David Yakobovitch
Beth Partridge
Today’s guest speaker shares her wealth of knowledge and expertise in the talent race for enterprise AI and strategy, listening as Beth Partridge at Milk+Honey, and I explore topics, including what cultural barriers must be overcome to build a data-driven culture, whether the three competence centers, that link teams together and how enterprises can build effective data science and AI research teams. This is HumAIn.
Welcome to HumAIn. My name is David Yakobovitch and I will be your host throughout this series, together we will explore AI through fireside conversations with industry experts, from business executives and AI researchers, to leaders who advanced AI for all HumAIn is the channel to release new AI products, to learn about industry trends and to bridge the gap between humans and machines in the fourth industrial revolution.
If you like this episode, remember to subscribe and leave a review.
David Yakobovitch
Welcome back everyone to the HumAIn Podcast, bridging the gap of humans and machines in the fourth industrial revolution. This week, I have a special guest with us. This is Beth Partridge. She is the CEO, chief data scientist at Milk+Honey, she’s coming to us all the way from Silicon Valley and is working in the industry very special near and dear to my heart, which is not only about employing data scientists, but discovering whether the best skills and the best way to hire data scientists as well.
Beth, thanks for joining us.
Beth Partridge
Thank you for having me.
David Yakobovitch
This is so fun and I everyday teach and everyday build curriculum and everything, interview candidates myself, and the industry is changing so fast. There’s so much misinformation and so much going on. What have you been seeing going on at Milk+Honey? And tell us a little bit about what you guys do.
Beth Partridge
So our charter is similar to yours actually, in that we’re trying to help bridge the gap between business and data science for the rest of the world. The non data native companies, everybody else. And we’re seeing exactly that there’s confusion starting with job titles, too. How to organize teams too, really what data science means in terms of organizational structure requirements and cultural change requirements.
And it’s still, we’ve been at this for a couple of years now, and it’s still just a giant mess. I really every day, there’s innocent, but misinformation, in the press constantly, and it’s especially hit hard for recruiters because it’s technical, but it’s not technical in the same way that other technical jobs are where they could pick up names of some tools or, and categorize them because there’s just no definition for data scientist yet.
And back from the HBR article, when was it? Five years ago it was declared that data scientist is the sexiest new job of the Century. A lot of people that aren’t really data scientists call themselves data scientists, and it’s just adding to the confusion.
David Yakobovitch
It’s so amazing. I got my start working in Fortran and C#and COBOL and all these, more classic languages, what Excel analysts and actuarial analysts and all them with work in. And it’s like the whole industry has merged. If you’re a business intelligence, if your data, if you’re visualization, whatever, it’s just all jumbled into data science.
Beth Partridge
It is. In fact, we had to actually create our own internal, very detailed profiling tool in order for me to personally not have to interview every single candidate. So I created a tool where we ask a bunch of questions and ask them to put in some details about their projects. And we actually kind of have to cross-reference between their tool set that they’re familiar with and the roles that they say they do on their projects and the whole package in order to really figure out who’s who, and then even once you figure out who’s who.
The other side of the equation is that our clients aren’t familiar with how to form teams with all of the right skills. So you’ll get teams that hire a PhD data scientist, thinking that that’s all they need. And then the PhD data scientist ends up without the right support. No, in data engineering support or, no data analyst support.
And it’s just this complete lack of understanding about who’s going to do what. You can have the best data scientists in the whole planet and then the most committed C-suite. Putting we’re willing to put whatever resources they have into making this transition to adopting enterprise AI. And if you don’t have somebody in the middle, then it’s still not going to work and that’s what’s happening over and over again. There’s, it’s just, there’s a, with that what’s starting to happen now is that there’s like a catalyst sort of person missing in the middle that can connect the dots and make it happen.
David Yakobovitch
At Galvanize we call this catalyst person, this may or may not be the right name, but we call it the data translator. Someone who, like they work with the business, they work with the data, they piece it together. And, I don’t know if that is the right title, but it’s in the right direction. Because when you think of PhD data scientists, this person is someone who’s really smart. They’ve been spending years of research. They’ve been studying, they’ve published, but that doesn’t mean they’re a pro at R Python, maybe they only did a couple things.
