Skip to main content
Search form
  • 00:05

    [Developing New Methodology to Measure Technology's Impacton the Future of Work]

  • 00:10

    MORGAN FRANK: My name is Morgan Frank,and I'm a PhD student at the MIT Media Lab.[Morgan Frank, Graduate Student, MIT Media Lab]I am part of the Scalable Cooperation research group,which is Professor Iyad Rahwan.This Scalable Cooperation group is interestedin society and governance, and howthings like technological change reshape that.So with artificial intelligence getting more advanced,

  • 00:30

    MORGAN FRANK [continued]: and with more real world applicationsfor this type of technology, like self-driving cars,and the implications of robotics and manufacturing,we're finding that daily life is changing as a result.And governance, and regulations, and politics are allsusceptible to change as a result of this.We want to understand this a bit better.And our approach for doing this is to leverage

  • 00:51

    MORGAN FRANK [continued]: computational social science.So rather than being the guys whoare developing these algorithms, wewant to do a more meta analysis and understandwhat the implications are for this typeof technological change.[How did you get interested in this field?]My interests are more focused on the future of work.

  • 01:12

    MORGAN FRANK [continued]: There's a lot of different things going on in this group,but my job is to try to understandthe implications for technological changeon the future of work.And I got interested in this because Ifelt like traditional economic work--they have a very defined toolset,and it's not really embracing the complex systems approach.And this is my sort of academic upbringing.

  • 01:34

    MORGAN FRANK [continued]: So I'm always tempted to find new areasto apply this way of thinking.And if there's new information to be gained because of that,it seems like a good area to explore.In particular, the way I got introducedto the future of work was, we weretrying to understand the impact of automationin different cities.And actually, it's sort of a romantic story.

  • 01:56

    MORGAN FRANK [continued]: I was driving with some collaboratorsin between Abu Dhabi and Dubai in the Middle East.And these are both two huge cities with nothing in between,just desert.So we were wondering, are the peoplewho are living in these rural areasmore or less susceptible to the negative impactof technological change by comparison to people

  • 02:17

    MORGAN FRANK [continued]: who are working in the cities?And we weren't really sure what the answer should be,right because technological change reallyhappens in cities.Cities are innovation hubs, and hubs for economic productivity,and tech startups happen in cities.So maybe just by proximity, the people in citiesare more susceptible.On the other hand, the people who are living in rural areas

  • 02:38

    MORGAN FRANK [continued]: tend to rely more on physical work and physical labor,and rely less on social and cognitive labor.And traditionally, social and cognitive laborare thought to be more resilient to technological change--less automatable, and these skillsare associated with higher wages.So our early work was to try to estimate that-- this impacton different cities.

  • 02:59

    MORGAN FRANK [continued]: And what we found for the US is that small cities facegreater impact from automation.And the task became to understand why.What's leading to this differential impactacross cities of different sizes?And indeed, it has to do with cognitive and physical skills.But we've found that when we focus on more specific skills--so if we consider workplace tasks and skills with increased

  • 03:21

    MORGAN FRANK [continued]: resolution, we actually see an improved story.We're able to predict important aggregate labor trendsmore accurately-- things like how people are movingbetween jobs on their career path,and also how people are migrating between cities.And this is crucial, since we knowemployment opportunities are the leading factor in people's

  • 03:41

    MORGAN FRANK [continued]: decision to relocate.Another new thing that we can do,because we're focused on improving our methodologyfor studying workplace skills as a complex systemis, we can understand how the system adapts to perturbations.This is sort of a really common idea in physics,but it's kind of absent from general equilibrium models

  • 04:01

    MORGAN FRANK [continued]: that economists are used to.So in particular, if a specific new technology,like computer vision, becomes very matureand eliminates demand for a worker whocan perform a specific visual task,then, if we have this high resolution model,we can understand how the decrease in demand for workerswith that specific skill--

  • 04:24

    MORGAN FRANK [continued]: how that decrease percolates throughout the system,and changes both the nature of different job titlesin terms of their skill requirements,and also changes the nature of a workforce in a particular city.And of course, these changes can lead to migration patterns,changes in employment opportunities, and ultimately,impacts the health and well-being of different areas.

