Events Add an event Speakers Talks Collections
 
Michael Johnson
Professor and Chair, Department of Public Policy and Public Affairs at University of Massachusetts Boston
+ 2 speakers
  • Video
  • Table of contents
  • Video
IRAC Chicago 2020
September 24, 2020, Online, Chicago, IL, USA
IRAC Chicago 2020
Request Q&A
IRAC Chicago 2020
From the conference
IRAC Chicago 2020
Request Q&A
Video
Panel Discussion: Analytics to the Rescue: Ethically
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Add to favorites
8
I like 0
I dislike 0
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
  • Description
  • Transcript
  • Discussion

About the talk

Michael P. Johnson, University of Massachusetts Boston;

Obi Ndu, KPMG; and

Narasimhan (NS) Krishnan, AIMMS

About speakers

Michael Johnson
Professor and Chair, Department of Public Policy and Public Affairs at University of Massachusetts Boston
Obi Nadu
Director Advanced Analytics Modeling at KPMG
Narasimhan Krishnan
North America Customer Management at AIMMS

Dr. Michael P. Johnson is Professor and Chair in the Department of Public Policy and Public Affairs at University of Massachusetts Boston. He is a leading expert in modeling and analysis for public policy and public management and a contributor to research, teaching and public and professional service. Dr. Johnson’s research interests lie primarily in operations research planning models for public-sector facility location and service delivery, with applications primarily to housing and community development. His methods enable public organizations serving disadvantaged and vulnerable populations to develop programs and policies that jointly optimize economic efficiency, beneficial population outcomes and social equity. His work can be characterized as: "community-engaged operations research", "community data analytics" and "urban and community planning and development". His work has appeared in a variety of journals, edited volumes and proceedings. He is lead-co-author of a special issue of European Journal of Operational Research titled "Community Operational Research: Innovations, Internationalization and Agenda-Setting Applications" (Elsevier, August 2018). He is lead author of the book Decision Science for Housing and Community Development: Localized and Evidence‐Based Responses to Distressed Housing and Blighted Communities (John Wiley & Sons, 2016). He is editor of Community-Based Operations Research: Decision Modeling for Local Impact and Diverse Populations

View the profile

Accomplished analytics and engineering professional with demonstrated ability to develop and implement data-driven, decision-making strategies. Experienced in leading data science projects, machine learning model development, and Bayesian statistical analyses; from predictive and descriptive analytics, to decision theory and resource optimization. Proven track record of building and managing skilled teams to implement artificial intelligence solutions.

View the profile

Strategic, outcome-driven IT executive with diverse leadership experience in: - Business and IT Strategy, Digital Strategy, Client Management, Service Delivery, Management Consulting, Program Management, Managed Services, Outsourcing, Right Shoring - Creating / Leading high performance teams spread across multiple cultural, geographical and demographic domains, interfacing at the C-level to enable business strategies - Aligning business and IT strategies in multiple industries and IT domains with a focus on operations (Innovation and R&D, Product Lifecycle Management, Sales & Marketing, Order to Cash, Supply Chain Management, Warehousing and Logistics, Finance). - Data and Analytics driven corporate and market decision making and strategy formulation - M&A industry analysis to develop market segmentation and business strategies - Successful business development (hunting and farming) driven by customer focused win-win solution-selling approach

View the profile
Share

Good guitar panel discussion started. So today are panel inside and rescue, ethically. So as to when we do these are special in these critical times that are a number of aspects that come into play that we'd be ready covers and about and I don't like to put it like, you know, it's different from the way we do and I'll take the routine. I'll text you and let you know regular times like all day today. I don't think and what do you want to do in the panel is too kind of explore some of those aspects of the

trade-off for the speeds of the governance some of the data challenges and some of the ethical considerations. First and foremost, let's welcome our panelists. Department of Public Safety and public affairs the University of massachusetts-boston David Johnson receive as patient or patience research from Northwestern University his research interest lies in the data analytics and insights for housing urban planning and Community Development inclusive policy and

planning interventions for climate education diversity equity and inclusion in orn Optics. Skip methods enable nonprofit in public organizations, especially those serving disadvantaged and vulnerable populations to develop programs and policies Advanced economic efficiency beneficial population outcomes and social equity. And we have dr. Rubina do he's the director of KPMG schedule artificial intelligence analytics and 19 years of experience. He specializes in

