Subscribe: Google Podcasts | RSS
The machines are coming! The machines are coming! Artifiicial Intelligence is here. But what is it, and how long will we have to wait for the technology to completely take over all analysis work? Dennis Mortensen — founder of x.ai — joins us on this episode for a deep dive into the topic. You will be surprised by how pragmatic and real AI seems as Dennis describes how he approaches it. And…then his last call will completely blow up the nice, cozy layer of downy comfort that you’ve settled into during the discussion. So it goes. Artificial intelligences and things referenced in this episode include:
The following is a straight-up machine translation. It has not been human-reviewed or human-corrected. We apologize on behalf of the machines for any text that winds up being incorrect, nonsensical, or offensive. We have asked the machine to do better, but it simply responds with, “I’m sorry, Dave. I’m afraid I can’t do that.”
[00:00:24] Hi everyone. Welcome to the digital analytics power hour.
[00:00:29] This is Episode 44. Imagine a future where your Web site will optimize itself based on what’s best for users.
[00:00:40] Artificial intelligence while it’s not necessarily a core part of what an analyst does today the concept of machine learning is already moving into every space. A couple of years ago Stephen Hawking famously said that artificial intelligence could spend the end of the human race. But until then it’s fun to experience it as it becomes more and more a part of the real world. Hi everyone I’m Michael Hoeben analytics practice leader at search discovery and of course I’m joined by Tim Wilson My cohost senior partner at analytics demystified. Hello Tim. Hey Michael. And who better to be our guest on this show than someone who has spanned both the digital analytics world as well as A.I. our good friend Dennis Mortensen. Dennis is a serial entrepreneur. He’s had many successful exits. He was the CEO of index tools which was acquired by Yahoo. He then founded Visual Revenue which he successfully had sold Outbrain and now he’s the CEO of X dot which is an amazing company that takes over scheduling meetings for you with the power of machine learning. Welcome to the show Dennis.
[00:01:46] Thanks much for having me.
[00:01:48] Yes. So Tim is curious that are we really talking to a real person or is this just another expression of the amazing AI that runs extort AI. So Dennis do you somehow have to prove your humanity to Tim apparently.
[00:02:09] I’m always suspicious always think there’s a curse going on. I never know.
[00:02:12] Something fishy is always going on. You’re absolutely right.
[00:02:17] There you go. Well to kick things off I think it might be great for our audience to first understand maybe the real definition of artificial intelligence or at least how you define it and then talk a little bit about how that has applied in terms of the work that you’re doing now.
[00:02:33] I am sure that any definition I come up with are really any definition that the three of us come up with today will at some point in the not too distant future be rendered invalid. And that’s been the case really for the last wots almost six decades. I do like this idea of taking what used to be a human process and see if you can somewhat emulate that’s s and the machine process and I think in the given kind of vertical that we’re attacking here for where we’re trying to emulate if not completely replicate that of the personal assistant way if you ask Kurt to set up a meeting. Our agent can do just that and that’s certainly a definition that continues to be true today. It was also true yesterday and I feel good about it being true tomorrow.
[00:03:28] So now I want to ask a follow up question which is so as I read up on the topic there’s sort of some people who would demarcate sort of a narrow versus broad AI kind of concept would extort AI be considered a narrow AI in that context because it’s focused on a singular task.
[00:03:48] Correct. We’ve suddenly had this fantasy of coming up with some human level AI and that’s not a single Hollywood movie that wouldn’t immediately suggest that it’s just around the corner and I they arrive it will most certainly be evil. However I think those fantasies and those doomsday scenario a little bit further out so we can just kind of pop them over here. But then there’s this idea that there’s plenty of narrow AI for where real applications and real use can be seen through applications today. Whether that is Facebook figuring out that you’re in the picture so you don’t have to tag of yourself or Amy being able to set a meeting for you. So narrow AI indeed.
[00:04:36] So how narrow will it stay for how long how about how do you to the perfect the perfect the narrow and then started to go broader.
[00:04:44] It’s a yeah it’s a good question right. I certainly don’t believe that there’ll be some morning in 2025 for where you receive a press release from Google or Facebook or some other company that will say today we’ve done it. And that’s the end. It’s the last piece of software that will ever develop because we now have human level AI I just can’t create that senile or create some sort of belief where I can see us get there. However I do see the near term and mid-term have some distinction between what I define as or Sunsilk AI or some sort of enabler and think of that as the Siri Cortana. And Leksa and so on so forth. And I set a vertical ice for where highly specialized jobs that you want done with the are essentially I cannot do will have to connect with. So the future is not one for where I compete with Apple and certainly not really one from where. Apple will consume all of the jobs that we need done more likely rhyme with you saying Hey Siri Could you have Amy at X today I reached out to my friend Tommy and set up a meeting for when he’s in Manhattan first week of September please. And he then becomes the horse sunnily ice job to figure out who is Amy and figured out what did I just collect handed over to that agent. See that agent do that job. Sometimes either immediately sometimes over days and upon conclusion come back and say Hey Siri that job you asked me to do. I’ve done it. You can now go back to whoever asked you that question.
