Source Material Assignment. ARTICLES MENTIONED BELOW ARE ATTACHED.

F E B R U A R Y 2 0 1 8 | V O L . 6 1 | N O . 2 | C O M M U N I C AT I O N S O F T H E A C M 29

V viewpoints

P H

O T

O B

Y A

L I

C I

A K

U B

I S

T A

/A N

D R

I J

B O

R Y

S A

S S

O C

I A

T E

S

ment in our economic, social, and po- litical history, as we mindfully navigate human-machine interactions.

The ways in which machine sys- tems influence our lives have become more explicit in recent years. A chief example that commands popular at- tention has been IBM’s Watson, serv- ing as an informative bellwether for human-machine relations. Its inven- tors and user community place Wat-

E M E R G I N G A N X I E T I E S P E R T A I N -

I N G to the rapid advance- ment and sophistication of artificial intelligence appear to be on a collision course

with historic models of human excep- tionality and individuality. Yet it is not just objective, technical sophistication in the development of AI that seems to cause this angst. It is also the linguis- tic treatment of machine “intelli- gence.” Headlines decry the existen- tial threat of machines against humans in various media outlets. But what is really at stake?

Are we truly concerned that we will be surpassed in our capacities as hu- man beings? Or is rhetorical slippage betraying age-old philosophical ques- tions on what it really means to be hu- man? To what degree do our shortcom- ings in acknowledging human dignity in all populations (regardless of skin pigmentation, linguistic system spo- ken, geographical location, or socio- economic position) emerge in ques- tions pertaining to power dynamics between humans and machines? And how might we usefully juxtapose a his- toric study of our past categorical tax- onomies of humanity to more subtly inform our navigation of human-ma- chine relationships? In the fall of 2017 we engaged these questions and more with first-year students at Carnegie Mellon University: 16 students from the School of Computer Science and

the Robotics Institute and 16 students from the Dietrich College of Humani- ties and Social Sciences. In a time of ac- celerating technological disruption, the next generation of leaders and in- novators are ill-equipped to navigate this boundary chapter in human-ma- chine relationships. Perhaps our stu- dents can learn from how humans have treated humans to determine viable roadmaps for this challenging mo-

Viewpoint Teaching Artificial Intelligence and Humanity Considering rapidly evolving human-machine interactions.

DOI:10.1145/3104986 Jennifer Keating and Illah Nourbakhsh

30 C O M M U N I C AT I O N S O F T H E A C M | F E B R U A R Y 2 0 1 8 | V O L . 6 1 | N O . 2

viewpoints

As humans we are readers, we cre- ate, we imagine, we strive to under- stand. Our individual subjectivity al- lows us to do more than perform specific functions. And yet, present discourse on the potential for AI is of- tentimes laced with echoes of our anxieties pertaining to human dignity and its links to work or the distinc- tiveness of our subjectivity and agen- cy.4 Will we be ‘bested’ by the very ma- chines that we build? Will the next generation of technologists be equipped to consider these intended and unintended consequences for the tools they unleash? Or, for now, do we only dream quite wildly beyond tech- nological purviews about the actual sophistication of these tools?

Reading In the historical context of globaliza- tion, labor, human dignity, and edu- cation in the West, few rival the nar- rative potency offered by Frederick Douglass. In The Narrative of the Life of Frederick Douglass, An American Slave

he writes: “Very soon after I went to live with Mr. and Mrs. Auld, she very kindly commenced to teach me the A,B,C. After I had learned this, she as- sisted me in learning to spell words of three or four letters. Just at this point of my progress, Mr. Auld found out what was going on, and at once for- bade Mrs. Auld to instruct me further, telling her, among other things, that it was unlawful, as well as unsafe, to teach a slave to read.”5

The power of literacy and its capac- ity to equip individuals with the neces- sary tools to dismantle exploitative and unjust systems of power are illus- trated in Douglass’ work. His capacity to articulate the features of a power negotiation that undermines the very core of the master-slave relationship in a post-Enlightenment era is cap- tured in a human capacity to learn. In the context of the West and its politi- cal and social systems, literacy is an opportunity to assert agency. But hu- man-to-human capacities to assert equality, to facilitate Douglass’ ability to be ‘of no value to his master,’ to ren- der him to be ‘forever unfit … to be a slave’ due to his capacity to read, are not the tenets used to describe ten- sions between an AI machine and the tool’s “master” [read programmer or

son’s clinical knowledge squarely with- in the social context of the medical community, ascribing agency and a ca- pacity to learn to a sophisticated ma- chine with a human name: “Nobody can read it all,” Miyano said. “We feel we are a frog in the bottom of the well. Under- standing cancer is beyond a human be- ing’s ability, but Watson can read, un- derstand and learn. Why not use it?”6