of the US government background investigation process. He also serves as an adjunct faculty and John Hopkins Applied Physics laboratory fan NASA, Dr. And we have an indiscretion a helix the customer and account management in North America for a imms experience in operations Consulting supply chain and analytics and is passionate about helping companies make better decisions with prescriptive analysis question. Before we get to the actual panel. I just wanted to quickly go over these structures so

that everyone like in previous years like a palace have had a chicken where we kind of talked about a little C form about the greatest aspects and from there the richest out a few freaky questions. So we're going to kind of our discussion based on these key questions. We were supposed to keep question. We will have some full of questions either for the Margaritas or from the palace themselves to each other and we spent about five minutes on a few questions and some questions and then we move on to the next one and we will we will

give the people. 10 minutes for audience questions and answers. Of course. I can if your presence of the discussions feel free to post them we have like I was collecting these questions and never will be And that's kind of the general structure structure that new photo. so would that make we would start with a very with a brief opening a statement from each of the palace so Stop Johnson. Would you like to go first? Yes, please. Thank you. And thanks to all of you for attending the

session what this is how can I purchase research and analytics make positive contributions to address in the most important social problems of our time while addressing issues of governance and is a public servant organizations Better work on behalf of stakeholders, especially those who are marginalized disadvantaged or discriminated against engagement that is work in Weston for communities and community members in a way that supports code learning for problem formulation solution and implementation

and finally ethics that is making a careful determination of what problems or issues to address that have the potential to increase voice representation and agency by those who have traditionally not had great influence over how important problems are solved. Services provided our products made by Services New initiatives that would represent the interests of sections of societies in for one of these is a belief that all are in analytics should use our know how to increase to pass the impact of resource admission German nonprofits to volunteer Consulting the salt

of problems. This was the origin of diversity equity and inclusion and more recently social and racial Justice in ways that can change how we think about what we do we study this is the basis for the committee also. African Dimas Blossom Carnegie Mellon University in 2007. I brought a passion to collaborate Stanley based organizations, but felt my work wasn't having an impact. I wanted around the time of the Great Recession hit I was able to get funding to work with

any development corporation develops Vision models for foreclosed house and Acquisitions and development. This enabled me to deepen my understanding a 400 or an analytics to be deployed and adapted needs to needs the community's and nonprofits that serve them the special is low and moderate-income communities and communities of color. So I hope that research and service and teaching that reflects a commitment to work with on behalf of groups and communities that are not as often reflected in the examples and applications that we are taught in our courses of study and enable all of us to think

different and do difference in support of high-impact policy and planning interventions. Thank you. Certainly, thank you. And first let me thank the the organisers and off my fellow panelists care for this very very timely topic and and also given the opportunity to participate in and what hopefully he's going to be a very fulfilling conversation. Did reference how relevant this topic is in light of our current times are but one thing that I like to surface during these

conversations is the how dr. Johnson spoke to a lot of the societal and policy aspects of this issue of ethics and bias and in the use of analytics one thing that I'd like to as I mentioned some of this conversation to some of the techniques and some of the methods that may be available to Advanced analytics practitioners need a scientist and bills into a i m l domain how to leverage some of these two Who's and methodologies and constructive speech on implementation to shut the gate in the effects of bias, and I

hate your phone Doctor Johnson and again happy to be on this panel and look forward to a very engaging and type of conversation. Thank you. And that Krishna would you like to go next? Thank you very much. You know, it's a tough act to follow both pumps are Cochran and my fellow panelists. So I'm tired. I'm going to try and do my best year. It's only a very relevant Cambodian topic for the times of b r a background of operations. I am not an academic by any stretch of the imagination the hopefully what I speak about today is going to be followed by what I have

experienced and the insomnia what OB start checking on the house, right? What should we do as one of the teams are going to do with an Analytics? And such were always an opportunity cost for doing something or not doing something or taking the position or not taking a position and my vision is getting without anybody waiting. What is the best possible outcome given the variables in the parameters and the constraints that one is aware of and one structure and qualify. What does a walk more than just the mathematical product