[00:06:36] And I think we’ve almost kind of seen this before. If you ask me just like. If they have the capacity to create this Oracle they can do all the jobs we want done or have all the answers to all the questions that we might have. That seems almost to rhyme with the idea that there shouldn’t be any app store that why didn’t they just do all the apps themselves which seems ludicrous when you say it out loud. And not only would they never have the resources to do so. They wouldn’t even have the imagination to know what apps do a 43 year old Danias American with two teenage daughters need. They have no idea what I’m looking for in that app store. I think it would be the same with these eyes that there needs to be some eijun marketplace or intelligence marketplace. Way you go hunt for intelligence that you need to do what you want done.
[00:07:37] So that that seems like an incredibly pragmatic and believable and plausible way to approach it.
[00:07:42] So I didn’t know if we were going to go it seems like you were part of the Hollywood’s one example but you also look at you know Silicon Valley and maybe that’s maybe it’s because you’re you’re grounded you’re a grounded east coaster you know talk about that we are kind of on the verge of the singularity and what you’re describing is much more of a if you have these smart components so they’re not. I say do X and it does x. I say I need to get to Y and you figure out how to get me there. And like you know Google Maps or whatever being the example. So if you have those kind of targeted I guess kind of more on the narrower front. And then if you have a large enough catalog of those then then you have I guess they are the horizontal AI is is just kind of a connectors into that. So stuff that I could never give two shits about is that my horizontally is never going to go and use those specialized components.
[00:08:43] I think you’re right. And I think where it becomes really sexy where things start to rhyme with whatever fantasy we have for AI is when the sawhorse until agents start to talk to each other. So right now if you think about any piece of software I think there’s two changes about to happen. So the first one is that most software today whether you’re an accounting and analysts a programmer a designer software will assist you in doing a job a little bit faster a little bit more accurate but you are certainly the one doing the job. It’s not like you can go back and say I had a piece of software do the job for me. No you did it. But hopefully the application make sure you can do it much much faster. I think that is about to change the way you will start to describe objectives of what you want it done and then the software will do that for you such as. In. The most simple of ways. Amy would you be so kind and set up a meeting between Denny’s and I when I’m in downtown Manhattan. Second week of August. That is not you doing the job that is you describing the job which you want done. And I think it will be plenty of software. They’ll move from this one paradigm to the next and it will start with little chores like setting up meetings and so on and so forth. But then it would come almost all of us in any piece of software for way. No. I have the question.
[00:10:21] That’s the hotpot I really just wanted to figure out whether the margin on our customers on these given campaigns especially on the East Coast this is what we do in Europe. And that’s the question. Somehow I think there’s a disconnect. So you go do that for me. Whatever analyst agent I might have. And once you’re done it come back and tell me. So that’s the first kind of major shift I see happening in software. The second major shift I see happening is that almost all into action with software is a very syntax driven environment even applications if they show the way you use buttons and drop downs and this and that’s however good you are you x it might be it is still since Syntex driven environment or even when to machine speak to each other it’s kind of an API driven environment and that puts the burden on whoever Scott the question is if you want something done you need to teach yourself the UI that you saw the syntax. I think that needs to flip away the burden should be on the receiver as in if I ask a human some something. One of my employees could you go do this for me. The burden is on him to understand what I just asked him to do and if he doesn’t understand that he needs to tell me but otherwise I’ll just walk away and assume he’ll do that job. Why is it not the same with software success. If I ask Expedia. Could you be so kind and find a trip to Miami this weekend.
[00:11:58] For me my wife and my daughter us less than a thousand dollars and need to be back one day by 10:00 a.m. for meetings. That’s my request. You go figure out how you want to interpret that. I don’t care whether you use Muppet’s in outer space. People in Manila machine learning just do what I told you to do. And I think those two combined that you describe objectives and the burden is on the receiver that will allow us to see these agents kind of come full circle you start to talk to each other. So if I ask Amy to set up a meeting with one of my potential partners in Miami come next week then she’ll have the whole meeting set up or somehow things change things get delayed. Then on Musse rather that Amy reached out to my travel agent make sure that the whole thing rebooked my travel agent reached out to my receipt agent that keeps all my receipts. So 150 dó rebooking charges. And I would much rather that my receipt agencies out to my CPA and have the whole thing stored and I want them to be able to talk to each other. But none of them should think about ooh we need like 17 different API eyes. Now just tell Julie the travel agent and Julie could be human. She could be a machine. Doesn’t matter. The burden is on her. So the old saw suddenly the near term changes I see. And then you can see it becomes super fucking sexy when all these agents start to talk to each other.