  • 04:45

    MORGAN FRANK [continued]: [What research methods were used in your research project?]When we're trying to decide how to approach these problems,there's a lot of outside interest in automationand technological change in AI.And it's becoming flashier and flashier to talk about.So what we need to be careful of when we approach these problems

  • 05:07

    MORGAN FRANK [continued]: is not to give in to the craze, and try to be grounded.And for me, one way we've achievedthat is to rely on very standard data sets, so far.So most of our work has been leveraging datathat anyone could get from the US Department of Laboror from the Census Bureau or from the Bureauof Transportation Statistics.

  • 05:29

    MORGAN FRANK [continued]: And the key has been to use this data in a new way.So it's really a new methodology.It's a new framework for studying labor.And it's not about using flashy new data.And that's been the key so far.Now having said that, of course, when you have new methods thataren't the traditional use case for a data set,you can run into sort of the edges

  • 05:51

    MORGAN FRANK [continued]: and the boundaries of that data set.And indeed, we're finding a bit of that too.So I'm hoping that because of this friction,we can begin a conversation about whatnew data sets might enable further explorationon this pathway.So in terms of statistical foundationsto understand the type of work that I'm doing,

  • 06:11

    MORGAN FRANK [continued]: I really, whenever possible, try to use the simpleststatistical methods possible.If I can get away with Pearson correlation, then that's great.If I need more complicated models,the standard in labor economics isto leverage linear regression, and thencompare different regression models in great detail,

  • 06:32

    MORGAN FRANK [continued]: including very specific things about the coefficientsassociated with different variables.And I find that even this level of specificitycan be rather complicated.So whenever you can boil it down to the cleanest storyand the simplest statistics, then you'redoing a really great job.I also think that this mentality plays into howI try to design figures.

  • 06:52

    MORGAN FRANK [continued]: I want figures to be really clean.And I want to use the appendix and supplementary materialsto provide the robustness checks, and the extra details,and things like regression table analysis.I think that figures and statistics should highlightthe key part of the story you're trying to tellat that stage in your paper.And that should be the only focus.

  • 07:13

    MORGAN FRANK [continued]: [How was the data collected and analysed for your researchproject?]We're using data from the Department of Labor and the USCensus Bureau--really traditional data sets.For example, census demographics in different cities.One thing we're really interested inis how educated is the population in a given city.

  • 07:34

    MORGAN FRANK [continued]: And we're also using, in particular,this rather new data set from the Department of Laborcalled the o net database.And this data set is produced from surveys--annual surveys of workers of different job titles.And they ask them how important really specificdifferent skills or tasks are to the completionof their daily work.

  • 07:55

    MORGAN FRANK [continued]: So this allows kind of an annual censusof demand for different skills.And connecting these things with workforce statistics--things like demographics, and where people are gettingeducated--can allow us to understand how the labor system in the USis adapting and evolving to technological change.[What were the motivations behind using new methodology?]

  • 08:19

    MORGAN FRANK [continued]: Our motivation for this new methodologyactually comes from ecosystems and the study of ecology.In those systems, there's two big piecesthat describe the ecosystem, generally speaking.There's these governing equations thatdetermine population dynamics.So these are things like birth rates and death

  • 08:41

    MORGAN FRANK [continued]: rates for different species in the ecosystem.But there's also interactions between different species.So the classical view from ecologyis to consider ecosystems as a bipartite network, wherethere's competition and mutualism.For example, maybe different speciesof bees and different species of plants that the beesinteract with.And it turns out that you can consider both the governing

  • 09:04

    MORGAN FRANK [continued]: equations for that system and the structureas two different pieces, both of whichallow you to say something about the health and resilienceof the ecosystem.So translating this now to the labor systemin a specific city, maybe we can consider different citiesas an ecosystem.And their workforce and, in particular, different

  • 09:26

    MORGAN FRANK [continued]: occupations and different skills as different speciesof bees and plants.And if this analogy holds, then wecan use the structure that we're identifyingwith our complex systems approachand our improved resolution on workplace skillsto say something about the resilienceof different urban workforces based on the structure.

  • 09:47

    MORGAN FRANK [continued]: And we can do this without focusingon these general equilibrium modelsthat labor economists have used traditionally.So what I mean by this structure is perhaps,based on the different skills workershave, if demand for those skills changes,how suited our workers to adapt to a different skillset thatis perhaps nearby, but they're currently not leveraging.