is always a range and we talked about that little bit about the range of resiliency vs. The accuracy. So dealing in supply chain analytics down at the order fulfillment level. You need to be supremely IQ Turn It Down For What in hungry goes to which customers to on which day I'm sorry that I was deciding on investments in a network. You need to be resilient in terms of the decision being good enough to make a strategic decision. And what kind of information is required to make it happen. What

are the parameters that govern how people use Solutions like this? And so I think the end into the ethics of it and like I said being biased towards action of all my alarms of parking will be talked about doing the next hour or so is going to be dictated by colored by one we see some of our customers doing when they use a platform and so would really taken the philosophy of democratizing the colon ability. So it's not about only the mathematical genius music venues really the people that are in the front and center

so weather is the nurse who is dealing with a coback patient and being able to be required to make a nanoleaf. RN Network Executives have to make a decision and use platforms and capabilities. I hopefully will get into something and of course data is ultimately the blood that flows through all the plumbing that we have and we can quantify because it's a big data available today that you have to figure out how to Abrogate. Of a human leg back with the with the structure part of it and I think that's

where we really want to focus in on to make battle capabilities going forward. Thank you so much personal want to talk about challenges encountered either positive because of what is an issue. How do you suggest you best address disturbances? Are you really soft the question that is front-and-center on almost any analytical mind, right, but they're absolutely fundamentally through the one number one. Right and I can always see some smiles and laughter in your nose right to pain is

we in any analytical situation analytics challenge situation. You never know. What kind of data you need and you also don't are you just going to wear what kind of bees are you at? Andreas damage amount that you can bring into a problem can be so your mom that you get you can really get paralyzed by the analysis and you don't want that to happen. That's what I mean. When I say that is not a bad thing. Every analytical solution does not have to be a roll soup a supremely beautiful technology solution. You

can stop at Excel you can get data in there. You can play around and that's what I mean when I said earlier about them up as I don't have to be in the hands off the experts only they can be in the hands of the decision-makers in the operational people that are actually fun time Center and that's what I meant when I said creating something. A nurse who's in the front like a building with a patient and absolutely critical. So think about the problem and

it's just a fan. More time on Devine upfront what is doable? And if you do those things that's almost always a guarantee of success sooner rather than later. I'll keep it at that for now and then we can finally go into more details as the conversation goes along. I just want to point out that somebody in there is a chance of putting the gay men, but he said you can provide a lot of inside a question right away, especially when we're building tomorrow for like

liquidy the data and the landscapers like changing weddy. Weddy fast flicks. Are there some best practices are some guidelines that you can share on how to kind of you know, we we are constantly learning from data. We're trying to build models and we tried to protect rights and we can do that have any fast-paced. So what would you have some the guidelines or suggestions on that? You know how to do that effectively. Recommend to write so the more you parameterize the more

you can let decision-making and configuration of the decision-making happen at whether the weather does operation needs to happen. Right? That's number one number two, is it still affect development? So the more you can design and the more you can take the technology element out of the equation of foster, you can get the people that know the data that are in the operations. Bring that experience of both into solutioning and cell which means you get to a solution that

is quicker you get to a solution that's more on text you and you get to a fine tuning the wind healing process is extremely rapid which means you get from design or talk to execution literally a matter of weeks rather than months in years. We took we ate a little bit of our own dog food earlier this year when we created a framework for anybody in the industry, not just our customers but anybody to use and we actually have a lot of our customers were all too many

prospects using a cold 1912 planner if you well and so we can allow me to open a platform to allow that kind of Rapid decision-making. We allow a lot of people in the trees that are now all of a sudden with how to react well to keep product register products. Do you need what and when and where I end and you might not have systems of those things in your own toolbox. And so having the capability to do something like that rapidly where all you do is bring it bring your data in Excel. Yeah, I really

wanted to just add a little bit more to the comment on parameterization the way you pose a problem. If it's it's the models and the Predictive Analytics capabilities seem to be chasing the data right where we're always getting new work more and more up-to-date information at High Velocity high speed. But if we go back to the fundamentals of statistical modeling the data in and of itself is a representation of the underlined generation process. So something there is is contributing that information that data and spitting it out