[00:13:38] You don’t even know why you staying an extra day in Miami but somehow these agents agreed that it was better that you stayed the night and you just assume that I guess I’m staying home by night. And here we are.
[00:13:52] Well it seems like one of the travel was an interesting one because I’ve I’ve grappled with you know when you go to travel you don’t realize how many factors you’re actually dealing with when you think that you’re.
[00:14:04] You just need to get to Miami and you need to be back by this amount of time and you have a price point but then it turns out that well you really don’t want to fly through O’Hare and at 30 you’ve got all these other other criteria for the agent too. I mean where does kind of the learning preferences over time. Because part of it seems like it’s natural language processing part of it seems like it’s having a very robust ruleset that can kind of kind of grow kind of organically and then as there is there a learning part of over time I’m going to get better at booking you these sorts of things.
[00:14:37] Is that a fall following one of those two.
[00:14:40] So travel is a funny one right away we’ve gone backwards so we used to have travel agents the way Querrey I just described would be the exact Querrey I would walk into my travel agent and give him and he would somehow have some relationship with me for a way that wouldn’t be crazy talk and he would work out a few options and I would pick one of those options. But then we were fooled into believing that. No you don’t need that. What you need is to sit alone in the dark at 1:00 a.m. on all bets Expedia AnCo for 30 a.m. and sold this yourself. Oh yeah I guess so. No it’s just fucking retarded. But somehow we got Leard into it and now for 15 years Slate the three of us have been travel agents and none of us wanted to be travel agents. I just want to be in Miami so somehow we need to figure out how to bring this back. And I think what you described and certainly what we’ve figured out is that for any one of these agents to survive we need to have them exist within the universe that is so well-defined that when I say a scheduling agent you and I think the same. Otherwise my universe is different to yours and Amy can exist in these two different universes. What makes travel agents at least in my opinion quite high is that when I say travel agent my view of that is different to yours. There’s not an immediate overlap. We have a clear understanding that I asked Amy to set up a meeting. The input is clear.
[00:16:18] Her negotiation is clear. The output which is named by it is clear and that half suddenly this end up being possible a travel agent might actually be 7 different distinct agents that we need to solve first before we have this more casual agent that’s I just described before so I think this other agents before the travel agents bots look no further than to the inbox.
[00:16:47] I think anything arriving in that inbox have already been digitized by the pure fact that is already in natural language in your inbox and completely ripe for potential automation and thus for an agent to take over.
[00:17:02] I like this vision of the future. This sounds exciting it sounds realistic. I mean that’s. Well I mean it’s already here for scheduling meetings right. So we’re but it’s not.
[00:17:13] It’s not Wattson right I mean that’s the thing is that what you’re describing is I would say almost almost exactly the opposite of how IBM presents Watson in the media as look at any of these brilliant minds. They can sit down and have this conversation and just throw data at it no will solve it. So that’s kind of like the machine learning of just throw data at the machine and it’s going to figure out it’s going to figure out what your inputs are what your outputs are it’s going to figure out where your deepening variables are where your independent variables are what the relationship is between them and that is kind of this amorphous broad thing where what you’re describing is saying no. Break it down into the components that can be repeated. And it turns out that’s challenging. If you bake those really well then you could start hooking them together right.
[00:18:01] I also have the same fantasy. I also want the same as they want some Oracle for where whatever pain I might have or whatever. Choi arrived in my inbox. I can just give to some machine and you do it so I can just run around the office with a diet coke. Work on the white but think grand thoughts. But I just don’t see any immediate pointers just being able to deliver on that promise. So the only way we’re going to get from here to today I think is if we break it down and to give you a good example here. So we are 70 people who spent almost three days in the basement trying to solve this single agent which can only schedule meetings. Does that really well though. But that’s the only thing she can do. Imagine then of all the things you want done just multiply that with 70 and 35 million dollars. By the way which we raised and then that receipt the end of that dinner that will be pretty darn expensive and I’m not even sure it becomes possible because that’s just some steps away. You can go from A to D. You need to go through these steps to kind of get there. And I think suddenly without this turning into some sort of therapy for me and it’s not even into me anymore there’s just Danny’s crying on Skype and we do that once a week.
[00:19:31] But I think one thing that people underestimate here is the dataset required so many of those stories or neural nets that you have seen people playing around with will mimic something so you’ll see any half decent paper mimic a conversation on some topic. The interesting thing about that is that the key word here is mimic as in the conversation is actually not between the human and the machine. The conversation is in your head as you can see what the machine said where the human said where the machine set and somehow you look at that and conclude that that’s a conversation. Notice it’s a conversation. If the machine understands what we talk about not as if it comes up with a string that somehow relates to the spring before it and you believe it’s a conversation. And I think that’s certainly where we cannot cheat as sin. If you ask Amy to set up a meeting between the three of us in Atlanta at a given conference for an hour at that hotel in that lobby she cannot not understand that she cannot write something back that looks like we are having a conversation. She needs to write something back away. Everything you said I understood as I understand the people the locations the dates the times the constraints your preferences all past and all of that I can look back and say given that my response will be this not a set of strings. No. Given that understanding my response will be this and that a sudden the way in all honesty we’ve underestimated the amount of effort we needed to apply to do two things to bring Amy’s life one define this universe in which she exists and it sounds almost too grand like Dennis don’t use the word universe. When you talk about meetings you some of them were just not that.