And as you just mentioned that sometimes sets them up to fail because they do need someone who’s been a hardcore Oracle DBA or data engineer for many years off the cloud or on the cloud. Where someone who’s been doing reports and analytics for quite a while. And so what you’re mentioning is all about community of practice and building up these best protocols for data science organizations. Most companies don’t even have data science teams still.
Beth Partridge
Some many have tried, most are trying like at a project level, which is a whole another topic about. But you really can’t approach data science by starting it in a corner and then having it grow, it takes cross-functional teams and commitment from the top and all that, the cultural stuff. But that the piece that’s even more critically missing is even classes in expertise in application of data science is the big missing gap.
It’s the people that can take, like my favorite example is like a machine learning model to minimize downtime, like for factory equipment, one of these preventative maintenance models. It’s like you have a factory manager who would do anything to this comes from my background. I didn’t tell you nowhere. The fab goes down and it’s, I don’t know, millions of dollars a minute or whatever. Every minute, every hour counts. So the factory managers are willing to do whatever they have to do to minimize how often their machines go down for preventative maintenance. So there’s a lot of ways to solve that.
Now they do a really conservative estimate, whatever the manufacturer of the equipment recommends to make sure that it doesn’t go down because downtime is the worst, but there’s many different ways you can solve that problem. You could do a recommender that says what’s the percentage that the next machine is going to fail. Or you could do a classifier that says it’s good or bad, or there’s a bunch of different ways you can approach it.
And probably in the beginning that you only have so many options, but depending on how much data you have, but you don’t have to wait until you have this perfect model that can predict how many hours till the machines, till each machine fails to use it because you’re starting from this really conservative worst case. So if there’s somebody that has enough confidence and understanding of the business and confidence in the models themselves, then you can do a plan where, as the model is maturing and developing, you could start allowing it to say, tweak the preventative maintenance plus or minus.
So you can wait a little longer and, over time you can, as you get more data, the right data, you can move to a different kind of model and the confidence is constantly growing, but what’s happening is that, there’s just not that bridge in between. So somebody says, they’ve heard of GE or, one of the well-known prevented maintenance model examples, and which is this beautiful model that says exactly how many hours until the engine fails or whatever.
And so they want that and they wait for that where, and it just doesn’t happen. They get frustrated. They don’t understand that machine learning projects are more like R&D projects. You can’t really plan on a development time. And so things are just dying on the lab floor and it’s really that, it takes somebody there’s so much mistrust, also kind of from the press and just because there’s so much missing. Even the basics of the terminology, AI versus machine learning versus big data is confusing.
And there’s just so much lack of understanding that it’s almost causing the stalemate, and the, what did you call it? What was the name for that job?
David Yakobovitch
The data translator.
Beth Partridge
So we have been calling it internally and we were trying to be really careful not to invent even more confusing job titles, so we don’t use them outside, but we call it a data strategist.
And so it’s somebody that understands the business, but then understands machine learning enough to understand the different types of approaches and what it means in terms of risk and accuracy. And, you have to really have somebody that has that crossover point, but we’ve encountered people that were looking to fill this role and the job titles that they posted or anything from technical product manager to, I can’t even remember the others, but they don’t know how to describe the role that they’re looking for. So they weren’t getting any response at all. That’s how big the disconnect is right now.
David Yakobovitch
And we’ve been seeing some other titles coming on for data scientists, like in the past couple of years, customer data scientist and market data scientist, and even solution engineers. So some of them are trying to get there. And they’re like this hybrid of technical and business. And you’re exactly right in what you mentioned, Beth, that R&D projects are getting stuck in the lab and dying on the floor, whether it’s consumer facing with students who I’ve worked with or enterprises I’ve seen dozens and dozens of projects like yourself.
And so many of them are just technical, but there’s no business case. And I always tell our clients like you have to operationalize your results and be able to translate or speak to what does this percent of accuracy mean? What does this false positive for the manufacturing defects or predictive maintenance, like you mentioned, result in. And perhaps people not be able to strategize that has been causing a mistrust in the process.
