so much of marketing. This is an example of a system called GPT. Two from open AI, which has a pre built model trained on few hundred million articles on the web. And then you give it some additional data to tune it up, and it will start writing stuff for you. In this case, I took the marketing AI Institute's tweets from the last year, fed it into GPT to said, Okay, I want you to learn how Paul and his team right, and then write me net new tweets. I also did this with a well known politician who's not known for making intelligible tweets. And I fed that person's tweets in and it spit out really good imitations. I told it As a starter, please declare war on North Korea. And it came up with these holding all these things that were if you didn't know that it was machine generated, like wow, that that seems pretty credible. I guess they really are really bad people, those agents.


You'll be using this to generate blogs, email newsletters, long form content, articles, scripts for your videos, scripts for your podcasts. And as the technology gets better and faster, it'll become more and more accessible. This today requires some pretty intensive hardware that you can actually rent from Google. This cost me 44 cents to do this project because it was Google's compute time. I should add By the way, all the other examples that we've been showing are not hundred million dollar computers. Almost everything in this presentation was done on my laptop. Deep fakes you can do on a video gaming machine if your if your machine can play. You know Call of Duty your machine can do deep fakes. So I said at the beginning, in the future, there are going to be two kinds of jobs. Either you are going to manage the machines, or the machines are going to manage you. And there's not much between the two.


You're already managed by machines. Partly How many of you look at this thing as the first thing you do in the day? Right? You're being managed by a machine right now. So the question is, How much more do you want to turn over? or How much more do you want to take charge of your life, your career, your marketing, and be the one who conducts the orchestra of machines. Thank you very much.

Transcribed by https://otter.ai

Related Posts

Chris Penn Has an Important Reminder About AI: It's Math, Not Magic

Paul Roetzer | February 25, 2020

Christopher S. Penn, Co-Founder and Chief Data Scientist at Trust Insights, sat down with Marketing AI Institute to share some advice about AI.

Marketing Artificial Intelligence Conference (MAICON) 2020 Sponsorship Kit Now Available

Sandie Young | September 17, 2019

Connect your marketing AI technology or service to an ever-growing community of marketing leaders seeking to drive change in their organizations. Request the Marketing Artificial Intelligence Conference (MAICON) 2020 sponsorship kit today for details.

How to Transform Your Content Marketing Strategy with AI [Free Ebook]

Mike Kaput | February 16, 2018

Content Marketing Institute and the Marketing AI Institute, share how marketers can use AI to boost their content marketing performance.