[00:21:40] But that universe however simplistic it might sound us and then just let me describe it for you here. It’s called location participant date and time. What else do you need. I think I just find it for you. But we probably spent a year and a half two years alone just describing it. Forget about 1 machine learning algorithm over another one neural net over another just defining this universe. Because if you don’t have that define how can you predict any outcomes within this universe. If you don’t know what the boundaries are. That was hard then the second major challenge is again completely underestimated by us included. Here is the data collection Chalons. It’s almost like the three of us figured out that we want to build a tunnel from Manhattan to New Jersey and all we talk about is the tunnel and how great it is going to be Trump style. And that sounds like a good idea. But the tunnel is not the job. The drill is the job. Oh we should be talking about is what drill do we need to get from here today and all the work that we need to do to actually do the tunnel is in assembling the drill and once the drill will reach the middle. Those two drills will get disassembled and probably never used again with the drills with the jaw. And that analogy comes back to. All of the work that we’ve done have been on the data annotation Kunzel the annotation guidelines not on picking out features in some machine learning model to increase precision or recall but annotation like. Just grunt work is so unsexy that.
[00:23:32] There’s no word for it but that’s what. That’s what’s needed.
[00:23:37] Well I mean it’s funny you’re you’re talking about defining the universe than the having a common common definitions and kind of being a digital analytics podcast and thinking of how often we run into you know on a Web site they get to crunch the data and give us the insights and make the recommendations and we stumble across that we we haven’t really gotten crystal clear on what successes. I mean that’s I mean I think that the easiest way for web optimization is optimized to conversion on the Web site and you know AB testing multivariate testing bandit problem whatever crank it through and as you’re talking I’m kind of hearing the same thing you still have to define what that problem is and you’re still splitting the. Yeah there’s a bunch of stuff you can throw at it. The tools to come to crank it through but that may be a falsely trying to draw draw a connection between not scheduling meetings and optimizing a Web site.
[00:24:38] Not really. So we come from the same place and there’s very few that will collect even good clickstream data. Even though we spent the last decade and a half fight to persuade people to do just good not even great just good but just getting to good is hot as in really high. I don’t think I ever looked at.
[00:25:07] Just the most rudimentary clickstream data sets thinking Damn that is stellar. You almost always just a little bit disappointed but I guess we’ll make the best of it. That’s your starting point. And I think that is suddenly that’s data sets and then the next question is then you tell me what you want to achieve here.
[00:25:30] And that could be multiple objectives sometimes even competing objectives. But if you don’t tell me what you want to achieve how do I help you get there with.
[00:25:41] All poor data sets. So it’s not even like I’m going to be selling there. So you don’t know where we are headed. And the boat is kind of leaking. So I guess we’re heading to sea now. Yeah. Okay. So I could certainly relate. Ed have. Nightmares from the past that are very similar to the current White House. And I think we’re you know as you’re talking about it it does express why it’s been so difficult to do things at our machine learning in analytics and things like that because the universe is not defined well enough and one universe for one company may be completely different than another and we make all sorts of.
[00:26:23] Assumptions. And obviously I have in my little venture here the challenge away I have to insist in a place for way. I can’t allow time. For me to educate the U.S.. You will look at Aimi make a set of assumptions for what she’s supposed to do and if it’s not crystal clear a few just fail. We just happened and I’m heavily biased here to stumble upon something for way we believe certainly there’s a good agreement amongst most people on what an agent that can set up meetings is supposed to do and we can see that in the intense that we try and predict. That’s a finite list. It’s not a never ending list of things people want Amy to do and that was really to last. We made a bet and the company said that listen up and finite. We would have died trying Assane we could not. Have. Some set of. Infinite amount of intense S.N.. What is it you want her to do. And really her skillset right and that needed to end and perhaps year year and a half into its we saw kind of getting to the end of that and that was obviously a ride towards. You know you keep paddling your further to sea hoping that you know there will be something beautiful out here and we suddenly got to the end of it. But that was a little bit of a bet. I think many other people who star on other agents will also be sailing into the sunset but they’ll be there’ll be no islands.
[00:28:05] When they get there they’ll just keep sailing and there’ll be an infinite amount of intense.
[00:28:09] So when you say intense or is that I mean is that a list of you have a list of 10 things or 20 things or 30 things or are two things is that is that how you knew you were there when when we all week went by and there was nothing else that wasn’t fit within that list. Is that how it worked.