Watson’s capacity to process data rivals that of a practicing physician and, in some domains, outpaces hu- man abilities. It is positioned as a tool that will rival human capacities in diagnostics to serve as release time for the practicing physician to dedicate more time, energy and in- tellectual bandwidth to patient-phy- sician interactions. The optimized functions of the Watson apparatus have limitations but they are certain- ly becoming more sophisticated rap- idly: “Before the computer can make real-life clinical recommendations, it must learn to understand and ana- lyze medical information, just as it once learned to ask the right ques- tions on “Jeopardy!” … The famed cancer institute [Memorial Sloan- Kettering] has signed up to be Wat- son’s tutor, feeding it clinical infor- mation extracted from real cases and then teaching it how to make sense of the data. ‘The process of pulling out two key facts from a “Jeopardy!” clue is totally different from pulling out all the relevant infor- mation, and its relationships, from a medical case … Sometimes there is conflicting information. People phrase things different ways.’”2

Read, Understand, Learn IBM’s Watson is personified as an independent agent in most press cov- erage. In contrast, at expert confer- ences like the “Humans, Machines and the Future of Work” conference at Rice University, AI systems like Watson are described as tools. Per- sonification is more tightly regulated when discussed or presented to tech- nologists who are not beholden to the mysteries of the black box, but rather its deconstruction into computation- al techniques. In the public domain, however, journalists ascribe person- hood to the learning machine, which is not necessarily corrected by engi-

neers or physicians, by describing the machine’s functions as reading and learning. Is Watson’s information processing and model-building truly reading or understanding? Does such a machine learn? Why do we ascribe features historically associated with humanity, subjectivity, and notions of a human self to built machines? And what chapters of human interre- lationships are threatened when we readily ascribe human characteris- tics to engineered systems?

Our society is locked in a stance of both anxiety and ambition in regard to the future of AI. We believe it is cru- cial that students embarking on un- dergraduate studies, as budding tech- nologists, writers, policymakers, and a myriad of other future leadership roles, should be better equipped and better practiced in engaging these dif- ficult questions. As automation will be a distinguishing feature in the next chapter of global economies, under- employment threatens the dignity of much of our human labor force. Yet as humans, most individuals would argue they are considerably more than a simple labor force driving a (al- beit pervasive and powerful) global economy. An insistence on our capac- ity to be more than what we might produce as commodities in a market is a distinguishing feature of human dignity in the 21st century. This is a concept, however, that needs to be tested, explored and seriously consid- ered as students prepare to enter this labor force and shape its direction for the coming generations.

Our society is locked in a stance of both anxiety and ambition in regard to the future of AI.

F E B R U A R Y 2 0 1 8 | V O L . 6 1 | N O . 2 | C O M M U N I C AT I O N S O F T H E A C M 31

viewpoints

user]. In the context of Douglass’ nar- rative, the prospect of literacy sug- gests the slave as worth more than his or her labor. Instead, the slave’s ca- pacity to learn, to engage in civilized discourse by joining what Benedict Anderson calls “an imagined commu- nity,” suggests his equal position with other members of the society in con- trast to the juridical category of slave as property.1 It is not the same scenar- io with AI because the machine is not human, although they are becoming increasingly social. This is relevant be- cause for a long period (arguably a pe- riod that we still occupy) humans have treated other humans as tools. The in- stitution of slavery, and its cousin in European colonial systems world- wide, used humans as machines com- posing a labor force. These were be- ings who were not extended existential features like agency, subjectivity, indi- viduality, or intelligence. As Joseph Conrad wrote in Heart of Darkness, the enslaved Congolese were “black shad- ows,” they were “bundles of acute an- gles sat with their legs drawn up.”3

Reading, however, marks agency for Douglass. Reading, in the context of an AI system, suggests anthropo- morphic undertones and perhaps hu- manity. The hierarchical power rela- tionships suggested in these examples are not in the service of hyperbole. The themes they introduce are also not entirely new. The historical analy- sis of human subjugation offers a rich tapestry regarding agency, identity, au- tonomy, labor, dignity and citizenry— issues at the heart of how future AI systems and humans will interrelate in our near future. Reading and learn- ing, the very verbs ascribed today to Watson, are central to Frederick Dou- glass’ discovery of the ways in which illiteracy in enslaved populations re- inforces the hegemonic power struc- ture of the master-slave dynamic be- fore abolition. Through reading, Douglass asserts his freedom in deed if not in legal standing.