Beth Partridge
It is, there’s just so much and it’s mistrust going both directions. We did it a piano with both old and new. We did a couple of different panels of data scientists to find out what they’re experiencing in the job market.
And a lot of the younger ones said that they go into some of the big fan companies and the interview for a supposedly data scientist role. And then they get hired and they’re not doing data science, they’re going, they’re doing reporting or whatever. And then on the other side, you start to see articles from disgruntled and frustrated, and appropriately companies that say, we hired the best data scientist and we put a bunch of money into it and got nothing. It’s really.
David Yakobovitch
And we’re the Kaggle grand master, they were number one at Leetcode. But they can’t do data science.
Beth Partridge
It’s so frustrating. And then, there’s an opportunity to hear that, when I get some spare time, I still spend some time with Berkeley. That’s where I got my masters. And, I was speaking with the career center, the person that runs the career center for the med students there. And she came to me actually saying that the data science programs, now they try to solicit not just CS or computer science, undergrads, they’re trying to get a good cross section since data science is so, requires a domain experience, but the ones that come into the program without having a CS or computer science background, they get through the program and they understand machine learning, but then they don’t feel confident enough to come out and get to try and look for machine learning jobs or even data engineering jobs. Because that’s not their background. So they’re going to the career center saying, what should I do?
And here we are out here saying, that’s, what we really need is these people that understand the business and understand machine learning enough to draw those, to make the connections and to really be that catalyst.
So I really do think that there’s an opportunity to bridge this gap. We’ve just got to let it evolve and just kind of name it. Actually, it would be great. And then we need to create coursework in serious applications of machine learning and business.
There really aren’t any, and lately there’s all kinds of how to get started in AI classes popping up everywhere. All the continuing education and stuff. And very often they go through like one example of each type or they go through some of the more common applications, which is great. It’s a great place to start, but there’s really that there’s no class where anybody could sit down and say, here’s a business problem. What are the different kinds of ways that you could approach it? What kinds of models could you do? And, how could you evolve it and how can you do a deployment plan and get it integrated into the business processes in a way that you don’t have to wait?
David Yakobovitch
That’s so key what you mentioned, it’s like, how do you get to model production? How do you get to model persistence? How do you get to these deployment plans? But to get there, there’s so much required before that. And funny enough, I’ve actually recently featured on the HumAIn Podcast one of the professors from the mids program, Alberto Todeschini. So he’s been on the previous episode and we’ve talked about similar points to what you just mentioned, Beth, about the business case, he’ll often get a lot of these executive students, who are mid-level managers at companies who may not have that full technical background. And they want to have more of it so that they can hire data scientists and scale out an organization. But it’s like, how much of that do you get not only in the program, but then to be able to implement it.
And that is a missing piece. And I’ve noticed that the industry is continuing to fragment in certain regards. We now have this, this new term in the past few years called MLOps. There’s a lot of MLOps conferences coming out about Kubernetes and Docker and replicated and deployments like Terraform and Ansible and a lot of that. So there’s so many new segments of the data science and machine learning, which is so fascinating.
Beth Partridge
The emergence of that segment that you’re talking about now, and it starting to the term data engineering is starting to stick for that, which is great because that’s now starting to happen. And there’s been a lot of press about four to one data engineers to machine learning engineers. So that one’s on its way, but this one that we’re talking about, the more the catalyst role of applied data science is still missing. It hasn’t really been broadly recognized and we need to find a way to describe what it is and label it and start getting some coursework along those lines. Now it really is at its heart, its applications of machine learning and business period. There’s not even really a framework for it.
David Yakobovitch
Isn’t that what students do in undergrad and MBA programs. They go through rotations when they get hired by companies and they do these case studies where you might work in four divisions of the company for a year on these rotations in business.
And that’s the traditional Excel person. But how about for new companies, when you get hired, do rotations as this apply data scientists to uncover the insights and what’s going on in the organization, maybe that’s it. You mentioned earlier about the PhDs being set up to fail and everyone’s rushing in. Because of the hype, we got to be part of data science, or we got to be part of AI. But it is rushing worth or will that cause maybe the fourth AI winter.