[00:28:26] Yes just a longer list and a longer time period. So it seems like one of those things that you and I can quickly start to imagine those intense new meeting council meeting meets get the meeting running late. Adding participant Mikoto is optional change conference number and so on and so forth lights up those are all logical buts is when you get to the end of that list where things might not be logical and when you start to see there might be some overlap here that can’t be any overlap in those intense. There needs to be unique and mutually exclusive so that I can predict one from the other and that have been a long journey. Even just so forget about the intent prediction. There’s three primary entities which we need to deal with. So to pool data location and people and if you think about to pull data just time in Dennis don’t tell me you spent more than 20 minutes defining time. I’ll tell it to you. August 1st 2016 two 100 hours. Thirty nine minutes. Daylight Saving. But you know what. They’ll be cool if everybody talks like that but they don’t and they’ll say things like Let’s do one one what. First of all this 1:00 p.m. 1:00 a.m. You’re on the west coast talking to me talking about your own time zone. Or even worse later. Let’s meet up later this week. Today. Your next. Or even worse. Let’s meet up upon my return. Way at myco. When are you coming back. And so am I traveling. Exactly.
[00:30:30] And somehow you need a time model for where you are trying to communicate time. And I still need to understand what you’re trying to communicate and somehow be able to restore that so that have taken a lot of time. No pun intended.
[00:30:48] This is awesome.
[00:30:49] I’m really enjoying this conversation. One thing that we’ve got to do at the beginning that we should probably go back and do now is maybe Dennis just talk a little bit about your journey from analytics into this field and sort of kind of what steps are what key points kind of can you point to that kind of got you to move to where you’re doing this today.
[00:31:11] So here’s the story that I’ve come up with. The funny thing is that I believe it but you are allowed to call bullshit come the end of it. And it goes like this which is that back in the mid 90s we did its enterprise web analytics consulting Afaf mostly raw files or rudimentary Web trends.
[00:31:37] Version one point oh by the way and did really well and the way we look and that was really a venture for where we would go find the data take it back analyze it come up with some sort of insights go out on a monthly bi monthly quarterly basis and provide that insight to management teams. That was my first venture. The next one was one for web that seems heavily delayed. How about we create a platform for way instead of me providing the insights I create some software away we will still collect store the data but allow you to retrieve and slice and dice it in a traditional enterprise web analytics SAS piece of software and suddenly a next step in that journey at VR we took it one step further.
[00:32:34] So instead of me analyzing and coming up with the insight instead of me giving it to user you come up with the insight. How about to create a recommendation system so that not only do I collect store retrieve. I will model on the data and tell you what I suggest you should do. And now is the next natural step if you think about that AI that’s kind of a fourth step on that journey for where I will collect store. The model suggests and take the action. So instead of me suggesting you meet up next Thursday at 4:00. How about I just have my agent figured that out foil and do it. So that’s my 20 year justification for how I got from one place to the next and it might just be all based on looking back at this.
[00:33:20] I can see that you could you could sell that you saw you had that vision from the from the get go you were you knew or ahead you just the world couldn’t couldn’t move quite as quickly as as you had a plan. I’d buy it.
[00:33:32] Only a computer could be that like hate it.
[00:33:35] Now we’re back to the Turing Test. We need that little.
[00:33:42] Boy comfort test. You need. So the questions that you should have prepared for me.
[00:33:47] Yes that’s right Wolf. But it is interesting because you know in the world of digital analytics there are lots of people who are trying to do things around you know discovery of insights or hey let’s do some things to try to push some of the things you should be looking at to you programmatically. And there’s there’s sort of this balancing act in these more I would say this is a more complex system than the one that you’ve chosen to tackle. Hence you know it’s called The Plant times 70 times 70 with the other millions and millions of dollars right that it would take to evolve it. Is it something where we actually should be thinking more like human with intuition and sends out agents on specific tasks and those agents should just know how to go pursue a very specific set of tasks within the data that the user then takes in is my intuitive understanding of what we need to accomplish and is that a better pursuit in the short term Russia we just keep on trucking. I
[00:34:49] think so because what you describe in the most extreme sense is really just a company away.
[00:34:59] One KPI let’s call it profit and everything else is optimized before that and no human need apply saying I just want max profit. You can do campaigns you can not do campaigns. You can allocate more finely. You can nauts you can hire people you cannot. You can do whatever you want as long as I’m maximize on this KPI I think we are far away from that.
