welcome to ai matters 6(1)sigai.acm.org/static/aimatters/6-1/aimatters-6-1.pdf · ai matters,...
TRANSCRIPT
AI MATTERS, VOLUME 6, ISSUE 1 6(1) 2020
AI MattersAnnotated Table of Contents
Welcome to AI Matters 6(1)Amy McGovern, co-editor, Iolanda Leite,co-editor & Anuj Karpatne, co-editor
Full article: http://doi.acm.org/10.1145/3402562.3402563
Welcome and summary
ACM SIGAI Activities Fund 2020: Callfor ProposalsSIGAI Executive Committee
Full article: http://doi.acm.org/10.1145/3402562.3402564
Funding Call for AI Activities
AAAI/ACM SIGAI Job Fair 2020: ARetrospectiveMichael Albert & John P Dickerson
Full article: http://doi.acm.org/10.1145/3402562.3402565
Report on SIGAI Job Fair
AI Policy MattersLarry Medsker & Farhana Faruqe
Full article: http://doi.acm.org/10.1145/3402562.3402566
Policy issues relevant to SIGAI
Towards AI IngredientsCameron Hughes & Tracey Hughes
Full article: http://doi.acm.org/10.1145/3402562.3402567
Contributed Article
AI Fun MattersAdi Botea
Full article: http://doi.acm.org/10.1145/3402562.3402568
AI generated crosswords
Links
SIGAI website: http://sigai.acm.org/Newsletter: http://sigai.acm.org/aimatters/Blog: http://sigai.acm.org/ai-matters/Twitter: http://twitter.com/acm sigai/Edition DOI: 10.1145/3402562
Join SIGAI
Students $11, others $25For details, see http://sigai.acm.org/Benefits: regular, student
Also consider joining ACM.
Our mailing list is open to all.
Notice to Contributing Authorsto SIG Newsletters
By submitting your article for distribution in thisSpecial Interest Group publication, you herebygrant to ACM the following non-exclusive, per-petual, worldwide rights:
• to publish in print on condition of acceptanceby the editor
• to digitize and post your article in the elec-tronic version of this publication
• to include the article in the ACM Digital Li-brary and in any Digital Library related ser-vices
• to allow users to make a personal copy ofthe article for noncommercial, educationalor research purposes
However, as a contributing author, you retaincopyright to your article and ACM will referrequests for republication directly to you.
1
AI MATTERS, VOLUME 6, ISSUE 1 6(1) 2020
Submit to AI Matters!
We’re accepting articles and announce-ments now for the next issue. Details onthe submission process are available athttp://sigai.acm.org/aimatters.
AI Matters Editorial Board
Amy McGovern, Co-Editor, U. OklahomaIolanda Leite, Co-Editor, KTHAnuj Karpatne, Co-Editor, Virginia TechSanmay Das, Washington Univ. in Saint LouisAlexei Efros, Univ. of CA BerkeleySusan L. Epstein, The City Univ. of NYYolanda Gil, ISI/Univ. of Southern CaliforniaDoug Lange, U.S. NavyKiri Wagstaff, JPL/CaltechXiaojin (Jerry) Zhu, Univ. of WI Madison
Contact us: [email protected]
Contents Legend
Book Announcement
Ph.D. Dissertation Briefing
AI Education
Event Report
Hot Topics
Humor
AI Impact
AI News
Opinion
Paper Precis
Spotlight
Video or Image
Details at http://sigai.acm.org/aimatters
2
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
Welcome to AI Matters 6(1)Amy McGovern, co-editor (University of Oklahoma; [email protected])Iolanda Leite, co-editor (Royal Institute of Technology (KTH); [email protected])Anuj Karpatne, co-editor (Virginia Tech; [email protected])DOI: 10.1145/3402562.3402563
Issue overview
Welcome to the first issue of the sixth volumeof the AI Matters Newsletter.
This issue is smaller than usual as a reflec-tion to the disruptions caused by the on-goingCOVID-19 pandemic on the scientific commu-nity. We open with a call for funding by SIGAIExecutive Committe Chair, Sanmay Das, tosupport AI activities promoting outreach. Inour regular articles, we provide an event re-port by Michael Albert and John P Dicker-son on the AAAI/ACM SIGAI Job Fair. In thepolicy column, Larry Medsker and FarhanaFaruqe cover a new series on AI and Bias onthe Policy Matters Blog with a commentary onBias, Fairness, and Discrimination in the con-text of AI, along with discussions on AI pol-icy issues with respect to work and timeframefor AI impact. Another regular column is ourAI crosswords from Adi Botea. We have onecontributed article from Cameron Hughes andTracey Hughes on what constitutes the essen-tial ingredients for AI in a world where AI is in-creasingly pervading in every walk of life. En-joy!
Copyright c© 2020 by the author(s).
Special Issue: AI For Social GoodRecognizing the potential of AI in solv-ing some of the most pressing challengesfacing our society, we are excited to an-nounce that the next Newsletter of AI Mat-ters will be a special issue on the themeof “AI for Social Good.” We solicit arti-cles that discuss how AI applications and/orinnovations have resulted in a meaning-ful impact on a societally relevant prob-lem, including problems in the domains ofhealth, agriculture, environmental sustain-ability, ecological forecasting, urban plan-ning, climate science, education, socialwelfare and justice, ethics and privacy, andassistive technology for people with dis-abilities. We also encourage submissionson emerging problems where AI advanceshave the potential to influence a transfor-mative change, and perspective articlesthat highlight the challenges faced by cur-rent standards of AI to have a societal im-pact and opportunities for future researchin this area. More details to be coming soonon http://sigai.acm.org/aimatters. Pleaseget in touch with us if you have any ques-tions!