Beth Partridge
I actually don’t think there’s too much momentum now for that to happen. And there are resources now. It’s so ironic that like, when I’m my I’m one of the guys that I graduated with from the mids program, which was, at the time, one of the top. Three or four or five or whatever. And he had 10 years of experience as a data engineer at a major company. He could not get a job coming out, he just could not get a job. And we’re seeing that now where many people that are hiring specifically say, I’m sorry, we’re not taking entry level.
And that’s not just, there’s some debate about the certification programs and the bootcamp programs and how effective those are. But, this is even coming out of the accredited university programs that they’re just because of burn, they’re feeling burned and in a way it’s appropriate because you really do need to have some understanding of business in order to effectively do it. And so the, again, it comes down to that gap.
David Yakobovitch
I agree. I sit down on a lot of RFPs and proposals for cities and governments and, training and re-skilling workforces. And one of them was with the heads of engineering of Instagram at sea and charitable just a few months back. And we were in that same exact conversation. They were saying, if we’re going to hire anyone for data, we ideally are going to look at maybe PhDs or computer science grads, but they’re going to get that preference over the bootcamps. Sure, you got great programs. I can talk about Galvanize, Flat Iron, Medis, all day, every day. But if you’re coming from a completely non-technical background and then doing the three month bootcamp to be a lead data scientist, I don’t know. I remember when I was graduating business, my degree in undergrad and doing Excel. And I joined the actuarial department making $11 an hour as an analyst doing very advanced work.
Beth Partridge
But it’s confusing. Is it even more that there’s actually, we’ve built out a framework to describe the applications of AI for business, and there’s really two major buckets. One is, truly automation applications that are more related to NLP and computer vision and those kinds of true automation, for minimizing head count or whatever.
And then there’s analytics, the automation models you really don’t need to be. You don’t need to have as much data science experience because for the most part, you’re just adapting the models that the cloud vendors. Offer, you’re adding some transfer, learning, layers, to a model. And a lot of them are accessible just by API calls.
And then there’s the traditional question of make versus buy. There’s a lot of software as a service companies out there, especially like marketing, especially we’re seeing a lot of them because, tends to be compatible from one company to another, but there is room for people that have that data science experience. But unless you have somebody that’s doing the strategic plan that understands that there’s those different levels of expertise required and doing like a strategic plan that takes into account those make versus buy decisions, then, you can’t take advantage of it.
Another way is under utilized is that 80% of building a machine learning model is data wrangling. It’s just getting it healthy and figuring it out and moving it around. And there’s such an opportunity to bring in young data scientists to assist with, doing that data wrangling. So it could be stretching our machine learning resources further while simultaneously training these younger data scientists with the kinds of practical experience that they need.
It just comes down to just a basic. Just a lack of understanding, still just kind of industry-wide about how it works and how that can happen, but I really have hope that, like you said, that data scientists or the data engineer, people, the Ansible and that the, just the recognition that there’s that role that’s happening now pretty quickly. And I’m hoping that this data strategist or data, whatever, we ended up calling them that data application of data science, I’m hoping that that will be next.
David Yakobovitch
Of course ML ops is happening so fast just because the cloud is giving more mature, but on the applied data scientists, we’re moving in that direction. I’ve coined previously on HumAIn Podcast, a few terms, the word data science is a service, which we’re seeing, platforms now like Weights and Biases and Neptune and H2L and Spell and DataRobot and a lot of these companies. Where you’re as you mentioned it’s make versus buy. So maybe you buy the solution or you integrate there. That might be something that’s going to be in the emerging trend in the remainder of 2019 and 2020.
Beth Partridge
I hope so. That a lot of the auto ML tools that the cloud vendors are bringing out, that they’re ending up getting a bad name because you don’t really end up getting fully completed models from them. But boy, they’re awesome tools for doing, like, as we go set out to build a strategic plan with a client, we first look at all the opportunities for us from a business perspective where their biggest ROIs. And then we go through and look at what data is available, and talk about what kinds of models would be best.