  • 10:09

    MORGAN FRANK [continued]: And this type of insight has potentialto inform models for job polarization,for worker mobility, and for the well-being of different citiesand their workforce.So for example, an example of thismight be people who are good at mathematicscan more likely obtain programming skills.

  • 10:30

    MORGAN FRANK [continued]: And this is because it turns out that math and programmingrequire both cognitive and numerical skills.And we know this from an aggregate viewjust by looking at the co-occurrence of these skillsin this o net database.However there's not a lot of co-occurrencebetween mathematics and certain physical skills.

  • 10:52

    MORGAN FRANK [continued]: And that sort of makes sense, intuitively, too.So this distinction between, perhaps,cognitive skills like math and programming,and physical skills, things like maybestamina and manual dexterity--this distinction between skills is underlying--the distinction between cognitive and physical jobpolarization that's happening in the USand actually happening in cities as well.

  • 11:14

    MORGAN FRANK [continued]: [What skills are required for this kind of research?]The day to day work on this line of researchis a lot of ideation and understandinghow these ideas fit into the canonical work of laboreconomics.Even though I'm not really leveraging general equilibriummodels and the traditional approaches

  • 11:36

    MORGAN FRANK [continued]: that labor economists use, the ideais to augment this toolset with new ideas, and a new approach,and a new way of viewing data.So I have to spend a lot of time interactingwith traditionalists in this field.And I've been very fortunate to work with people at MIT whocome from a variety of backgrounds,including Business School economists,

  • 11:58

    MORGAN FRANK [continued]: economic department economists-- and I understand--I'm told there's a distinction between the two--and also working with physicists and people whostudy urban science.So I think that the day to day workcomes down to boiling down these ideasfrom these different flavors of research,and understanding, even though they feel like they disagree,

  • 12:19

    MORGAN FRANK [continued]: and they're perhaps siloed in different areas,can we find intersections between their ideas?And does that allow us to see something new?So once we identify some new ideas,the key is to understand how to work with the datato test that idea, and make sure that you're doing a reallyfair test, and you're controlling for thingsthat the classical theory says that you should control for.

  • 12:42

    MORGAN FRANK [continued]: And for me, this has been a lot of data science work.So it looks like a lot of me sitting at my computerwriting Python and making pretty plots.And that that's been a lot of it, actually.[What tools and resources are helpful for students lookingto use this method?]I like to use existing packages in Python whenever

  • 13:04

    MORGAN FRANK [continued]: I can for this type of work.For example, I leverage pandas a lot, and numpy, and matplotlib.Those are great packages.I've also recently been using igraph, which is an alternativeto networkx, because it's a bit faster,and I think that the algorithms built into the packagefor plotting are a bit more useful.

  • 13:26

    MORGAN FRANK [continued]: But even though I like using these packages,I find myself wrapping them a lot in my own codethat allows me to use specific functionalities in a waythat I find more intuitive.So for people who are familiar with Python and matplotlib,they will probably agree that matplotlibcan be rather complicated.

  • 13:47

    MORGAN FRANK [continued]: So masking those complications with my own personal codefor use cases that I run into a lot is a really useful thing.And it makes my workflow smoother in the future.So what I find when I'm running into a situation

  • 14:07

    MORGAN FRANK [continued]: where I am deciding between--OK, maybe I should wrap this code up nicelyand spend some extra time building this customtool that I know I'm going to use, as opposed to copyingand pasting code all the time or constantlytrying to finnick with matplotlib to get somethingto look just right.It's definitely worth it to take a pause from the research

  • 14:30

    MORGAN FRANK [continued]: and build this tool for my personal use.I find that that's extremely useful,and it's something I would recommend to everyone who'sdoing a lot of programming.[How do you manage large amounts of data?]Data cleaning is essential.And I think you'll find that a lot of datascientists and computational social scientistsmight joke that data cleaning is like 60% or 80% of the work,

  • 14:54

    MORGAN FRANK [continued]: and the science is actually a small part of it.And I have to say that working with census data and datafrom, for example,which is an official portal for getting accessto official government data like Department of Labor data.It's super cumbersome to clean the dataand to get exactly the data you're looking for.