Submit to AI Matters!Thanks for reading! Don’t forget to sendyour ideas and future submissions to AIMatters! We’re accepting articles and an-nouncements now for the next issue. De-tails on the submission process are avail-able at http://sigai.acm.org/aimatters.
3
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
Amy McGovern is co-editor of AI Matters. Sheis a Professor of com-puter science at the Uni-versity of Oklahoma andan adjunct Professor ofmeteorology. She directsthe Interaction, Discovery,Exploration and Adapta-tion (IDEA) lab. Her re-search focuses on ma-
chine learning and data mining with applica-tions to high-impact weather.
Iolanda Leite is co-editorof AI Matters. She is anAssistant Professor at theSchool of Electrical En-gineering and ComputerScience at the KTH RoyalInstitute of Technology inSweden. Her research in-terests are in the areas of
Human-Robot Interaction and Artificial Intelli-gence. She aims to develop autonomous so-cially intelligent robots that can assist peopleover long periods of time.
Anuj Karpatne is co-editor of AI Matters. Heis an Assistant Profes-sor in the Department ofComputer Science at Vir-ginia Polytechnic Instituteand State University (Vir-ginia Tech). He leads thePhysics-Guided Machine
Learning (PGML) Lab at Virginia Tech, wherehe develops novel ways of integrating sci-entific knowledge (or physics) with machinelearning methods to accelerate scientific dis-covery from data.
4
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
ACM SIGAI Activities Fund 2020: Call for ProposalsSIGAI Executive Committee (ACM SIGAI; chair [email protected])DOI: 10.1145/3402562.3402564
Overview
ACM SIGAI invites funding proposals for ar-tificial intelligence (AI) activities that can takeplace entirely virtually and that have a strongoutreach component to students, researchers,practitioners not working on AI technologies,or to the public in general.
We funded several proposals with a simi-lar outreach thrust last year, but those werelargely based around physical interactions.With the necessity of responding to the cur-rent situation and the Covid-19 pandemic, weare instead focusing this year on activitiesthat can both take place virtually and pro-vide resources, material, or engagement tothe broader community through virtual means.Nevertheless, a focus of, and knowledge of,local populations that this could reach, and ev-idence of the ability to reach such populations,will be positively viewed.
The purpose of this call is to promote a betterunderstanding of current AI technologies, in-cluding their strengths and limitations, as wellas their promise for the future. Examples offundable activities include (but are not limitedto), virtual panels on AI technology (includ-ing on AI ethics) with expert speakers, creat-ing podcasts or short films on AI technologiesthat are accessible to the public, and holdingAI programming competitions virtually. ACMSIGAI will look for evidence that the informa-tion presented by the activity will be of highquality, accurate, unbiased (for example, notinfluenced by company interests), and at theright level for the intended audience.
ACM SIGAI will fund up to five proposals. Thefunding will be provided either as reimburse-ment (up to $2000), or as an honorarium (upto $1000), keeping in mind that the honorar-ium option has tax consequences for the re-cipient. In the case of honorarium payments,we plan to ensure that the final “deliverable”is in keeping with what was promised in theproposal before sending funding, and that you
Copyright c© 2020 by the author(s).
produce a writeup of the project for submis-sion to AI Matters, the SIGAI newsletter.
We will prioritize the following types of propos-als:
a Proposals from ACM affiliated organizationsother than conferences (such as ACM SIGAIchapters or ACM chapters).
b Out-of-the-box ideas that can be deliveredrapidly,
c New activities (rather than existing and re-curring activities),
d Activities with long-term impact, ande Activities that reach many people. We pre-
fer not to fund activities for which sufficientfunding is already available from elsewhereor that result in profit for the organizers.
Note that expert talks on AI technology cantypically be arranged with financial supportof the ACM Distinguished Speaker program(https://speakers.acm.org/) and arenot appropriate for funding via this call. Like-wise, webinars can be hosted in partnershipwith ACM SIGAI (e.g., https://learning.acm.org/techtalks/ethicsai). If youwant to participate in programs such as theseplease reach out to us separately.
Submission Guidelines
A proposal should contain the following infor-mation on at most 3 pages:
• A description of the activity (including whenand where it will be held);
• A budget for the activity and the amount offunding requested, and whether other orga-nizations have been or will be approachedfor funding (and, if so, for how much). Thiscould also be a request for an honorariumpayment instead;
• An explanation of how the activity fits thiscall (including whether it is new or recurring,which audience it will benefit, and how largethe audience is);
5
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
• A description of the organizers and otherparticipants (such as speakers) involved inthe activity (including their expertise andtheir affiliation with ACM SIGAI or ACM);
• A description of what will happen to the sur-plus in case there is, unexpectedly, one;and the name, affiliation, and contact details(including postal and email address, phonenumber, and URL) of the corresponding or-ganizer.
Grantees are required to submit reports toACM SIGAI following completion of their ac-tivities with details on how they utilized thefunds and other information which might alsobe published in the ACM SIGAI newsletter “AIMatters.”