But with these ML productivity tools, like the data robot, you can do a really easy, quick and dirty feasibility analysis, just, throw your data in and it runs it against or like every single Kaggle model or whatever, and you don’t get anything close to a finished model, but you get a head start at figuring out how to approach it algorithmically and you get a good idea of what the baseline is. Which ones that you’re already reasonably close to, that you could focus on first. So there’s definitely a place for them. And it would be again, the tool for this role, a really good tool for this role that we’re talking about of the person who’s doing the strategic planning and identifying the solutions and building out the plan and making them make buying decisions and all that kind of stuff. It’s really valuable for that.
David Yakobovitch
And, with the industry keep evolving, we’ve seen the 2012 to 2015 rise of big data. That’s funny 2015, the 2018 rise of data science. Now the rise of AI, and, as you keep mentioning Beth, the skills of people are not necessarily matching the roles and you’ve built out this profile to better understand people. If you’re a company today and you want to do hiring, what are some tips and tricks of the trade that you can recommend to companies looking to build out AI for enterprise.
Beth Partridge
The first thing to look at is, and the first thing to address is the cultural barriers, the corporate wide things that have to happen. Like there’s, you really have to have a data-driven culture that is really absorbed in their bones and you have to have a C-suite that’s fully committed to riding out the wave and the problems and all of the retraining and all the stuff that it takes to actually, make the adoption, get fully adopt enterprise AI.
And then the next step after that. And frankly, I don’t think anybody’s figured out the, or there is no one answer probably to this is how do you organizationally bring in AI? Do you put it in your IT department that generally doesn’t work great or do you bring it into marketing? If that’s the first model you’re going to do and that doesn’t work so well either.
So, we’re finding that the most successful models are to create some sort of a, almost a competency center that links everybody together.
So that’s the second thing. So first check the for cultural holders, second figure how you’re going to implement it. And then, probably the most important piece is really sit down and understand what resources are necessary for it, data science team to be successful because there’s really three elements. There has to be the business domain expertise, the machine learning expertise and the data engineering expertise.
And those are generally, three different specialties. You can find some data scientists that can do all three, but in general, there are three really different focus areas. And then even within those three focus areas, the folks that do the strategic planning stuff tend to be more business focused and more strategic. And then the folks that are doing more of the implementation stuff, it could be the younger ones that are actually building it out. So there’s actually kind of six roles and it’s really important to understand.
Not, you don’t even have to look at it from a role perspective, but it’s important to understand what elements are required on a data science team to be successful. And making sure that you can put a name in each of those boxes, do a little bit of homework upfront and really make sure that you understand what the new data science team looks like.
David Yakobovitch
And the recurring theme I’m hearing as you’re sharing about these competency centers is that data engineers check it’s the new MLOps, machine learning, it’s the dev, it’s the data scientist, it’s the new AI researcher and so forth check. But the business domain, who’s doing that, is it the business analysts, the data strategists, the data translator. And that seems to be this consistent missing piece, which might be why a lot of AI is getting left in the lab.
Beth Partridge
We try to stay away from that. Trying to label them for exactly the reason that we’re talking about from company to company. There’s such vastly different definitions that we have. We use a matrix where it says that actually we’ve broken it out into six phases of development.
And then those three separate work phases. And it actually just says what they need to do. Like this person needs to identify business opportunities with high ROI. And then now this person needs to scope and consider what are the different possible solutions, instead of like saying who they are, we literally say, this is the activity that has to happen and the expertise that they have to have, and it doesn’t matter what they’re called.
And sometimes you can find when we profile data scientists that comes out in this like grid and some of them kind of straddle machine learning and data engineering, some of them straddle machine learning, engineer and business, and then some are more strategic than others. And we actually get, kind of a profile of where the data scientists fit. So it’s not always. It’s not always a one-to-one, different people have different backgrounds and different crossovers. The most important thing is that somebody on the team is able to actually do that function. That’s really the only way to evaluate it without running into all these problems with the job titles.
David Yakobovitch
That’s right. And if you can do all six of those or all three titles, you’re pretty much found their material.
Beth Partridge
No, that would be the unicorn, the famous data science unicorn.
David Yakobovitch
It’s so good. It’s incredible. And, as you mentioned, we’re looking for one of my mentors, Amy Webb, I love all the work with her new book The Big Nine.
