AI MATTERS, VOLUME 3, ISSUE 3 SUMMER 2017 Celebrating the Past, Present, and Future of Computing Timothy E. Lee (Carnegie Mellon University; [email protected]) Justin Svegliato (University of Massachusetts Amherst; [email protected]) DOI: 10.1145/3137574.3137581 Abstract Timothy Lee and Justin Svegliato, two Student SIGAI Scholars, cover The 50 Years of the ACM Turing Award Celebration, which con- vened in San Francisco last June. The semi- centennial celebration addressed the past, present, and future advancements of comput- ing, ranging from deep learning and ethics to augmented reality and quantum computing. As Student Scholars sponsored by SIGAI, we are grateful for the opportunity to be a part of The 50 Years of the ACM Turing Award Celebration. For two days in June, hundreds of professors, researchers, and students from across the globe gathered together in San Francisco to celebrate the legacy of the Tur- ing Award (often referred to as the Nobel Prize of computing) and the incredible advances in computing over the last 50 years. The semi- centennial celebration also honored this year’s Turing Award recipient, Tim Berners-Lee, for inventing the World Wide Web and related net- working technologies such as the Semantic Web. After opening remarks from a Turing laure- ate each day, we heard from several panels that spanned the field of computing, ranging from deep learning and ethics to augmented reality and quantum computing. Every panel featured a distinguished moderator and sev- eral panelists that included Turing laureates, prominent researchers, and rising stars in the field. Interleaved with the panels were sev- eral short films. These films featured the life and work of the father of computer science, Alan Turing, and highlighted the Turing lau- reates’ contributions, including those made to the field of artificial intelligence by the AI Tur- ing Award laureates. We were honored by the attendance of several AI Turing Award recipi- ents: Judea Pearl (2011 Turing laureate), Ed Feigenbaum (1994 Turing laureate), and Raj Reddy (1994 Turing laureate). Copyright c 2017 by the author(s). Figure 1: SIGAI at The 50 Years of the ACM Tur- ing Award Celebration. Pictured from left to right: Timothy E. Lee, Yolanda Gil, and Justin Svegliato. The bronze bust of Alan Turing unveiled during the conference is also shown here. The first panel Advances in Deep Neural Net- works was particularly relevant to the SIGAI community. Moderated by Judea Pearl, the panel featured Michael Jordan (UC Berke- ley), Fei-Fei Li (Stanford), Stuart Russell (UC Berkeley), Ilya Sutskever (OpenAI), and Raquel Urtasun (Toronto). As a popular area not only in AI but also in computing in gen- eral, deep learning has emerged as a powerful approach for enabling machine intelligence. Sutskever explained that neural networks are essentially tunable circuits that learn high- dimensional mappings from data. Deep neu- ral networks have emerged from the conflu- ence of several factors: the recent advances in hardware (the “oxygen” of neural networks ac- cording to Sutskever), the availability of mas- sive datasets via the Internet, and the accel- erated progress of data science. Despite their promising results, a common theme emerged from the panel. Although ef- fective in particular domains, deep learning in its current form cannot be the fundamental ab- straction of machine intelligence sought after by researchers. There are many questions 22
Transcript
AI MATTERS, VOLUME 3, ISSUE 3 SUMMER 2017
Celebrating the Past, Present, and Future of ComputingTimothy E. Lee (Carnegie Mellon University; [email protected])Justin Svegliato (University of Massachusetts Amherst; [email protected])DOI: 10.1145/3137574.3137581
Abstract
Timothy Lee and Justin Svegliato, two StudentSIGAI Scholars, cover The 50 Years of theACM Turing Award Celebration, which con-vened in San Francisco last June. The semi-centennial celebration addressed the past,present, and future advancements of comput-ing, ranging from deep learning and ethics toaugmented reality and quantum computing.
As Student Scholars sponsored by SIGAI, weare grateful for the opportunity to be a partof The 50 Years of the ACM Turing AwardCelebration. For two days in June, hundredsof professors, researchers, and students fromacross the globe gathered together in SanFrancisco to celebrate the legacy of the Tur-ing Award (often referred to as the Nobel Prizeof computing) and the incredible advances incomputing over the last 50 years. The semi-centennial celebration also honored this year’sTuring Award recipient, Tim Berners-Lee, forinventing the World Wide Web and related net-working technologies such as the SemanticWeb.