  • 15:16

    MORGAN FRANK [continued]: So in my experience, this has been a really expensive partof it.But it's absolutely worth it.And getting it done and being able to relate thingsacross different data collection effortshelps you to identify trends that are really true.If you see the same pattern of human behavioracross different surveys, for example,that's stronger and stronger evidence

  • 15:38

    MORGAN FRANK [continued]: that the human behavior is really a pervasive thing.[Were there any challenges and how did you overcome them?]My work on the future of work has had some big road blocks.And the biggest roadblock has beentrying to work with traditionalists in this area,and to develop a shared language with them so that they

  • 16:01

    MORGAN FRANK [continued]: can understand how I'm thinking about the problem,and I can understand how they're thinking about the problem.And in the case of my background, whichis complex systems and applied math, and tryingto interface with traditional labor economists,there's so much divide.Basically, we both know math, and that'sthe only thing in common.

  • 16:21

    MORGAN FRANK [continued]: In terms of how we write papers, for example,it's drastically different.Econ papers tend to be very long,have very long lit reviews, and all the figuresare in black and white.And again, there's this real focus on regression models,in my experience.On the other hand, if we look at the papers comingfrom the complex systems community,there's very short lit reviews.

  • 16:42

    MORGAN FRANK [continued]: Papers are very short--usually seven pages.Figures are very colorful.And you avoid tables at any cost, pretty much.So this has been the hardest part,but it's also been the most rewarding part.So I've been fortunate to receive an NSFgrant with my advisor to organize a workshopwhere we got to bring together people

  • 17:03

    MORGAN FRANK [continued]: from different disciplines-- complex systems peopleand people studying urban science, but also BusinessSchool professors and people working in econ departmentsfrom around the country.And we actually got everyone to sit in a room for a few days,and play nice, and share their ideas,and understand where everyone agreed new questions stilllive.So how do we refocus this shared effort

  • 17:25

    MORGAN FRANK [continued]: on answering those questions?And this has opened the door for some morecollaboration across these different disciplines.So in working with these already existing data sets,the challenge is to be creative.Because these data sets are open to everyoneand have existed for a very long time.

  • 17:45

    MORGAN FRANK [continued]: So I'm thinking of census data, for example.The challenge is to provide novelty,and to discover something new.So some strategies I've found to help with thatare to work with a rich variety of different peopleto come up with genuine new ideasthat you might be able to just use this data to answer.

  • 18:07

    MORGAN FRANK [continued]: Another thing is to combine different data setsin ways that haven't been combined before.So an example from my work is, I'mvery interested in understanding worker migration--the flow of people between different citiesdepending on their skillsets.And I think that technological change changesdemand for specific skills.And this perturbation to the labor system

  • 18:27

    MORGAN FRANK [continued]: can lead to aggregate trends like worker migration.So if we want to study this, we needto take both migration data from the census and skills datafrom the Department of Labor, and also, I'vebeen interested in flight data from the Bureauof Transportation Statistics, and combine all these thingsto make sure that there's a cohesive story across all

  • 18:48

    MORGAN FRANK [continued]: these different data collection efforts.And when we can do that, it suggeststhat something really true is happening,and we're detecting it.Another challenge I've faced with this line of workhas been to share information and our insightin a way that crosses norms in different fields.

  • 19:08

    MORGAN FRANK [continued]: So even though my academic upbringing is in applied math,and I work with a lot of physicists and complex systemscientists, we often get people from geography and peoplefrom labor economics as reviewerswhen we try to share this work in academic journals.And to this end, I think that the role of editorsin helping connect our style of storytelling

  • 19:31

    MORGAN FRANK [continued]: with the norms of these other fieldsthat the reviewers come from has beenreally essential in our ability to share our work.So for example, we often get reviews like,where are the regression tables?Why are there no regression tables in the main text?And this has to do with the style of information thatis portrayed in these general science journals,

  • 19:52

    MORGAN FRANK [continued]: for instance like Science Advances and PNAS.I think that it's essential for editorsto foster interdisciplinary research, and to not give into reviews which are sort of very shallow--things like where are the regression tables?I think that the role of the scientist on the other hand,

  • 20:13

    MORGAN FRANK [continued]: is to provide every possible viewonto this insight and the supposed resultthat they want to share.But it's not that every viewpointneeds to be in the main text.And this is a good use for the supplementary materialsin the appendix.So what I'm saying is, if I thinkthat I can tell a story using a figure,