bring it back to your model and construct where he parameterize attributes or is that you can deduct a drive from the day. I said it selfie parameterizing stings on top of that the characterization of the contribution of uncertainty around those parameters where you're not just providing Point estimates on those parameters. Now, you've you have a distribution of possibilities around those parameters with that you drive from your data. Context in addition

with the after the decision framework with the decision spaces, you are now giving yourself more room to actually marry your decisions and align them with the acceptable thresholds within the uncertainty characterization of your parameters. So it is critical we're now as you didn't want more information more and more data updates the state of the parameters that you have corrective characterized. Now, you you draw your decision threshold. Why is it aligns with the distribution of the spirometer? So you're no longer saying well, you know Point estimate, you know, clippy 50%

Alexis value is absolute value. Which one lense to the best decision for us in our particular use case. I just wanted to you know, Echo the sentiment around from parameterization as a way to get a hand of a head of the influx of War. So you're not continually chasing the data with your model number, which is a Pisces right to answer to the question. We wanted to kind of asked was the what are some ways. He's biting model and how actually pervasive these

biases are and if you can share some examples and if you could pick it up and you only mentioned him directly from data, what are what are some other sources and what are some examples are no absolutely. That's all thank you, you know, we already started to touch on that. It's the data for the most part. That's a majority of it models are trained off of data weather in supervised models or unsupervised. AutoZone, whatever statistical approach that you take into your marvelous 2019 Harvard Business review article essentially

Define bias and a I always advise getting into any I do know that multiple ways. I could happen one is your ear training your models off of a set of information that is derived from a process where their heads from society whether it's labeled data from outcomes that you've observed in invariable invariably you might start to see effects of biases that I inherited Society creep into those in tables sets of how comes on which you train your data these decisions of these these biases, you know, often times reflect

no historical sites in equities and even being a variable Still exist within the outcomes observed in the dentist at 2. That's that's predominantly the the one way is when you're training your models off of these the next day. They actual the use of the results come in from Models train on data that contain these inherent viruses typically outcomes or output about models, you know, an informed decision capacity or another whether it's all the same a chicken with its decision to enter a market space where there is an intervention program that the model dictates were calculated

activities May mitigate certain buy a speeder level gymnastics radio models off of that you're making decisions now that are based off of bias predictions on how comes in these models. We seen recently Amazon stopped its use of a hiring algorithm after finding that essentially applicants were favored based on having you work like executed captured and do some some previously d a show that these terms are more frequently a line of found on on men's resume. So that is that is a

man feel more comfortable using some of these more action-oriented more pound in the hand on the table type words and the algorithm starts to learn that those attributes and treat are preferred by hiring managers. So that's one example several more. Are you all remember a couple years back where natural language? Wortham processing to wear on a I was being trained and he was being trained to respond in a certain with open source. We had you know, young people teenagers started seriously corrupting the xeelee training data. And before we knew it

the algorithm to hear myself cuz it responding and very very inappropriate manners. And so it's it's this symbiotic relationship between the AI learning quite a lot. I'll probably invite... To Johnson here to speak on some of those observations from from the policy and community of decision-making perspective. And I know do want to like bring about an example examples that you mention nothing very true and then I happen to know about Really don't know the bike is gone

by season. Right. That's an excellent. Excellent question since I I talked about being interested in the house on our end starting to do we are training the model of development capability your predictive capability off of data if initial starts to reveal that there is some imbalance across the data set some things that we started to explore our techniques that require models to have predictive hop equal predictive power across groups. So Indian actor named

rather simplistic way, if you think of a of a logistic regression algorithm that's got covariance that are aligned with Race For example doing some testing a priori to ensure that you have equal predicted power across all levels of race. If you have multiple levels that represent different ethnicities on Facebook groups. That's one way. A priori ensuring that your models do for these protectors protected Cove area that you do see equal predictive power or another ways to ensure that if you do remove

the those are protected. Attributes that you're not seeing an inability to explain the outcomes from the Romanian variables in the dataset took it's either you ain't sure that it's equally predictive across groups or that the introduction of those protected characteristics. Don't sway the model one where the human in the loop where decisions are being made off of models that may have been trained on bias data. We we need to we need to continue to pull the thread on explain ability and transparency of RTI model such that when the outcomes