And then we talk about the AI industry and there’s overlap 40 major trends this year, and then it continues to evolve. And one of the challenges is it’s moving so fast. In fact, one of the most significant conferences in the AI spaces is NeurIPS, which is for doing a lot of computer vision and NLP and all these papers and research.
And they said for this past conference, it’s sold out in 12 minutes. Everyone wanted to go to the conference. So for the first time in the December, 2019 conference, doing a lottery system, you can’t even pay to go to the conference. It’s a lottery cause so everyone’s in on the ML, AI.
Beth Partridge
There’s other ways that we’re out of sync too. I love to keep on, keep up with all of the neural, the brain kind of research because that deep learning is still just total brute force. But anyways, besides that there’s different industries that are coming up with amazing approaches that aren’t even getting shared industry to industry like legal is doing some amazing stuff with NLP and NLG, doing automatic drafting of briefs and, doing some really innovative stuff with projecting which arguments are going to be the best and everything.
And then there’s other industries that are getting really good at like the marketing side. But it’s just, even at that level, it’s not getting shared, there’s so much happening at once. I try to keep up and I just, I still feel like I can’t read half of the things that I want to, it’s really crazy.
David Yakobovitch
And there’s so many papers coming out. I’ve read that between last year and this year more than 80% of all the David’s science and AI papers all came out in the past two years. So it’s such a scale, but that does beg a question, whether you’re a professional, who’s retraining, maybe going to midsor a program to pick up data science or you’re a fresh college grad or computer scientists going to a bootcamp.
And you’re trying to get into data science. So much going on right now. What should you do? how should you make sure you could be employable because there is that risk of the skill trap. If you keep studying, keep learning, keep picking up things.
Beth Partridge
Especially with the fear of hiring the newbies. So what I tell people that are trying to get into it right now is get the education, get the training, get solid on, at least your machine learning basics, and then find a job at a company that’s next to data science, where maybe they do the data engineering, or maybe they do the MLOps part if they come from a DevOps background or whatever, but I recommend that they get in someplace next to it and then so that they can then move into it once they’re in, because it’s just nearly impossible they get straight in they have to find an opportunity where, or maybe it’s somebody that has a particularly strong, specific domain expertise or domain background that they could find a special niche.
But in general, you have to kind of find a sideway in right now. And I honestly believe that once we find once we get this resource of the strategy resource, that can kind of make a plan that all these people fit in somewhere, there is a need for all these people right now. The 80% or whatever it is of companies that haven’t successfully adopted yet, they need these folks.
It’s just that we’re stalled because we can’t find a, we just don’t have a framework for plugging the people in where they need to be. It’s going to happen. It’s going to tip at some point this, we’re going to get past this business understanding and people are gonna start to go, okay, I get it and start to ramp up and then there’s not going to be nearly enough I suspect.
David Yakobovitch
There’s going to be a gap. And maybe then, the report about data science being the sexiest job of the Century will come true in that sense. But, several things resonated with me that you just shared. So the data science next to them as data science, adjacent roles. So you’re absolutely right. Come in as a data analyst, but you’re working with ML or data science teams. So then you’re seeing how they do code review, how they productionized their models and over your one to two years in that role, you gain some mentors and experience and then you can laterally make that move. It sounds really effective.
Beth Partridge
In fact, a lot of companies aren’t even calling machine learning engineers, a lot of them like Microsoft tend to call them software engineers, and then they may be add on machine learning, but there’s a lot of crossover, especially at that machine learning space. And in DevOps, actually DevOps to AIOps is, you have to learn all the distributed, but so many companies are moving to cloud anyways for just normal operations, not even the AI distributed stuff. So that’s starting to happen more quickly too.
David Yakobovitch
It’s incredible. And the industry continues to evolve. One thing I’ve worked with a lot of enterprises is I’ve also been working to coin the term, the data science workflow. So it’s important that you have that structure, but exactly like you just mentioned that it’s constantly evolving. Like there’s no perfect workflow. At one point I had 13 steps than 5 steps now, 7 steps. It’s always, always changing, always iterating.
And, we all want to make something that’s agnostic to any industry or to any title, but, it’s a process. And like you mentioned, it’s that strategy, like how can we do talent as a service so that we can effectively ramp up teams and ramp up divisions to succeed as enterprise AI companies?