The deadline for submissions is 11:59pmon June 15, 2020 (UTC-12). Propos-als should be submitted as pdf documentsin any style at https://easychair.org/conferences/?conf=sigaiaaf2020
Committees
The program is organized by the SIGAI Ex-ecutive Committee, and will be evaluated bythem with the help of other SIGAI officers andexternal reviewers.
The funding decisions of ACM SIGAI are finaland cannot be appealed. Some funding ear-marked for this call might not be awarded atthe discretion of ACM SIGAI, for example, incase the number of high-quality proposals isnot sufficiently large.
Contact
In case of questions, please first check theACM SIGAI blog for announcements andclarifications: https://sigai.acm.org/aimatters/blog/. Questions should be di-rected to Sanmay Das ([email protected])and Nicholas Mattei ([email protected]), the SIGAI Chair and Vice-Chair.
The SIGAI EC consists ofthe chair, vice-chair, sec-retary/treasurer, and pastchair of SIGAI.
6
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
AAAI/ACM SIGAI Job Fair 2020: A RetrospectiveMichael Albert (University of Virginia; [email protected])John P Dickerson (University of Maryland; [email protected])DOI: 10.1145/3402562.3402565
Introduction
For the sixth year running, AAAI and ACMSIGAI jointly ran the popular AAAI/ACMSIGAI Job Fair. In lockstep with the growthof AAAI and the growth of the greaterartificial intelligence and machine learning(AI/ML) community, our once-small job fairalso grew. This year, thirty-eight com-panies and universities formally attended—typically with a booth, team of recruiters,swag, and other representatives—increasingfrom twenty-six companies during the jobfair’s previous run in 2019, and twenty-onecompanies in the year prior to that. Lastyear, we purchased a dedicated domain—https://aaaijobfair.com/—for the jobfair. This year, we provided a link on that sitethrough which job-seekers—students, post-docs, practitioners, and maybe even a fewfaculty—could upload their resumes or CVs.We then shared that data and contact informa-tion for slightly under four hundred job-seekerswith participants on the other side: prospec-tive employers. Those employers are listed inthe section below.
Participating Employers• Association for Computing Machinery
(ACM)• AI Singapore• Alexander von Humboldt Foundation• Amazon• Apple• AppZen• Arthur AI• Audatic GMbH• Baidu• Beijing Century Tal• Bloomberg• Charles River Analytics
Copyright c© 2020 by the author(s).
• Dataminr• Elsevier• Google• Happy Elements• Hewlett Packard• IBM• Jane Street• Kitware• Mayo Clinic• Microsoft• NLMatics• Openstream• Point 72• Raytheon BBN Technologies• SGInnovate• SigOpt• Sony• SuperbAI• The Take• Tongdun Technology• United Technologies Research• University of Chicago Crime Lab• University of Southern California• University of Zurich• Waymo• Visa
We kicked off the job fair with a brief motiva-tional speech from the two organizers. Im-mediately following this—as in the two fairsprior to the present one—firms and universi-ties were given the option to speak for 60 sec-onds (often a touch longer, but who’s count-ing!) accompanied by a single slide. Most(over 75%) of employers chose to speak. As inprevious iterations of the job fair, this introduc-tory session served multiple purposes. First,it coalesced a group of interested job-seekers
7
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
in a central location. Second, it coalesced re-cruiters and representatives from employersin that same central location—they wanted tohear about their competition, and also matchfaces with the names on the resumes andCVs they were provided. Third, importantly,this served to introduce job-seekers to firmsthat they might not have heard of—start-ups,smaller non-profits, and firms that may haveless of a public presence. As usual, potentialemployers hailed from all over the world (e.g.,China, Germany, Norway, Singapore, Switzer-land, US) and from industry, academia, andgovernment. Job-seekers were as diverse asthe burgeoning AI/ML community is, presum-ably hailing from all over the world.
Figure 1: The job fair kicked off with a brief introfrom organizers. This helped gather job-seekersand representatives from employers.
Figure 2: A representative from each of the partic-ipating firms gave a 1–2 minute, single-slide pitch.
We hope that all participants (on both sides!)
Figure 3: A view from the action on the ground.
in this year’s fair enjoyed their time and foundthe experience worthwhile. We’d especiallylike to thank AAAI—especially Monique Abedfrom AAAI, for her boots-on-the-ground helpat the conference and her work, with CarolHamilton from AAAI, triaging emails in themonths leading up to the job fair. If you haveany comments regarding the fair itself, or sug-gested improvements, please get in touch!
Michael Albert is anAssistant Professor atthe University of Virginiawhere he has joint ap-pointments at the DardenSchool of Business andthe School of Engineeringand Applied Sciences.His research focuseson combining machinelearning and algorithmictechniques to automatethe design of markets.
John P Dickerson isan Assistant Professorof Computer Science atthe University of Mary-land. His research cen-ters on solving practicaleconomic problems us-ing techniques from com-puter science, stochastic
optimization, and machine learning.
8
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
AI Policy MattersLarry Medsker (The George Washington University; [email protected])Farhana Faruqe (The George Washington University; [email protected])DOI: 10.1145/3402562.3402566
Abstract
AI Policy Matters is a regular column in AIMatters featuring summaries and commen-tary based on postings that appear twicea month in the AI Matters blog (https://sigai.acm.org/aimatters/blog/). Wewelcome everyone to make blog commentsso we can develop a rich knowledge base ofinformation and ideas representing the SIGAImembers.