Understanding Ascription of features like inconsis- tency, induction and emotion to ma- chines prematurely suggests essential human characteristics upon our inven- tions. Yet technologists forge ahead, projecting personhood and agency,

coupled with anxiety and uncertainty, upon machines. Even in the case of ear- ly light-seeking robots in 1950, W. Grey Walter recognized elements of human psychology, from free will to personal- ity: “… the uncertainty, randomness, free will or independence so strikingly absent in most well-designed ma- chines. The fact that only a few richly interconnected elements can provide practically infinite modes of existence suggests that there is no logical or ex- perimental necessity to invoke more than ‘number’ to account for our sub- jective conviction of freedom of will and our objective awareness of per- sonality in our fellow men.”7

Core human characteristics be- come terms for making sense of com- plex robotic behavior, as if complexity is sufficient to justify giving our ma- chines subjectivity. Today, serious le- gal experts are considering granting personhood to self-driving automo- biles, because these AI-driven ma- chines will be so socially integrated into our transportation infrastructure that they need to be individually liable for accidents. Notably, corporations were historically granted limited per- sonhood to shield individual humans from responsibility and blame; per- sonhood ascribed to AI similarly shields both corporations and engi- neers. Personification trades account- ability with convenience for tort and liability, and further with product marketing. Watson, for instance, is named as one technology product across many use cases; but in reality, each disciplinary version of Watson is a separate, custom-made instantia- tion with its own silo of AI, data store, and interface.

Core human characteristics become terms for making sense of complex robotic behavior.

Calendar of Events February 5–9 WSDM 2018: The 11th ACM International Conference on Web Search and Data Mining, Marina Del Rey, CA, Co-Sponsored: ACM/SIG, Contact: Yi Chang, Email: [email protected]

February 21–24 SIGCSE ‘18: The 49th ACM Technical Symposium on Computing Science Education, Baltimore, MD, Sponsored: ACM/SIG, Contact: Tiffany Barnes, Email: [email protected]

February 24–28 CGO ‘18: 16th Annual IEEE/ACM International Symposium on Code Generation and Optimization, Vösendorf/Vienna, Austria, Contact: Jens Knoop, Email: knoop@complang. tuwien.ac.at

February 24–28 PPoPP ‘18: 23nd ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, Vienna, Austria, Co-Sponsored: ACM/SIG, Contact: Andreas Krall, Email: [email protected]. ac.at

February 25–27 FPGA ‘18: The 2018 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays, Monterey, CA, Sponsored: ACM/SIG, Contact: Jason Anderson, Email: [email protected]. edu

March

March 5–8 HRI ‘18: ACM/IEEE International Conference on Human-Robot Interaction, Chicago, IL, Contact: Takayuki Kanda, Email: [email protected]

32 C O M M U N I C AT I O N S O F T H E A C M | F E B R U A R Y 2 0 1 8 | V O L . 6 1 | N O . 2

viewpoints

contemporary consideration of hu- man relationships to machines. We take inspiration from Raymond Wil- liams’ Keywords and the Key Words Project (http://www.kewords.pitt.edu) to create a conceptual structure of core themes that will guide the semester (see the accompanying table).

In this interdisciplinary course, students will be introduced to both the historical development of AI and to the current state of the art. As we engage with core themes of power negotiations, political implications for advancing technology, and cul- tural response, students will use ter- minology from Key Words to build conceptual maps that make sense of technological advances and their societal implications. Students will develop mixed media ‘futuring’ as- signments by semester’s end that offer speculations on the future of human relationships to machines. Working in groups, they will create their own narratives, synthesizing a future ethic based on course ma- terials and explorations. While this course is a first experiment in con- necting the freshman experience to socio-technical issues relevant to all, we hope that several iterations of course refinement and deployment will yield an approach that can serve as a valuable scaffold for AI and hu- manity across institutions.