Beth Partridge
It’s getting there, it’s evolving so quickly. All the pieces are there and we’re just kind of just, it almost sometimes feels like we’re just trying to catch up, and put everything together. That’s already there because it’s going so fast.
David Yakobovitch
And even though it’s moving so fast, I love that we’re always seeing lots of new trends and new reports out there on the skills to be most relevant for the job of tomorrow. And so if we were to look out at data science and data engineers, or that’s probably enough to look at. But if we look at that over the next two, three years. Are there any signals that you see that those who are studying should be like, let’s pick up this language or this toolkit, or this is what we should focus on if it’s just one or two things to hyper-focus.
Beth Partridge
Definitely Python is the machine learning language of choice for sure. And if somebody just wants to get a baseline that isn’t quite sure where they want to end up then absolutely take machine learning courses. There they offer them through, anything from the free online Khan Academy has some really good foundation classes to Coursera. And there’s a ton of different ways to do that, but getting a baseline and, there’s some basic math first.
If the baseline education didn’t include, especially linear algebra, you need to have in order to really get into some of the machine learning, but any baseline that they can get an understanding of models and modeling and all of the Python library set and all that. Would be, give them the best foundation in order to go into whatever they decide to do with it.
And you really have to have even the data engineers who are doing the Dev AI Ops, as we’re calling it, they need to understand the models too, because productionizing a recommender is completely different from productionizing, some forecasting model. So everybody’s got to have a baseline foundation of how machine learning models work and what the options are, and like the business folks at least need to understand enough to understand, what is the risk really? And what is the bias really so that they can translate that into business risk? So nobody can go wrong with taking basic machine learning courses. In my opinion, it’s not just the next technology. It’s like a whole new generation of analytics really.
I like to compare it to, machine learning is to business, what the microscope was to medicine. We went from looking at symptoms or, kind of guessing at generalizations from looking at a body to actually being able to see individual things that are happening, that’s really what machine learning means for business. It’s no more surveys. It’s no more aggregate statistics, guesses and segments.
Now we look at each individual, we make predictions at that. At the individual level. And it’s a fundamental shift in the way that we use and look at data and the results are amazing that at least that’s pretty consistent in the press that everybody gets that if they have to do it, they have to make the transition and have just stalled at actually implementing it.
David Yakobovitch
And whether you are just getting started or you’re an enterprise company wanting to be AI enabled, it sounds like all use cases are accelerating. We mentioned earlier in the episode about transfer learning and how NLP is coming of age, lots of great papers, a lot of great code bases have been coming up this year. We’re even seeing the works from 2017, 2018 with computer vision now coming to age as well. So those are two of the big use cases that a lot of the companies lean into when they want to be AI first, but it’s more than just the use cases. It’s being able to build an organization.
That’s ready. You’ve said it best earlier several times about that. It’s about culture and it’s all about being data first and thinking about how to constantly innovate. And it’s not a one-year investment, but it’s a consistent process that you gotta be in for the long haul.
Beth Partridge
It is. And ironically, in the same issue of HBR that the sexiest job article was in, there was another article that proclaimed, we were calling it big data at the time, but it proclaimed big data, a management revolution. So we’ve known from way back then. They called it exactly right. We just haven’t quite found a way to translate it into action yet.
David Yakobovitch
But even though we haven’t yet, that’s the key word that we will we’re in that right direction. And, perhaps we’ll continue to be human in the loop enabled, perhaps we’ll be able to repurpose ourselves, whether it’s a future of Elon Musk’s style of having chips in our brain or running new cognitive tasks. That’s still early to tell, but that’s something that I’d love to continue to explore with you and really appreciate you for taking the time to appear today on the HumAIn Podcast.
Beth Partridge
Thank you. Thanks for having me.
David Yakobovitch
Thanks so much. It’s been my pleasure.
Humans, thanks for listening to this episode of HumAIn. My name is David Yakobovitch, and if you like HumAIn, remember to click subscribe on Apple podcasts, Spotify or Luminary. Thanks for tuning in and joining us for our next episode. New releases are every Tuesday.