After opening remarks from a Turing laure-ate each day, we heard from several panelsthat spanned the field of computing, rangingfrom deep learning and ethics to augmentedreality and quantum computing. Every panelfeatured a distinguished moderator and sev-eral panelists that included Turing laureates,prominent researchers, and rising stars in thefield. Interleaved with the panels were sev-eral short films. These films featured the lifeand work of the father of computer science,Alan Turing, and highlighted the Turing lau-reates’ contributions, including those made tothe field of artificial intelligence by the AI Tur-ing Award laureates. We were honored by theattendance of several AI Turing Award recipi-ents: Judea Pearl (2011 Turing laureate), EdFeigenbaum (1994 Turing laureate), and RajReddy (1994 Turing laureate).
Figure 1: SIGAI at The 50 Years of the ACM Tur-ing Award Celebration. Pictured from left to right:Timothy E. Lee, Yolanda Gil, and Justin Svegliato.The bronze bust of Alan Turing unveiled during theconference is also shown here.
The first panel Advances in Deep Neural Net-works was particularly relevant to the SIGAIcommunity. Moderated by Judea Pearl, thepanel featured Michael Jordan (UC Berke-ley), Fei-Fei Li (Stanford), Stuart Russell(UC Berkeley), Ilya Sutskever (OpenAI), andRaquel Urtasun (Toronto). As a popular areanot only in AI but also in computing in gen-eral, deep learning has emerged as a powerfulapproach for enabling machine intelligence.Sutskever explained that neural networks areessentially tunable circuits that learn high-dimensional mappings from data. Deep neu-ral networks have emerged from the conflu-ence of several factors: the recent advances inhardware (the “oxygen” of neural networks ac-cording to Sutskever), the availability of mas-sive datasets via the Internet, and the accel-erated progress of data science.
Despite their promising results, a commontheme emerged from the panel. Although ef-fective in particular domains, deep learning inits current form cannot be the fundamental ab-straction of machine intelligence sought afterby researchers. There are many questions
about its long-term viability as the bedrockof machine intelligence. As Jordan argued,today’s neural networks are deep architec-turally, but not semantically. Pearl questionedwhether these networks could reason aboutcausality, a central theme in his foundationalwork on Bayesian networks.
Outlining the weaknesses of today’s deepneural networks segued into a discussion onthe types of intelligent behavior that humansexhibit but these networks currently lack, suchas semantic understanding, contextual rea-soning, abstraction, and reasoning under un-certainty, all of which are easily handled by hu-mans despite little training data. Russell drewa fitting analogy between Allen Newell, CliffShaw, and Herbert Simon’s General ProblemSolver and the need for exponential comput-ing with deep learning and the need for ex-ponential data. In Russell’s opinion, hopingto achieve “tabula rasa” machine intelligencewith only deep learning may be infeasible insome—or all—domains due to the data de-mands, and so we must continue to searchfor better techniques. Li offered a similaranecdote from her work with ImageNet. Withenough data, deep neural networks are by farthe state of the art in object recognition, butthey perform poorly and cannot reason effec-tively without massive datasets. In the caseof robotics, Urtasun noted that being unableto model uncertainty well in deep learning is aconsiderable drawback in her current work onself-driving vehicles where algorithm robust-ness is critical.
Still, even with these shortcomings, deeplearning performs quite impressively in nar-row problems, such as computer vision, imagecaptioning, and object segmentation. In somecases, such as AlphaGo, it enables decision-making capabilities that are superior to humanintelligence. Several of the panelists agreedthat deep learning has matured enough to beused in industry, but the search for machine in-telligence must continue. Ultimately, the panelcould be best summarized by Li’s comments:we are entering the “end of the beginning” forAI. Deep neural networks may be one of ourbest existing tools for enabling the develop-ment of intelligent agents, but even greaterbreakthroughs are yet to come.