References 1. Anderson, B. Imagined Communities. Verso, London, 1991. 2. Cohn, J. The robot will see you now. Atlantic. (Mar. 2013). 3. Conrad, J. Heart of Darkness. Bantam Books, NY,

1981, 26–27. 4. Deleuze, G. and Foucault, M. Intellectuals and power:

A conversation between Michel Foucault and Gilles Deleuze. L’Arc (Mar. 4, 1972).

5. Douglass, F. Narrative of the Life of Frederick Douglass, an American Slave. Penguin Books, NY, 1982, 78.

6. Gaudin, S. IBM: In 5 years, Watson A.I. will be behind our every decision. Computerworld (Oct. 27, 2016).

7. Grey, W. An imitation of life. Scientific American (1950), 42–45.

8. Williams, R. Keywords: A Vocabulary of Culture and Society. Oxford University Press, Oxford, 1983.

Jennifer Keating ([email protected]) is Assistant Dean for Educational Initiatives in Dietrich College, Carnegie Mellon University.

Illah Nourbakhsh ([email protected]) is Professor of Robotics at The Robotics Institute, Carnegie Mellon University.

Copyright held by authors.

Learning Rapid progress in AI/machine learn- ing and its central role in our social, economic, and political culture sig- nals its salience to the next genera- tion of students entering universi- ties. Building next-generation AI is currently a hot topic. At Carnegie Mellon, we have no trouble filling such classes. And yet, a nuanced understanding of the contributions that technologists are currently mak- ing to the world, an indication of how the next generation of computer sci- entists, engineers, and roboticists might shape the world that human- ists and social scientists study, is not at the forefront of our undergradu- ates’ minds. So, how might we en- sure this is something they consider throughout their undergraduate career? And that, instead, societal consideration shapes their under- graduate studies from their first year onward? We propose to introduce AI and Humanity in the first term of the undergraduate career. Humanities students will sit in class beside their colleagues from the Robotics Insti- tute and the School of Computer Sci- ence. They will be taught each class by a team of faculty with an intertwined pedagogical approach: a roboticist and a humanist.

Artificial Intelligence & Humanity is part of a new fleet of first-year courses called Dietrich College Grand Challenge Interdisciplinary Fresh-

man Seminars. These encourage fac- ulty teams to propose courses that at- tend to historically persistent problems facing humanity, demon- strating an interdisciplinary ap- proach to attending to these prob- lems whose solutions continue to elude us or demonstrate boundary work that a single discipline is often ill-equipped to solve. By harnessing the methodological approaches of various disciplines to demonstrate the complexity and the range of ap- proaches to problem solving in the academy, students are exposed to ar- gumentative structures and efforts to juxtapose historical human-to-hu- man relationships with future narra- tives of human-to-AI relations.

In Artificial Intelligence & Humanity, students will respond to historical examples of negotiations of power between individuals and communi- ties, then develop language to de- scribe contemporary and historical taxonomies of human-to-human and human-to-machine power rela- tionships. Starting with a survey of narrative forms that explore human relationships that include written memoirs, dystopian television shows, documentary films and sci- ence fiction novels, students will consider the various ways in which we narrate our relationships between humans from a variety of perspec- tives. They will consider how these relationships might manifest in our

Sample of syllabus keywords (left column) and related materials for analysis in seminar (from Williams8).

Agency Frederick Douglass: Narrative of the Life of Frederick Douglass, an American Slave Black Mirror: Men Against Fire

Self Black Mirror: Be Right Back Jerrold Seigel: The Idea of the Self

Technology Adam Hochschild: Bury the Chains Werner Herzog: Lo and Behold

Equality and Exploitation

Andrew McAfee: The Second Machine Age Star Trek: The Measure of a Man

Surveillance Minority Report Ian Ayres: Super Crunchers

Labor and Digital Labor

Joseph Conrad: Heart of Darkness Simon Head: Mindless, Why Smarter Machines are Making Dumber Humans

Citizen Kurt Vonnegut: Player Piano Black Mirror: Hated in the Nation

Narrative Richard Powers: Plowing the Dark David Herman: The Cambridge Companion to Narrative

Copyright of Communications of the ACM is the property of Association for Computing Machinery and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.