[00:35:27] What the thing is much more likely is this second Sanaya which you describe which says what I tend to see as the bring your own agent generation away used to be certainly the three of us young kids had a Bring Your Own Device generation away. We could bring our own laptop bring our own smartphone. Stachel access company data through those devices. Now I think we’ve just reached a point for way that’s just normal. Nobody will bitch about you bringing your own mobile phone or even reading your company e-mail on your mobile device. I think the next step is bring your own agents S.N.. When I hire Tim to do a job or come work together with me for the next two years we are hiring. By the way a customer acquisition analyst you should go apply on to the next point and say Hi it’s him. Who am I hiring here and hiring Tim. I’m hiring Tim who. Have employed 11 distinct agents that he worked with for years. Machine agents and he trained those agents. So you think they are super optimal not just in the universe and last but in their relationship with them and that I find interesting. I find that super interesting that I don’t want just Tim alone Tim coming alone would seem handicapped sane. That’s all. I’m not sure we’re ready for that. I would issue some campaign optimisation agent for content that you’ve been working with for years that’s you go deploy if there’s any questions in that domain you obviously wouldn’t do that yourself.
[00:37:22] Or are you telling me that you do keyword optimization or keyword analysis and spreadsheet. Just sell him. That’s a little bit eccentric. We probably need some agent to help us with that. So I think that is certainly what’s going to happen near term and then is enjoying who is paying those agents. So if I pay Tim his six figure salary is he paying his own agents because I kind of hired him under the eyes of those aHUS agents because they keep those agents as he walks out the door. So he needs to pay them. So I probably need to somehow see some settling away. I need to pay for this little army of agents that he bring along. And that’s I think this is interesting and realistic by the way.
[00:38:16] If we look at in an individual at our careers and for the for those of us who are in consulting absolutely we’re supposed to show up when we’re working with clients and saying I’m not starting off saying well let me. When we learn how to use a spreadsheet let me learn how to use your your web analytics platform. There’s kind of an expectation that you’ve got experience so that’s kind of the that’s maybe the more and more office you’re more efficient because you’ve done this but it does seem like there’s the potential to say whether you’re a consultant or inside a company is you’re growing your career. What are those conceptual agents the techniques that you’ve codified process. Be it a tool you’ve developed. There’s certainly a wealth of technology vendors out there that are saying they’ve built the magic agents and you can just plug in their software which doesn’t happen you still need to have a human being who can operate those. But some kind of maybe extrapolating a little too far but thinking that that’s certainly what I feel like a lot of what I’m doing right now is trying to figure out what can I do. What can I do. What what machine little mini machine. Can I build that I can apply in multiple places and then grow and develop those those agents. And that’s kind of a technical agent but I guess there’s a process piece too. I think I’m on the same page as you are.
[00:39:35] Well the first the first thing that has to happen is a pattern or system has to be developed as the right or best way to approach that problem like again in software development. There’s design patterns.
[00:39:48] And I think if we look at the jaw of a junior analyst even that is way too sophisticated to dream up some agent they can do the job even of a junior analyst. I do think though if you look back. At the first six months of this year there’s many tasks which you’ve done over and over again and some of those tasks as you repeatedly do them might actually be up for automation and some agent could do that for you. And I think the interface to that agent will not be a drop down menu or some other interface that will be national language where you will describe to that agent what you want done on whatever given data set and that agent will do that for you. And the more you work with that agent the more that agent will learn both your language your objectives and what you want done and the more you train it the smarter it becomes. Even one agent from the same company versus the same agent from the same company with another person yours might just be better trained because you honed a skill set over the last two decades. So it’s not about necessarily just the vendors and their agents. I think it’s about the people who deploy the agents and how they train them.
[00:41:11] So is there but if you’re bringing your own agents is there the there’s there the potential the challenge you were talking earlier about needing to know the syntax that if I’m using my own agent and the think I’ve certainly run into all sorts of people who’ve kind of built their own their own tools and they talk about how this thing they’ve built is super smart super super flexible and in that they’re kind of illumines who can interface with it which is still fine if I’m asking them to do a task and they become the agent to get to the agent to do that. But is that does that short circuit design the ability to short circuit half of the equation that you were describing.
[00:41:50] I don’t think so. So are the agent you go employ will be an extension of.
[00:41:57] Tim and expecting to be able to communicate with that agent. I do think though that the agents will communicate in plain English. I think you will learn your preferences. But you’ll still be able to communicate with other similar agents or whatever agent universe exists within your world the way it needs to also communicate with these other five that you have employed and he should be aware of the fact that they exist and things sometimes will happen because they had a conversation not behind your back but in front of you because that would just be natural. Why wouldn’t they. And that’s I think Sunday is that exciting step but I so want to lower our ambition and not have us believe that there’s some junior analyst coming along. I think they’ll be some highly specialized agent for one type of campaigns in one specific setting for a given dataset that somebody might come up with. And instead of spending two hours every Thursday on that task you have an agent and that agent can do would you ask you to do on that dataset in that given Nyau that will be the very first beginning.
[00:43:15] So don’t they just productize it and then go and sell it to the masses and then they will because whoever has the best agent who should be able to sell that everyone so we can get the best they’ll put it in the App Store. We’re back. We’re back to where we started. And horizontal vertical. Many many vertical eyes. So that raises another question of which horizontal is the best right now.