AI and DC
News Items for February, 2020
OECD launched the OECD.AI Observatory,an online platform to shape and share AI poli-cies across the globe.
The White House released the American Ar-tificial Intelligence Initiative: Year One AnnualReport and supported the OECD policy
Bias, Ethics, and Policy
The Policy Matters blog has started a serieson AI and Bias, with posts on backgroundand context of bias in general and then fo-cused on specific instances of bias in currentand emerging areas of AI. The information isintended to inform ideas and discussions onpublic policy. We look forward to your com-ments and suggestions. Extensive work suchas “A Survey on Bias and Fairness in MachineLearning” by Ninareh Mehrabi et al. is oneof the background resources for the conver-sation. Additional resources are provided byBarocas, et al. The guest co-author of thiscolumn is Farhana Faruqe, doctoral studentin the George Washington University Human-Technology Collaboration program.
AI Bias and Discrimination
Discrimination, unfairness, and bias are termsused frequently these days in the context of
Copyright c© 2020 by the author(s).
AI and data science applications that makedecisions in the everyday lives of individualsand groups. Machine learning applicationsdepend on data sets that are usually a re-flection of our real world in which individualshave intentional and unintentional biases thatmay cause unfair actions and discrimination.Broadly, fairness is the absence of any prej-udice or favoritism towards an individual or agroup based on their intrinsic or acquired traitsin the context of decision-making.Direct Discrimination. As described byNinareh Mehrabi et al., “Direct discriminationhappens when protected attributes of individ-uals explicitly result in non-favorable outcomestoward them“. Some traits like race, color, na-tional origin, religion, sex, family status, dis-ability, marital status, recipient of public assis-tance, and age are identified as sensitive at-tributes or protected attributes in the machinelearning world. It is not legal to discriminateagainst these sensitive attributes, which arelisted by the FHA and Equal Credit Opportu-nity Act (ECOA).Indirect Discrimination. Even if sensitiveor protected attributes are not used againstan individual, still indirect discrimination canhappen. For example, residential zip code isnot categorized as a protected attribute, butfrom the zip code one may find out about racewhich is a protected attribute. So, “protectedgroups or individuals still can get treated un-justly as a result of implicit effects from theirprotected attributes”.Systemic Discrimination. In the nursing pro-fession, the custom is to expect a nurse to bea woman. So, excluding qualified male nursesfor nursing position is an example of system-atic discrimination. Systematic discriminationis defined as “policies, customs, or behaviorsthat are a part of the culture or structure of anorganization that may perpetuate discrimina-tion against certain subgroups of the popula-tion.”Statistical Discrimination. In law enforce-ment, racial profiling is an example of sta-tistical discrimination. In this case, minority
9
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
drivers are pulled over more often than whitedrivers. The authors define “statistical dis-crimination is a phenomenon where decision-makers use average group statistics to judgean individual belonging to that group.”Explainable Discrimination. In some cases,discrimination can be explained using at-tributes like working hours and education,which is legal and acceptable as well. In awidely used dataset in the fairness domain,males on average have a higher annual in-come than females because on average fe-males work fewer hours per week than malesdo. Decisions made without considering work-ing hours could lead to discrimination.Unexplainable Discrimination. This typeof discrimination is not legal as explainablediscrimination because “the discrimination to-ward a group is unjustified”. Some re-searchers have introduced techniques duringdata pre-processing and training to removeunexplainable discrimination.
AI Bias and Fairness
In terms of decision-making and policy, fair-ness can be defined as “the absence of anyprejudice or favoritism towards an individualor a group based on their inherent or ac-quired characteristics”. Six of the most useddefinitions are equalized odds, equal oppor-tunity, demographic parity, fairness throughunawareness or group unaware, treatmentequality.
The concept of equalized odds and equal op-portunity is that individuals who qualify fora desirable outcome should have an equalchance of being correctly assigned regardlessof an individual’s belonging to a protected orunprotected group (e.g., female/male). Alongwith other concepts like “demographic par-ity” and “group unaware” are illustrated by theGoogle visualization research team with nicevisualizations using a “simulating loan deci-sions for different groups”. The focus of equalopportunity is on the outcome of the true pos-itive rate of the group. On the other hand, thefocus of the demographic parity is on the pos-itive rate only. Consider a loan approval pro-cess for two groups: group A and group B. Fordemographic parity, the overall number of ap-proved loans should be equal in both group Aand group B regardless of a person belong-ing to a protected group. Since the focus for
demographic parity is on overall loan approvalrate, the rate should be equal for both groups.Some people in group A who would pay backthe loan might be disadvantaged compared tothe people in group B who might not pay backthe loan; however, the people in group A willnot be at a disadvantage in the equal oppor-tunity concept, since this concept focuses ontrue positive rate. As an example of fairnessthrough unawareness “an algorithm is fair aslong as any protected attributes A are not ex-plicitly used in the decision-making process”.All of the fairness concepts or definitions ei-ther fall under individual fairness, subgroupfairness or group fairness. For example, de-mographic parity, equalized odds, and equalopportunity are the group fairness type; fair-ness through awareness falls under the indi-vidual type where the focus is not on the over-all group.
A definition of bias can be in the three cate-gories data, algorithm and a user interactionfeedback loop: Data – behavioral bias, pre-sentation bias, linking bias, and content pro-duction bias; Algorithmic – historical bias,aggregation bias, temporal bias, and socialbias falls; User Interaction – popularity bias,ranking bias, evaluation bias, and emergentbias. Bias is a large domain with much to ex-plore and take into consideration. Bias andpublic policy will be discussed in future blogposts.