are presented to the human in the loop to a human decision make make that they have the ability or they have other avenues to prove the AI recommendations for the results coming out of a to understand the impact of these disparate groups and show that those drivers are well characterized. Ensuring that the human does me can control in the way they interpret the outcome to ensuring that some technical methods are implemented to ensure Equity across groups and predictive capability. We also start to put on more high-level

mitigative approaches 1 ensuring diversity across the domain space in terms of the practice when we have practitioners across the AI ml data science group that do represent the representative of society-at-large. I think we Empower will provide us us a necklace and to be able to spot some of these biases better vs. Balanced set of practitioners to do one thing and of course, that's why you know, educationally, haha, haha. Haha and trying that you're you're introduced in the curriculum early enough. Exposing multiple people to today to science experts to to

bring into the domain finally ensuring and continuing to ensure that we I'll put it in place regulations and policies that mandates or equal representation across groups and data collection efforts. What do I mean by data collection efforts you imagine you're doing a clinical drug trial? How do you insure a representation across all groups that that drug that is on the test is the demonstrates equal level of ecstasy have crossed when you can also expand at 2 to Market in 2 to a funeral product trial tests and so on and so forth

mitigative approaches that we can start to to take to ensure that we accounts are in the effect of bias in the eye. Thank you very much. I really like the point where he said making the annals Community took the words right across from different perspectives and color like the area of the current state of ethical analytics and assessment of the current state of like how a warrant in Texas like helpful for social problems like this. And then the question is

what would you think? young you muted story So stipulate for you know the folks out there who do public sector a couple of oriented alarm that has a has a rich history going back 50 years there have been some of the Innovations in all in areas, like education transportation and security covid-19. The amount of presents that informs has in this space is truly amazing. I just learned about it and it's in the past few days. African-American I have a commitment to do research and teaching do service with on and on behalf of food has a color low and moderate-income

communities and marginalizing disadvantaged populations and a mi. Westmont is that this disinterest has has not always been as well represented in the space of public interest on applications as it as it as it could be also our engineering orientation towards making things more efficient and more effective. Can make it hard for us to think very different about the core purpose of our work that is how it could serve larger models of of social change. Here's an example of the movement for black lives came out for years ago with a policy platform that

addressed education Community Development criminal justice in the areas and those proposed the proposals that they they they put out. They were really bold and transformative like in a remake in the education system to eliminate the school-to-prison pipeline so forth, but at the time I thought of it as as really interesting and provocative but not really immediately useful for the kind of empirical analysis that that we often do in our community like wolves. Where is the the beat for what we want to do the worldwide protest?

What does defending the police really mean? What do policy and planning interventions? The True Value black lives look like in practice. So we have opportunities to to combine bold and Visionary thinking with the that the message that we know really well to trace some Transformer Solutions illustrate that I work with the recent PhD graduate to use qualitative decision analysis to learn with clients and providers of homeless services to improve performance management a lot of

times when figuring out what to do how to do things. We don't consult as much with the clients special clients who are March on. Another is the work that I do with the current doctoral students work with residents of urban slums in India about programs and services that can improve their learning conditions and how we can measure such Improvement ideas that when government wants to design an intervention, they come up with dramatically different Notions of what ought to be done. Then if we talked to the recipients themselves a third as a project to work with their birth stakeholders

in Boston to devise a portfolio policy interventions that conjointly address climate and housing crisis as they affect me some is vulnerable communities, but I can really do to make it impact on important social problems a lot of us and I believe that some in analytic space Macy social issues as not so congruent with The stuff they're doing that is really of a meeting important and that's a challenge to bridge that Gap. Thank you. Thank you for sharing

that. There's a lot to unpack into Lexi Perez likes to let me about like the example that you mention rice or barley the work about the example from India. Where is elected by the government is trying to design policies against the recipients have a difference in a box Justin Mike. It reminds me of like the same problems as well as the game TV models that we have an inelastic over how how how does one go about incorporating both sides and and still making the right decision for the

people that would be some of the talks in how to solve the problem. I think you start by asking who is traditionally at the table when proposing or devised a solution approaches and in this problem context and of course, I would defer to the doctoral student. Traditional approaches to have government bureaucrats decide what kinds of interventions ought to be done for example, tear down this slum and reload it. Everyone's a newly-developed office blocks. This is not so different than the