[00:43:41] See that’s a good question and there’s obviously multiple answers to that. Which one of these horizontal agents have the most nahl it’s s and if I just shout out a random question will it have the answer to it. But I think let’s just go a little bit further up and I can tell you what you have turned me on with this. The form factor of the Amazon Echo. Away. You have a bigger speaker better microphones Freule always on. It’s a game changer. I don’t think they have the best nahl it’s the best and LP and the best kind of concept of the universe at last. But that always on BET on microphones have changed everything. If I look at how much I’ve talked to Alexa over the last couple of months it has surpassed how much I’ve talked to Siri for eight years. Right.
[00:44:45] Right. Or how much my kids have talked to Alexa. No I think that’s right. It’s not the amount of knowledge in the horizontal.
[00:44:52] It’s more how vertical can it be and how fast and you can see that they have a very similar attitude towards tomorrow as I have. And they’ve done created that skill store. I believe they call it which is their vertical eyes. So they just assume that no we won’t have all the answers. No we won’t be able to do all the jobs. So how about people who just go engineer skills which can tap in to our ecosystem.
[00:45:23] So when can I expect Amy to be a skill and the Amazon Echo in the not too distant future.
[00:45:32] The. Yes and I think you’ll see us do to begin with. It’s actually just you being able to get status here looks at what is Amy working on so she can speak back. Hey I’m still trying to reach Tim. I’m on the second reminder. He’s a little bit spotty but hey you know Tim this was just an example by the way.
[00:45:55] So apparently you looked looked up our records before you got into this.
[00:46:01] Usually the other way around.
[00:46:02] But now that’s that’s amazing and that is sort of you know just in the past year I think Amazon has done something that Google and Apple who are ahead have failed to do a little bit more.
[00:46:20] But then again it’s about to happen to it’s going to go yet. Google gets it. Google just announced earlier this year their device. But in hindsight it’s just so obvious and given that it’s so obvious why didn’t Michael Tim or Dennis come up with the moniker.
[00:46:41] Home device. We will be killing it right now. Have we done so well somehow it was only obvious the day to arrive that you really just needed Semien as speaker. That said a really good microphone’s and plug it into the wall. So is always on.
[00:46:59] And I was I was too busy cleaning up my crappy clickstream data so I might get there. My excuse is only going out with the people with good ideas. I don’t personally know.
[00:47:13] But yeah. And it helps. Probably that you know that’s where you buy everything from.
[00:47:17] It was just a good idea.
[00:47:19] Such a good idea that it’s almost disappointing to see that my new Apple TV isn’t always on with Siri because that is the one thing for way of crying out loud. I’m staring into it as we speak and it doesn’t talk back to me.
[00:47:36] It’s like they you could just add two microphones and everything would have been hunky dory but they didn’t all go to Amazon while spotted jotting down things I want to go invent because my head is exploding right now and I can’t do them but my agent and that’s pretty exciting and for someone who has never had like I’m not a data scientist like some other people on my team like Tim is learning or like. Those are the skills I don’t possess. But what if I had agents who I could deploy with the intuition I have and then make someone like me much more able to go and do great analysis but without necessarily have to learn all the ins and outs of how to utilize those systems.
[00:48:27] And that’s the key word I think you just describe what I believe will be one of the most valuable skills in the future. So it’s not the one who have the best answer answers will be cheap. It’s the one where the best question. So ANSYS I can just get an agent to go work on that. So they are monicker chip chip no questions. That is gold and that will see I think the dynamic in the workplace changed dramatically. Asain right now we reward people with answers and we should because that’s the environment we live in. But I think we don’t fully realize how valuable the questions are.
[00:49:15] I feel like I’ve been beating the drum of. Because right now I think a lot of times analysts are expected to provide they’re asked to provide answers in the absence of a question. Now give me insights and recommendations or to your other early example when I say well what let’s talk about what you’re really trying to achieve. I just want to maximize profit. I’m like well that’s great. But you’re selling toilet paper so and you’re asking me about social media. So we’re going to have to get to questions that are a little more focused that rely on your intuition. So but like this is really helping me bridge the gap just in web analytics trying to get people to ask questions generate hypotheses you know because you look at a question is I have a hypothesis. I have a question I can state that is I. I believe the world operates this way whether I really do or not it doesn’t matter. And I want to go validated. And to me that’s gold for an analyst because the analyst just then has a very specific thing they can operate on. I like this actually helped me bridge the machine learning to the softer world of discovering hypotheses so nice group.
[00:50:22] Yeah I had no idea where the conversation would go and this has been awesome.
[00:50:29] This has been a great discussion and I’d love for us to continue but we do get to start wrapping up. One thing we do on this show is we have to go around and do a last call which is anything you think is interesting. Right now you’ve run across in the last few weeks or days. So anybody thought out there anybody can start.