AI and Work
The AI and Work session in the recent AAAIFSS-19 Symposia was a good example of ex-ploration and research that should inform pub-lic policy making. All topics are related to, orcould be expedited by, good public policy andaware policy makers. The deployment of AItechnologies in the future will likely require hu-mans to collaborate with AI systems, and thisrealization highlights the need for more sus-tained research on how to design such sys-tems. High levels of autonomy and the abilityto learn and interact with other systems, in-cluding humans redesigning work and rethink-ing incomes with bold ideas to improve thelives of workers and provide more interestingjobs with more meaning, purpose and dignity.
1. How do we design effective human-AI team-ing?
10
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
2. What does participatory design look like forAI in the context of work?
3. What training do people need to be able towork successfully with smarter systems?
Time Frame for AI Impact
An interesting IEEE Spectrum article “AI andEconomic Productivity: Expect Evolution, NotRevolution” by Jeffrey Funk questions popularclaims about the pace of AI’s impact on pro-ductivity and the economy. He asserts that“Despite the hype, artificial intelligence willtake years to significantly boost economic pro-ductivity”. If correct, this will have serious im-plications for public policy making. The articleraises good points, but many of the examplesdo not look like real AI, at least as a dominantcomponent. Putting “smart” in the name of aproduct does not make it AI, and automationdoes not necessarily use AI.
On a broader note, we should care about thetechnology language we use and beware ofthe usual practices in commercialization. Asdiscussed previously, expanding the mean-ings of terms like AI, machine learning, andalgorithms makes rational discourse more dif-ficult. Some of us remember marketing of ex-pert systems and relational databases: com-panies do a disservice to society by claim-ing each breakthrough technology actually isin their products. Here we go again today,with anything counting as AI depending on thepoint you want to make and the products youwant to sell.
Another issue raised by the article relates tostartups as the leaders of economic impact,as opposed to innovations from establishedindustry and government labs. Any technol-ogy has an adoption curve, going from earlyadopters through the laggards, of about sevenyears. If you add to that the difficulties of mak-ing a startup succeed, a decade or so is prob-ably the minimum timescale. A better per-spective on revolution versus evolution couldcome from longitudinal evaluations looking attrends. In that case, a good endpoint for ahypothesis about dramatic impact on produc-tivity might be the 2030-2035 time frame. An-other difficulty of using a vague and broad no-tion of AI is that policymakers could miss therevolutionary impact of data science, whichcan, but may not, involve real AI. Data sci-
ence probably has the best chance of dramati-cally impacting society and the economy soonand has the advantage of not having to involvedesigning and manufacturing physical objects,and thus not waiting for consumers to adoptnew products. Data Science is already affect-ing society and employment through obvious,and not so obvious, revolutionary impacts onour work and lives.
Please join our discussions at the SIGAI Pol-icy Blog.
References
1. Ninareh Mehrabi, Fred Morstatter, NripsutaSaxena, Kristina Lerman, and Aram Gal-styan. “A survey on bias and fairness inmachine learning”. CoRR, abs/1908.09635,2019.
2. Moritz Hardt, Eric Price, and Nati Sre-bro. 2016. “Equality of Opportu-nity in Supervised Learning”. In Ad-vances in Neural Information ProcessingSystems 29, D. D. Lee, M. Sugiyama,U. V. Luxburg, I. Guyon, and R. Garnett(Eds.). Curran Associates, Inc., 3315–3323.http://papers.nips.cc/paper/ 6374-equality-of-opportunity-in-supervised-learning.pdf.
3. Martin Wattenberg, Fernanda Viegas, andMoritz Hardt. “Attacking discrimination withsmarter machine learning.” Accessed athttps://research.google.com/bigpicture/attacking-discrimination-in-ml/, 2016.
Larry Medsker is Re-search Professor andfounding director of theData Science graduateprogram at The GeorgeWashington University.He is a faculty memberin the GW Human-
Technology Collaboration Lab and Ph.D. Hisresearch in AI includes work on artificialneural networks, hybrid intelligent systems,and the impacts of AI on society and policy.He is the Public Policy Officer for the ACMSIGAI.
11
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
Farhana Faruqe is a doc-toral student in the GWHuman-Technology Col-laboration Lab and Ph.D.program. Her researchincludes work on impactsof cognitive assistants onhuman-technology collab-oration. She investigatesissues in AI and Ethics,
human-machine trust, and impacts of AI sys-tem bias on individuals and society.
12
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
Towards AI IngredientsCameron Hughes (Northeast Ohio ACM Chair; [email protected])Tracey Hughes (Northeast Ohio ACM Secretary; [email protected])DOI: 10.1145/3402562.3402567
We have AI in our cars, in our mobile phones,and AI in our video games. We have AI inmedicine, AI in the military applications, andAI in government agencies. It’s getting harderto find an aspect of our daily lives that doesn’tpurport to have some kind of interaction withAI. We are relinquishing more of the personaland professional decision-making process tovestiges of evolving notions of AI. Not onlyare we starting to defer to AI for the decision-making process, we are subtly transferring theultimate responsibility for the decisions andthe consequences of those decisions to theAI. The public’s acceptance and reliance onvarious aspects of AI is becoming normalized.One major problem with this scenario is thatwe as a society are unclear about what con-stitutes AI. Our social position on AI is: wemay not be able to concisely or correctly de-fine it, but we all know it when we see it, right?Clearly the integral part that AI has in our soci-ety makes this position untenable and we canand should do better with our definition.