voices of those who are directly affected are not consulted as much as they could to determine whether this type of solution is usable and preto improving and we had to decide to learn for who is not at the table beside a how we can talk to local resume. The inside is that some kinds of methods the qualitative data analytic method. I'm referring to is called Valley Focus thinking it's traditionally been done with professionals educated people can be applied very well to those who are not necessarily

educated as long as we focus on uncovering values and Center invoice. If we do that it can enable us to identify the widest range of alternative courses of action that are meaningful to a variety of stakeholders and determining Picture it was common in the chat from David. How would one put like a emotional issues because no not really quantifiable, but all are or tomorrow so very quietly the right to Howard Howard 100 incorporate those emotional metrics into

the modeling. And so we traditionally think of emotions as a orthogonal to rigorous analysis of who says that we can make better decisions when we uncover the values that determine what objectives we wish to optimize. So by asking what is important to us as we really are tapping into our our feelings because we have they have a valence we say we want more of this or less of that and we feel strongly about this rather than something else. So it's not about thinking of looking at emotions as diverting us from our made goal, but instead

saying the things that are essential for human beings values and be a central part of our decision modeling address emotions directly like you if you try to catch their preferences and then they can be Quantified and put the models ahead if you look at it, but it's all about designing the right ejected function, right if you have the right of the function, then you decide Play it all going back to the original tops, right we can trade off between different parameters or variables that affect

the objective function. So the question it's wanting to yesterday acknowledged not yet. There are emotions that possibly could be factored in and structure and Quantified in a way that allows us to do the analytics right to mechanized an allergic right but it's a completely and that can be defined by water from the left or the right question up front. If you have the right question upfront then you know what he need to go look for the answer that question. Hey

vanilla, I would like to actually jump on this on this earth right here. This is a business sons of the issue that we're dealing with my tutu. The previous two points Army and both sides wanted decision-makers and then be affected population come to a common understanding of how to measure fairness. Right and Equity, so what may seem fair and Equitable to the decision-makers I hear government in this use case of the PHD student. I'm looking at siding in India, you know, what is fair to those that are going to

be affected and what is the measure of fairness to those that are implemented those decision? I need you can characterize it right here. You can characterize and and of course the difficulty is what's fair to me may not seem fair to the other. So we now have to if I go back to the eye doctor Johnson Mission with royal engineers at heart near you if you remember the traditional stress strain curve, right? It is going to be that overlap of the two distributions. What is that sweet spot where you are making sure that your models are operating in that region where

there is Call the wife between both perspectives and you want to meet her or ship you out of lakes in such a way that you have maximum between the implementers view of fairness. And the user is all the affected view of vanished. But excellent, excellent point that mites that conversation there will be two. address life is one of the busiest suggested was to Fake data from all groups equal in rights what what are some groups are more affected than others, you know dealing

with overabundance boss's day at the same time across to does it do things theme Augusta to play, you know, I would like to know that's a fantastic question about managing imbalance in data sets there a message where I mean, they're so so for example being Bayesian inference is superbly equipped to handle disparate data sets or smaller than a sex buddy traditional. Ml. I mean, we we have a standard approaches to balancing minority vs. Majority class when it comes to classification models, you know, and approaches,

you know include, you know leverage in the minority class and generating instances of that theater set within the region of reality of the existing between two lines that represent beautiful horses and making the assumption that I longed that Continuum between the two data sets that you can generate multiple instances of data. So these are some of the coaches that we used to to handle class imbalance in and did a horse transformation what you want to normalize the data set, but they are still available but apologize if I'm getting too much into the

methods abound Exact exact function. So Course, once you're not getting to simulations, if you have reached a comprehensive or agreed-upon definition across multiple parties of what kind of Jackson function Ivy Sanders looks like and you start to simulate with your your techniques of over on the sampling on managing class imbalance you start to see which representation of the underlined in dataset starts to get you to your objective function. Wolves, this is been like we covered a lot of clown and I was just looking at the time and it's okay. We we need to