[00:50:48] I’ll throw one of them calling an audible during the episode based on the discussion I had a couple of podcasts I was going to recommend and instead I’m a rep recommend one specific episode of one specific podcast which is the Planet Money episode number 6 26 which is this is the end. It was from May of 2015 but they went around to economists there’s this whole there’s this idea that is the machines get smarter and all these agents are out doing stuff we’ll have more leisure time.
[00:51:19] And what does that mean. What’s the impact on the world if we actually achieve this point where we don’t need to work 60 hours a week. And I remember it being very fascinating is that one arguing how close are we to that which we touched on that a little bit of how close are we to this Nirvana of the machines just doing everything.
[00:51:40] But then also the economic impact of it. So it just seemed so yes to this episode so it was episode 6 26 of Planet Money.
[00:51:48] I will in the same vein if not clarity but perhaps further disconnect with where we believed we were because there is this blog post that I keep rereading every two months. You know that’s one of those where you fall in love. Then you have to read it again. And I keep coming back and it’s the one by Tim over at. Wait But Why. The one on the Fermi paradox. And whenever we think you’re in control and you have some understanding of the real universe and the whole thing almost makes sense. You read that blog post and nothing makes sense. Whatever guideline for tomorrow. I might have had. I have nothing I guess I’ll just go to sleep and when I wake up I’ll be much best. But I love that blog post because I can’t fathom. Why we’re LOL why they’re not others and why haven’t they talked to us yet. And that sounds like ludicrous. Then you read the blog post thinking they should have called this. Why didn’t they call us and then when you finally get to cannot believe that. Ah so there’s this great filter concept. That’s why they haven’t call us. But then we’re doomed. Ah so we do have them call us if they do call us they’re going to be so far ahead that we’re probably also doomed. And then you go to sleep again. Just one of those every two months. You should go read it. It’s a wonderful blog post 15 minutes.
[00:53:32] Where fucks the Great Filter is ahead of us. Oh this is going to be fun and awesome.
[00:53:39] All right so acto my last call is not that dissimilar in some senses.
[00:53:46] Although it’s kind of out there recently a lot of people have been recommending to me a certain show on television called Rick and Morty and so I’ve begun to watch it and it’s very interesting. So that will be my last call is that Rick and Morty which also deals with time travel and multidimensional universes. So not that dissimilar. Apparently the science is pretty solid so I hear I’m like few episodes in so I’ll keep learning. All right well this has been phenomenal. What a great discussion. You know as we’ve been talking you heard Dennis say the question is the thing so if you have questions hit us up on our Facebook page. We will relay them to Dennis and maybe if he’s not too busy working on what’s going on over at XTANDI. He can jump in and help us answer them or hit us up on the measure slack. We always love to hear from everybody. Also if you haven’t yet and you have a mind to feel free to rate us on iTunes. Apparently that helps us out in some way or develop an agent to go rate us on iTunes. We should work. We’re going to work on that as. Right now I think we’re the number one explicit digital analytics podcast on iTunes and we’re hoping to stay that way. But anyway it’s been such a profound delight Dennis to have you on the show and this conversation has been one that’s been sort of mind blowing a little bit and delightful in that it gives you so many great avenues to go down.
[00:55:21] So I’m excited to see some of the conversations that come after that. So thanks again Dennis. It’s been so delightful to have you on the podcast. Really enjoyed your insights. Thank you so much for joining us.
[00:55:36] Thank you very much for having me. This was absolutely fun.
[00:55:41] Michael Tim Gere’s and we’d love to hear from you as well. But as you know and Tilders agents to do it for you. Keep analyzing.
[00:55:55] Thanks for listening. And don’t forget to join the conversation on Twitter or measures like grit. We welcome your comments and questions. Facebook dot com slash analytics or analytics on Twitter. Smart guys I want to play a little.
[00:56:15] Bit. Yep yep we’re recording calling him. Tim. Gets me every time I’m already recording and I see something really secret. Jim will ask a lot of questions and I will try to get a word in. Oh I’m a user. Absolutely. Very good. You’re going to peel your face off here at the end of the. That’s right. Technically now that you’re in the world of a captive set your was more of a Bond villain. Because I won that money. So while you like on television television right. I mean it’s pre-recorded. Don’t want my camera.
[00:57:13] Rock again artificial intelligence.
Subscribe: Google Podcasts | RSS
This site uses Akismet to reduce spam. Learn how your comment data is processed.
https://media.blubrry.com/the_digital_analytics_power/traffic.libsyn.com/analyticshour/APH_-_Episode_220_-_Product_Management_for_Data_Products_and_Data_Platforms_with_Austin_Byrne.mp3Podcast: Download | EmbedSubscribe: Google Podcasts | RSSTweetShareShareEmail0 Shares
Subscribe: Google Podcasts | RSS
[…] #044: Artificial Intelligence with Dennis Mortensen […]