Even among AI researchers, educators, andpractitioners, there is some consternation anddisagreement about what constitutes AI andwhat doesn’t, and the fact that we are currentlyin an AI hype cycle doesn’t help matters. It’sno wonder that in the general public the term“AI” is routinely misconstrued and misapplied.In the AI community, we have a responsibil-ity to properly demarcate the tenets of Artifi-cial Intelligence, its mathematics, science, andapplication. We need to define things clearlyfor the laymen and the public at large. Buta clear, concise definition or presentation ofAI for the laymen or the public at large is atall order. AI research areas and techniquescover a wide range. Consequently, the com-mercial and government applications that de-ploy AI techniques from various research ar-eas can have significant differences. Table 1shows some of the research areas from theBio-inspired [1] approach to AI and the sym-bolic approach to AI.
Copyright c© 2020 by the author(s).
Table 1. Areas of Research from Bio-inspiredand Symbolic approaches to Artificial Intelli-gence.
Some Areas in Bio-Inspired Artificial Intel-ligence Research
• Artificial Neural Networks,• Cellular Automata,• Bio-inspired, nature-inspired algorithms,
search and machine learning algorithms,• Deep Learning,• Neural Networks,• Behavior-based modeling,• Swarm Intelligence,• Evolutionary Computing,• Nature-Inspired Metaheuristic Algorithms,
Some Areas in Symbolic Artificial Intelli-gence Research
• Knowledge Representation using frames,scripts, oav, conceptual graphs, modal log-ics,
• Expert Systems,• Logic-based Machine Learning e.g. Induc-
tive Logic,• Programming and Relational Learning,• Common Sense Reasoning,• Situational Calculus, Event Calculus,• Answer Set Programming,• Symbolic and Mathematical Logic,• Agent-Oriented Programming,• Associative Memory Models
Bio-inspired approaches to AI have differentassumptions, goals, vernacular, techniquesand tools than what are typically found in sym-bolic approaches to AI. They both representtwo very different schools of thought when itcomes to the possibilities of replicating hu-man intelligence and behavior by computerprograms or in computer hardware. While
13
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
there are hybrids of both approaches, most re-searchers tend to pick sides and consequentlythe boundaries and the fundamental definitionof what constitutes AI can vary dramaticallydepending on which side has the podium. Tomake matters worse, vagueness and ambigu-ity are introduced once these differing tech-niques, tools, and vernacular are manifestedin the form of commercial or government ap-plications. Further, many of these applicationscarry serious social implications and can havemajor positive or negative impact on society.Table 2 shows some uses of AI-based sys-tems that are used in the law enforcement, themilitary, and the legal system. The AI used ineach of these areas can irrevocably changethe trajectories of the human lives involved.
Table 2. Areas of Research from Bio-inspiredand Symbolic approaches to Artificial Intelli-gence.
LAW ENFORCEMENT
• Facial Recognition (FaceFirst, FACES,PoliceOne)Identify criminals and missing persons inpublic spaces and video footage.– Used by:– Police Departments,– US Airports,– National Human Genome Research Insti-
tute,– FBI.– Technology Used:– Principal component analysis using
eigenfaces,– Linear discriminant analysis,– Multilinear subspace learning.
• Smarter Physical Robots (DroneDeployin CA)– Bomb detonation,– Crime surveillance,– Crime scene investigation,– Accident Scenes,– Search and Rescue,– Crowd monitoring.– Used by:– Police Departments,– ICE,
– FBI,– Border Patrol.– Technology Used:– Unmanned and Remote controlled,– Autonomous,– Some equipped with Face Recognition
and other technology.
• Pattern Identification and PredictivePolicing– Identification of counterfeit goods,– Crime detection/prediction,– Forensic analysis,– Identification of potential perpetrators, vic-
tims and locations at increased risk ofcrime.
– Used by:– Police Departments (California, Washing-
ton, South Carolina, Alabama, Arizona,Tennessee, New York and Illinois).
– Technology Used:– Big Data Algorithms/ML behavior scripts,– Neural networks,– Databases,– Predictive Analytics.
• Bias Mitigation Tools– Removes racial biasness from police re-
ports that identifies a suspects race.– Used by:– Police Departments.– Technology Used:– Automatic Translation Information,– Identification, retrieval and information ex-
traction.
• Speech Recognition Interface (Nexgen,Dragon Law Enforcement for CAD/RMSSystems)– Police reports, incident reports and
search.– Used by:– Police Departments.– Technology Used:– Customized-Language/Statistical model-
ing.
• AFI System Data Collection and Min-ing(Project Maven/Algorithmic WarfareCross-Function Team)
14
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
– Used to identify individuals, associations,or relationships that pose a potential lawenforcement or security risk.
– Used by:– Homeland Security,– DoD.– Technology Used:– Computer vision algorithms,– TensorFlow APIs assist in object-
recognition on unclassified data.
• COPLINK:Develop information and knowledge man-agement systems technologies from hetero-geneous data sources.– Captures, accesses, analyze, visualize,
and share law enforcement-related infor-mation in order to solve cases and de-velop police reports.