try to are there any questions that you would like for a panelist to answer to address? What would you all think about like what we're talking about? Could see the pain center in the chat section. Cool so we can convince her but give them give them a bed so we could ask you once you take away that you would like for an audience to go home with from the discussion. What would you say? Thank you. I believe that Analytics. Super small one in an obvious Point academic. I'm not practitioner like my other distinguished analyst. So I tend to paint with a broad brush

should Encompass qualitative and quantitative data for evidence-based and values-driven decision-making in public policy and confined spaces. Mostly and so I know that I'm outside of our analytics who could greatest emphasis on voice and experiences other peers mostly inside Oriental is a related the greatest emphasis on rigorous and efficient and provably correct modes of inquiry that I have to say I don't feel them to believe are bound primarily by fydelity to Notions of justice and empowerment.

Important sure but Are there specific applications not so clear. My greatest wish is that this field is discipline and the profession can do good work embodies values that might be seen as controversial or insufficiently rigorous, but can really speak to on the special needs of this moment. Whether it's Housing Development or climate change. I want us to collect and use data and build a template models that that that contribute and intangible way to an emerging world of opportunity and justice. Thank you

Christian music store next it's been a wonderful conversation. You know, the quality of the conversation is really big and you can almost be oblivious to the audience when you get there. So immersed in the conversation is down and then we go to questions arise you realize all the election in always listening to you so that you know, Hindi Amazon example that Kobe find about and their stunt other examples as you know, we need to take people away from sentencing guidelines, right we need to make an automated but then they realize that the people that were designed the automated got guys

lined we're not. They were be biased. But the data using the crane tomatoes with themselves based on his maybe buy a happy resulted in perpetuating some of the things that you acknowledge and accepting that something like that could happen right and being able to stay away from that and having a technology platform that allows you to rapidly and refocus and recalibrate and come back as a problem without losing without the problems of a string out. There

are the two key takeaways and I would leave want to get knowledgement. I'm too hot in the technology capabilities to react to that knowledge mint and react fast enough would be the to be the rest of it is things I can be long but these are in my opinion and it speaks to the DNA of who could be successful in a speed like this. Definitely definitely and absolutely I would actually my takeaways are aligned with the with the hair stays where it has to be a reckoning and a sincere

acknowledgement that this is this is a real. Oh, you know phenomenal the the Introduction of bias in AI as a i stocks to take off and become more and more prevalent and its use across multiple lenses and multiple Endeavors that we engage in recognizing that it's real and not just that it's real chance by happenstance because of weight training dog with into using biosphere. There's also sometimes and Insidious or deliberate attempt to buy us data on which algorithms a trained. I think that is the even the

more pressing thing that we as an analytics space or practice practitioners at something that we have to stay alert to not going to get into everything that's happening world all day, but they're sometimes can be a deliberate attempt. To buy a step data so that you have these bias talcum. Where at where you don't end up stay. Well. Well, that's what the model said. So that's one 1 acknowledging it and not just acknowledging that it's real but that it's being sometimes but deliberately perpetuated

about it. Once you make that sausage man's right, that's part of the 12-step process. Once you accepted it. What do we do next? And on top let us continue the research into method refining those methods for mitigating the introduction by a sore spot in by that includes some of the policy impacts on the policy decision, but also incorporate into technological advancements, for example, I being published release the series of modules in Python called Fast 360

360 you to think that help you spot bias and algorithms to stay in. Top of what's available out there and incorporating them and lastly I would say the introduction of Ethics training in our engineering data science in statistical model in curriculum across School similar to What's Done in the Life Sciences building and then he'll carry medicine recognizing that we are making life-altering decision off of compute and off of things that we're doing with data. So early introduction to the impact of Ethics in in our practice

on level if I can help. Turn off devices that I can find in my behavior so bad if there are stomach viruses in them that we are using probably unknowingly. Just looking. By taking into account various representatives of the. I would say, we hope that everyone enjoyed the question which we could not get to I would ask you if you know what I mean, we can just pause for a second that question. Otherwise, you can follow influenced by misinformation. Obj want to start that's