In addition to the deep learning panel, the
opening day of the celebration also featuredfour other panels with many prominent re-searchers from industry and academia, alongwith a talk by 2008 Turing laureate BarbaraLiskov that explored the history of comput-ing. First, in Restoring Personal Privacy with-out Compromising National Security, WhitfieldDiffie, 2015 Turing laureate, along with sev-eral leaders in security, cryptography, and net-working discussed how governments couldobtain useful information using backdoors andother intentional vulnerabilities to aid crimi-nal investigations without jeopardizing the pri-vacy of society. Following a short film onAlan Turing’s life, we then turned to Vint Cerf,2004 Turing laureate, and several other distin-guished researchers in Preserving Our Pastfor the Future. They considered the problemof how to store data for centuries to come andwhether corporations or governments shouldfund such an endeavor.
Later that day, Moore’s Law Is Really Dead:What’s Next? headlined the 1992 Turing lau-reate Butler Lampson. The panel exploredthe ways in which the field can continue thetrend of exponential technological growth de-spite that Moores Law has continued to slowdown. During the panel, a common themeemerged: researchers will eventually leveragespecial-purpose hardware and quantum com-puting to push the boundaries of computingforward.
At the end of the day, we heard from RajReddy in Challenges in Ethics and Comput-ing. Given the increasing relevance of AI andmachine learning, Reddy believes that ethi-cal questions in computing have become moreimportant than ever. Noel Sharkey addedseveral important questions in light of re-cent progress in self-driving cars and machinelearning. How can self-driving cars make de-cisions that were once reserved for humans inlife and death situations? And how do we en-sure that data-driven algorithms escape biasagainst minorities in the justice system?
On the final day of the conference, there weretwo panels on some of the most rapidly grow-ing fields in computing following a talk fromDonald Knuth, 1974 Turing laureate. Quan-tum Computing: Far Away? Around the Cor-ner? Or Maybe Both at the Same Time? thatfeatured the 2000 Turing laureate Andrew Yao
23
AI MATTERS, VOLUME 3, ISSUE 3 SUMMER 2017
investigated the current state of quantum com-puting and how it might drive software devel-opment in the next 50 years. Like the deeplearning panel, John Martinis cautioned thatquantum computing is only a powerful tool incertain combinatorial problems but useless inothers. However, in areas like AI, machinelearning, and cryptography, it has the poten-tial to revolutionize the field.
The celebration culminated with a panel on anarea of computing that has recently seen rapidprogress: Augmented Reality: From Gamingto Cognitive Aids and Beyond. Fred Books,1999 Turing laureate, and Ivan Sutherland,1988 Turing laureate, reminisced about theearly work of augmented reality while theywere aspiring researchers. Peter Lee dis-cussed the impact of augmented reality onthe gaming industry. Other panelists consid-ered Google Glass, Pokemon Go, and OculusRift and explored the inevitable future of aug-mented reality in the home and at the work-place.
For all attendees across the spectrum of com-puting, the advancements of the last 50 yearshonored during the conference will undoubt-edly shape our own contributions for the next50 years to come. And, for the SIGAI com-munity, the experts in our field gave insightinto the ongoing search for machine intelli-gence and how deep learning might play arole. Given the recent groundbreaking ad-vances in AI, it was only fitting that the cele-bration of computing’s greatest achievementswas in honor of who many call the grandfatherof AI, Alan Turing.
Acknowledgements
We gratefully acknowledge the support ofSIGAI for the opportunity to attend The 50Years of the ACM Turing Award Celebration.We also thank Yolanda Gil for her help withthis article.
Timothy E. Lee is a M.S.student in Robotics atCarnegie Mellon Univer-sity. As a member ofthe Robust Adaptive Sys-tems Lab under Profes-sor Nathan Michael, Tim-othy’s research focuseson improving mobile robotintelligence to enable theautomation of challengingtasks in real-world set-tings. He is currently in-
vestigating robust, vision-based navigation ofa submersible robot to automate the precisioninspection of underwater infrastructure.
Justin Svegliato is asecond year Ph.D. stu-dent in Computer Scienceat the University of Mas-sachusetts Amherst. Inthe Resource-BoundedReasoning Lab underProfessor Shlomo Zilber-stein, Justin’s researchfocuses on bounded ratio-nality, real-time decisionmaking, and autonomous
agent architectures. He is currently develop-ing metareasoning techniques that monitorand control algorithms that trade decisionquality with computation time.