– Used by:– Police Departments,– ICE.– Technology Used:– Database assessment/integration.
MILITARY
• Lethal Autonomous Weapon SystemsCan independently search for and engagetargets based on programmed constraintsand descriptions. Current systems (as of2018) are restricted to a human giving finalcommand to attack.– Offensive (Drones, Unmanned Vehi-
cles): Autonomously search, identify, andlocate enemies but can only engage witha target when authorized by mission com-mand.
– Defensive: Autonomously identify and at-tack oncoming weapon systems.
– Used by:– Military.– Technology Used:– Facial recognition,– Decision-making algorithms.
LEGAL SYSTEM
• Prison Sentencing Recommendations(Compas-Correctional Offender Manage-ment Profiling for Alternative Sanctions)
• Predicting outcomes of future trials
– Combat biasness and provide consis-tency,
– Used to assist in the sentencing of defen-dants by human judges,
– Weighing contradicting legal evidence,rule on cases in order to help humansmake better legal decisions.
– Used by:– Court systems (New York, Wisconsin,
California, Florida, and other jurisdic-tions).
– Technology Used:– Machine Learning.
The applications and the domain areas shownin Table 2, express the seriousness and grow-ing deference to AI in many aspects of soci-ety. We in the AI community have a solemnobligation to define what we mean by ArtificialIntelligence, its limits, its applicable scope, fail-ure rates, risks, potential benefits and costs.We have to find a way to clearly and effec-tively educate and inform the public about thistechnology. How else does the domain expert,community advocate as well as the laymen orpublic at large navigate the potential morassof notions that can be attributed to AI?
Considering the severity of the consequencesof using AI in these applications and domains,what precautions and communications havebeen employed regarding the fallibility of AI?Humans are fallible. Our technology is falli-ble. Data models can be incorrect and incom-plete. They can be correct and complete buttransient in nature because of structure, ge-ographical or culture changes in the underly-ing sample sets. Rules learned by machinelearning can expire as a result of dramaticchanges in the environment from which thedata culled. Valid decisions made, or conclu-sions drawn by the AI today, might be invalid10 years from now when made in the sameenvironment but different circumstances. Wehave to effectively communicate the risks in-volved. But where to start? We’ve yet toproduce a clear concise, correct definition ofAI suitable. How do we communicate limits,risks, safety issues?
Could labels on the proverbial container help?That is, should we have labeling in AI suit-
15
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
able for public consumption? In the same waythat we are now starting to demand labels forfruits and vegetables that have been genet-ically modified, or meats that are synthetic.Should we develop labels that describe the in-gredients of the AI in these applications or ser-vices? Should we apply the notion of expira-tion dates and safe use to our AI applicationsand services in the same way that we placeexpiration dates and directions for safe use onfoods and other consumables? Data modelsthat are used as the basis of decision-supportsystems for Prison Sentencing Recommenda-tion systems or criminal profiling can quicklybecome outdated as the result of population,cultural, social, or geographic change. So thata model that may be appropriate now, may notbe applicable at some future date. Those datamodels are historic and may not take into ac-count the biasness that exist in the data col-lection. The cost and effort required to pro-duce data models, or rule-based systems thatsupport the AI might introduce reluctance toroutinely update those models or rules. Oncethese systems are put in place, there will beinertia to prevent change.
What Are the AI Ingredients?
Therefore, should we label the AI application,data model, or rule set with an expiration dateor other temporal restrictions? Being able todescribe AI ingredients such as:
• Epistemic metrics,• Reliability indices,• Expiration dates, and other temporal re-
strictions,• Applicability scope,• Failure rates,• Safety considerations,• Federal Regulations/Law Alignment,
require that we have shared concise and cor-rect definitions for the various AI technologiesfor which these ”ingredients” will define and af-fect. Perhaps the simplest path to educatingthe public is to provide labels that contain theAI ingredients of an application, device or ser-vice. Using labels would allow the public toknow what AI is being purported, under whichsituations it can be reliably and safely used,
and when it expires. Table 2 shows someof the domain areas where AI-based systemsare in use that have serious consequences forthe public. If we had labels that detailed the AIingredients of these applications then the pub-lic would be in a better position to ascertainthe value and legitimate uses of such applica-tions.
In AI Matters, Volume 5 Issue 2 entitled: WhatMetrics Should We Use To Measure Commer-cial AI?, we discussed the need to not onlydefine AI, but to also be able to measure theAI that is in any given application, device, orservice. Clearly and concisely definable AI,measurable AI, and labels that contain AI in-gredients are steps in the right direction of ed-ucating and informing the public with respectto the proper viability, applicability, and utilityof applications, devices, or services that claimthe use of AI. The notion of AI ingredients isrelated to transparent AI and explainable AI. Inthe next issue of AI Matters, we want to take acloser look at the notions of labeling AI, mea-suring AI, and transparent AI.
References
[1] Floreano, Dario and Mattiussi, Claudio(2008). Bio-Inspired Artificial Intelligence.Theories, Methods, and Technologies. Cam-bridge, The MIT Press.
Related Articles• Greengard, Samuel. Policing the Future.
Communications of the ACM. VOL. 55 No.3. March 2012. pages 19-21.
• Kugler, Logan. AI Judges and Juries: Artifi-cial intelligence is changing the legal indus-try. Communications of the ACM. VOL. 61No. 12. December 2018. pages 19-21.