  • 20:34

    MORGAN FRANK [continued]: then I'm going to do that, because thoseare the norms I'm used to.But if I can tell the same story using a regression tableor using another traditional tool for analysisthat I think reviewers will like,then I want to provide that in the supplementary as well.[What recommendations would you have for students looking to dosimilar research?]I would recommend to scholars who are entering

  • 20:57

    MORGAN FRANK [continued]: computational social science--I would advise them to embrace the interdisciplinary natureof the work.And by, that what I mean specificallyis, make connections to traditionalistsin these different fields.I think that it's tempting to thinkabout computational social scienceas this invasive species.We invade sociology, we invade economics,

  • 21:18

    MORGAN FRANK [continued]: we invade business schools.But that's not the right mentality.It's not that we're competing.We need to be thinking like we're augmentingthese different disciplines.We're living in a society now wherethere's more data than ever.And of course we need new methodologies, new approachesto go with that data.That data enables a new way to look at the world.

  • 21:39

    MORGAN FRANK [continued]: So these traditional disciplines--they need to be augmented by this approach as well.So what that means for new studentsis, at every possible opportunity,you need to make connections with traditionaliststo understand how they think about things traditionally,to not repeat work, to not come up with your own terms,for example, that they already have jargon for,

  • 22:00

    MORGAN FRANK [continued]: and to understand what the big open questions arefrom the traditionalist point of view.[What is next for your research?]What's next for our research on the future of workhas been to see what we can recoverfrom traditional models using this new framework with reallyspecific workplace skills, considering the whole system

  • 22:23

    MORGAN FRANK [continued]: as a complex system.So the first step is always to see what you can recoverfrom the traditional work.And in our case, we are not reallyin disagreement with traditional insightsinto labor econ and technological change,but what we're seeing is some new insightsthat were absent from the traditional approaches.So this is kind of step two.

  • 22:43

    MORGAN FRANK [continued]: What is given that we can recover?What already exists?Can we see something new?For labor economics and technological change,one big example is the interactionbetween bank tellers and automated teller machines.I think most people would expect that employmentfor bank tellers decreased with increased use of ATMs,but actually, the opposite happened.

  • 23:05

    MORGAN FRANK [continued]: Actually, national employment for bank tellersincreased proportionately to the increased adoption of ATMsnationwide.And there's two reasons for this.The first is pretty well understood by labor economists,I think.And it's this thing called demand elasticity.Basically, when technology makes a certain type of laborcheaper, sometimes the demand for that good or service

  • 23:26

    MORGAN FRANK [continued]: goes up non-linearly with a decrease in price.And so for bank tellers, this meant that more bank branchesopened up nationwide.The other reason is that the fundamental skillrequirements for this job title of bank teller--it changed from requiring clerical work--people who can handle money--to people who leverage social skills.So for example, the modern bank teller sort of

  • 23:48

    MORGAN FRANK [continued]: works as a customer service representative,or a sales representative for bank goods and services.So there's not really a lot of insightout there about this fundamental shift in skill requirements.And this is something we're trying to build towards.So this is an example of some new insights, and perhapsa new theory that we can try to developas we try to come up with a model for occupational skill

  • 24:11

    MORGAN FRANK [continued]: redefinition.[Further Reading][Alabdulkaeem, A., Frank, M. R., Sun, L., AlShelbi, B., Hidalgo,C. & Rahwan, I. (2018).Unpacking the polarization of workplace skills.Science Advances, 4(7), eaao6030][Frank, M.R., Sun, L., Cebrian, M., Youn, H. J., & Rahwan, I.(2018) Small cities face greater impact from automation.Journal of the Royal Society Interface, 15(139), 20170946]


Morgan Frank, PhD candidate at MIT Media Lab, discusses his research on the implications of technological change on the future of work, including research methodology; data collection and analysis; use of new methodology; skills, tools and resources to manage big data; recommendations; and future research.

Looks like you do not have access to this content.

Developing New Methodology to Measure Technology's Impact on the Future of Work

Morgan Frank, PhD candidate at MIT Media Lab, discusses his research on the implications of technological change on the future of work, including research methodology; data collection and analysis; use of new methodology; skills, tools and resources to manage big data; recommendations; and future research.

Copy and paste the following HTML into your website