a tough one. That's a tough one because again, is that aligns with what are internal? What is your internal Compass, you know on what's what's fair because that usually determine your susceptibility to the information. Sometimes we might end up in an hour and Eco chamber Echo chamber out where your you know where it would just reinforce in some of our underlying beliefs and biases so it is a larger it's at its largest size of question and not one that I can answer

immediately in the context of quantitative specific analytics, but I think these these the notion that each one of us, we have to examine internally our own understanding or attitude towards what's fair Equitable and then use it to adjust. From meter what we're observing be investigated be curious question things and challenge assumptions and and positions that are presented to you. And that's that's what I can answer it from a technical standpoint. Let yourself fall into that

bubble of Information Management. We talked about that. It is critical. I mean you got a place of knowledge that there's a problem to you and try to fix it. But again with the bias of how rapidly recalibrated and retest and an enso 335i and revalidate that your revisions to your models are give You truly are results now. I'm more in line with your what Your unbiased perceptions are real. You're you're so she was on the Block. So you might you might have knowledge everything and you might be great if not a problem. But if you're still

planning board go hunting at Islamic Center. we have to I purchase I'm really from qualitative and I wouldn't talk to individuals who are prone to be swayed by misinformation or disinformation or who may pay it and ask him. What their positionality is a sense of their there? They're their values and their receptiveness to evidence and empirical and if they're not open to a recalibration based on Purex, then that employers is not useful for the modeling that I would. Play

cool. I think they got this is I think we all can let go on and on forever. And even though these are all great topic and point once again like to thank you all the panelists. Let's kind of get back to the bedroom and like a kind of sunrise for the day so I can look and we won't have like a let me see if I can bring up my share my screen. Yes for today except for concession which is like right after they just want to take a moment to kind of tank all other speakers and panelists to thank our

sponsors and last but not least of cozaar participants for reading this bring up these questions like participating in the in these discussions. We kind of a hard in the in the Chino, right? So we we we talked about how we hurt me. Sheldon about how to use a thinking in navigating the world made for us. But you know, we have the tools and techniques to kind of man is risk address that I think that was a very very powerful message and then I'll we resolve the next election.

How how we think about these different issues and last but not least the panel, right? So we kind of talked about this parade is a space and hopefully they called us give you a a good sense of like, you know, when Alex is coming to the rescue like what are some of the aspects that we are covering that we need to be very cognizant of Bartlett wedding conscious about and do things otherwise, like we would be, you know, his morals and results in pokis, which one work for anyone right? So so I hope like you don't really really enjoyed enjoyed just as much as I did today.

So our next tomorrow let you know if your if you are on the fence coming back tomorrow like I would request you to know I think again, you know, we have a couple of similar great talk schedules. It's going to be another amazing date with that I take for the very next part is exactly what we when we should is a networking session. I've shared my screen, but we really want to hear from you right away. If you have something to share from wood from your perspective would the day was and how any ideas for us for a chapter, you know, what kind of events are working if you would like to

see what kind of other events for we are all ears right when we we need to hear from you. So I think we have about a motel room so we can interact in the more more one-on-one fashion, and then we can cover for the questions. Thank you all for joining. So just stay on the line. We will meet in separate living room stink you

Cackle comments for the website

Buy this talk

Access to the talk “Panel Discussion: Analytics to the Rescue: Ethically”
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free

Standard

Get access to all videos “IRAC Chicago 2020”
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Ticket

Interested in topic “Artificial Intelligence and Machine Learning”?

You might be interested in videos from this event

February 4 - 5, 2021
Online
26
147
ai, application, bot, chatbot, conversation, data, design, healthcare, ml

Similar talks

Anne Robinson
Chief Strategy Officer at Kinaxis
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Joris van de Klundert
Professor of Operations Management at Prince Mohammad Bin Salman College of Business & Entrepreneurship - MBSC
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Kirsten Martin
William P. and Hazel B. White Center Professor of Technology Ethics at University of Notre Dame - Mendoza College of Business
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free

Buy this video

Video
Access to the talk “Panel Discussion: Analytics to the Rescue: Ethically”
Available
In cart
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free
Free

Conference Cast

With ConferenceCast.tv, you get access to our library of the world's best conference talks.

Conference Cast
946 conferences
37592 speakers
14370 hours of content
Michael Johnson
Obi Nadu
Narasimhan Krishnan