• Goose, Stephen., Arkin, Ronald. The Casefor Banning Killer Robots. Communicationsof the ACM. VOL. 58 No. 12. December2015. pages 43-45.
• Pasquale, Frank. Law and Technology:When Machine Learning Is Facially Invalid.Communications of the ACM. VOL. 61 No.9. September 2018. pages 25-27.
16
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
• Chen, Hsinchun., Zeng, Daniel., et. al.COPLINK: Managing Law EnforcementData and Knowledge. Communications ofthe ACM. VOL. 46 No. 1. 2003. pages 28-34.
Cameron Hughes is acomputer and robot pro-grammer. He is a Soft-ware Epistemologist atCtest Laboratories wherehe is currently working onA.I.M. (Alternative Intelli-gence for Machines) andA.I.R (Alternative Intelli-gence for Robots) tech-nologies. Cameron isthe lead AI Engineer forthe Knowledge Group atAdvanced Software Con-
struction Inc. He is a member of the advisoryboard for the NREF (National Robotics Edu-cation Foundation) and the Oak Hill RoboticsMakerspace. He is the project leader of thetechnical team for the NEOACM CSI/CLUERobotics Challenge and regularly organizesand directs robot programming workshops forvarying robot platforms. Cameron Hughesis the co-author of many books and blogson software development and Artificial Intelli-gence.
Tracey Hughes is a soft-ware and epistemic visu-alization engineer at CtestLaboratories. She isthe lead designer for theMIND, TAMI, and NO-FAQS projects that uti-lize epistemic visualiza-tion. Tracey is also amember of the advisoryboard for the NREF (Na-tional Robotics EducationFoundation) and the OakHill Robotics Makerspace.
She is the lead researcher of the technicalteam for the NEOACM CSI/CLUE RoboticsChallenge. Tracey Hughes is the co-authorwith Cameron Hughes of many books on soft-ware development and Artificial Intelligence.
17
AI MATTERS, VOLUME 6, ISSUE 1 MAY 2020
AI Fun MattersAdi Botea (Eaton; [email protected])DOI: 10.1145/3402562.3402568
1 L 2 3 4 I N 5 6 7 A 8
L 9 10 11 I12 13 14 15 16
17 18 19 20
21 22 G 23
24 25 A 26
27 M 28
29 30 31 32 E 33 34 35
36 37 S 38
39 40 41
42 43 44 45
C 46 47 N48 49 E A N
Across: 1) AI research area. 6) Individualamong mondern-era Veneti. 9) A word toemphasize. 10) Woodwind instrument. 12)Flowers whose purple popular variety symbol-izes wisdom. 15) Helpful accessories that aregood to see. 17) Fish-eating mammal. 18)Young Flanders. 20) AI research center in Cal-ifornia. 21) Apple computer brand. 22) Hair-shaping accessories. 23) West African coun-try. 24) Smoking accessory. 26) Stupid per-son. 27) City in New Jersey, USA. 29) Liga-ment. 32) Necklace medallion. 36) Irish musi-cian. 37) Sings a tune softly. 38) *, an opti-mal adversarial-search algorithm. 39) Pale ,a type of beer. 40) El , one of the Universityof Texas locations. 41) Personal cover. 42)Convert points into a reward. 44) Himalayaslanguage. 46) Girl from a famous picture inimage processing. 47) Garlic in a coq-au-vindish. 48) Achilles’ weak spot. 49) AI researcharea.
Down: 1) Flat screen technology. 2) Add a fileto an email. 3) Christmas in Paris. 4) Logicgate type. 5) Cooks food into a light brown
Copyright c© 2020 by the author(s).
color. 6) Direct descendant. 7) Landlord. 8) AIresearch area. 11) Test , a dataset or plat-form for scientific evaluation. 13) Sch. of Eng.and Appl. Sci. at Harvard. 14) Grab workfrom another thread in parallel computing. 16)Cogito sum, the basis of Descartes’ phi-losophy. 19) Gods’ mountain in Greece. 22)Successfully seek an extension. 23) Commu-nicated orally. 25) Keyword in an IF statement.26) Fashion sector. 28) Evil spirit. 29) AI re-search area. 30) CPU status when waitingfor jobs. 31) Acupuncture utensil. 33) Water-soluble base in chemistry. 34) Mini canvas atone’s fingertips. 35) Muscle-shaping activity.37) Bright star in the Aries constellation. 40)
pals in a remote relation. 41) Be active inthe space of hard drive disks. 43) Snake-likefish. 45) File type based on JAR.
Previous solution: RAILED - BELIE - ENTIRE -ARRANT - ADHERE - SEARCH - LEAN - PAID - SHE -MAC - HERA - LEER - SNARING - BINDS - ISSUERS- PESTS - ELAPSES - RATE - BRIG - CAP - ORR -FAST - BASE - UNIBUS - IBERIA - SEVENS - SOCCER- TREND - TAKERS
Acknowledgment: I thank Karen McKennafor her feedback. The grid is created with theAI system Combus (Botea, 2007).
References
Botea, A. (2007). Crossword Grid Composi-tion with A Hierarchical CSP Encoding.In The 6th CP Workshop ModRef.
Adi Botea is an AI se-nior specialist at Eaton,Ireland. He holds a PhDfrom the Univ. of Alberta,Canada. Adi co-authoredover 80 scientific papers.He has published cross-words in Romanian, innational-level venues, foralmost 3 decades.
18