Artificial Intelligence: Its Impact on Human Occupations and Distribution of Income

ARTIFICIAL INTELLIGENCE: ITS IMPACTS ON HUMAN

OCCUPATIONS AND DISTRIDUTION OF INCOME

a pand moderated bv

Dr. NilB J. NilBBOn, Director Artificial Intdligwce Center and Dr. Sandra B. Cook, Manager Computer Securitv Program

SRI International Menlo Park, CA 94025

Panelist.~: Dr. ;\ian C . K ay, Chie f Scientist I Ice President Corporate Research Altni ht c. Sunnyvale, CA fl4086

Dr. Faye Duchin, Associate Dirertor /n.qlitule for Cconomic Analysis Nt;w l'ork Un£·rersity New York, NV 10003

Professor 1\far·gu.rel A Boden S'choo/ of Social Science.9 1'l1e University of Sussex J:rifthlon, J~ngland

Dr. Dennis Chamol , Asst. Director Department for PI'Dfes.sional Fmployees ,·\1"/,-C /0 Wa shington, DC 20006.

A:tificial Intelli gence (i\1) will have many profound so('ieLll <'fft>C'ts. It. promises potential benefits (and may 3lso po:;e ri~ks) ia ,, Jue:lticn, defense, business, law, and ~e icncc. In t . hi~ panel we explore how AI is likely to affect human occupations and the distribution of income. The pandist s have summarized their positions on matters

related to t.hPse topics below.

First, 1\ils Ni lsson a.rgu<'s that future technieal deYI'lopnwnl:; in AI wit! make it. economically attractive to replace h 11 mans by mac !Jines in many jobs, leading to a reduction in the total need for human labor. (Note: A written stat.em··nt. by Alan Kay for inclusion here was not available in time.) Next, Faye Dnchin analyzes in more detail t.hc errcels that these developments might have on employ nwnt and income distribution. Fears about unemployment involve both a fear of losing income and a fear of los ing mean ingful activity. While Dr. Duchin addre:>~cs f.he first. of these fears by discussing the possibility <, f deconpling income from employment, Ma.rgan·t Boden shows how AI can dramatically enhance our opportunities for richer and more human lifestyles. Finally, Dennis Chamot discusses many of the problems involved in the transition from our current economic systems to those that more properly take account of the potential of AT and related technologies.

ARTIFICIAL INTELLIGENCE AND Tim NEED FOR HU~-IAN LABOR

Nils J. Nilsson

ThPn~ a rl' two somewhat opposite ways to look at the pol·.·ntial of i\1 for rcdur ing t.tw need for human labor. Fir~ t , we mi g;!1t :~·; : ; urrH~ that AI is not really very differen t fr01n other technologies that hav.~ supported automation and incrPas ·.~ r.l productivity -- technologies such as l'l<·d.ronics , control engirwering, materials seience, anrl UJH'rat ion:. r\'searrh. A~ these t ('chnol0gies did , we might then cunciHde that AI also will ultimately lead to nn exp:t:Jdin :; t·ronorny wit.h a concomi tant expansion of emp loYiJH·nt opportunities. At worst, we might conclud<' tha t tlien· will besom'!, perhaps substantial , shifts in tl11' typ<'s of j0bs, but certain ly n0 overall reduction in the t.otal number of jobs. In my opinion, such a view is based on an over ly consenativc appraisal of the real potential of AI.

On t.he other hand , we might b~e our prediction on a rai her st. rong hypothesis 3bout AI -- a. hypothesis that sets AI quite apart from pn·Yions b .bor!saving teehnologies. Quite ~i rnp l y, this hypothes i~ is that anything (anything.') that. hu111 ans can do, AI ran do also. Certainly, AI has nol yet ac hieved anything like human-level performance in many important functions, but many of us believe that AI inevitably will equal and surpass human mental abilities. if not. in twenty y<'ars, then surely in fifty or one hundred. There is not vcr1 much that I can do in this short spare (or even in a tlJUch longer one) to persuade these who are presently unwilling to accept this hypothPsi,-. Perhaps many people, especially at a conference like this one, already believe it . Others may heli cvP ;nore-or-less weakened versions of it. (leading to on ly proportionally w<·akenPd conclusions).

The main conclusion of this assumption is that, even if AI crrat.<·s more work , no matter, this work ran also, in principle, be performed by AI d<'vices and, thus, will not llt:'Cessa ri!y lead to more jobs for humans. Of COUrse, just lweause some work can be performed automatically, doesn't mean that it will be so performed. Automation drrcnds on many fa.ct.ors, economic a.nd social The major economic factor would seem to be the relative costs of having humans or machines perform a given job (at a given rate and level of quality) . Again, AI differs from many previous labor-saving technologies in that it is (and will be increasingly so) a very low cost technology. (Remember, w mputa.tion is free!) .

AI can be expected to reduce particularly the need for burnan "white-collar· and service work. Included in this catq?;ory are many managerial functions, such as decision making, reporting, communicating and coordinating, fact gathering, and the supervision of similar activities by subordinates. Also included are many ·paper-handling· c!Pric:31 fun ctions. The · expert systems· and automatic planning prngra.ms currently being developed in AJ resl'arch bhoratorics and AI companies will be able to handle rna ny of thf"se tasks completely automatically. Because sueh AJ devices are mainly ·computational" (I. hat is, they involve only trivial amounts of energy and rnatr rials), we can expect their cost to be small compared to the cost of equivalent human labor, and thus we can conclude that these devices will in all likelihood be used instl'ad of human labor. Since it has been estimated that these · information-processing activities· occupy over half of ;\_merican workers, we can expect the automation of this work to have a significant. effect on the total need for labor.

Even though there are economic reasons to substitute mnrhines for human labor, we, as a society, may not choose to do so. That. is, we may decide to continue to employ humans in jobs ·next to the window· (as the Japanese say) merely as a way to distribute income and to giYe people something to do. But I leave that matter for the other panelists to ronsider.

COf·vfPUTERS, EMPLOYMENT, AND THE D!STrUDl iTION OF INCOt-.1E

Faye Duchin

There are severa.l related propositions underlying this pnncl that, fortunately , one begins to hear discussed with increa,<:ing frequency. The fundamental observation is that the lechnologica.l revolut ion based on the electronic, digital computer, now in its early stages, promises to he at least. as disruptive to the existing economic and social order as the agricultura l revolution of the tenth millenium B.C. or the industrial revolution of the eighteenth century. One of the direct consequences of the computer in the nex t few decades will he a significant shift in the composition of necessary labor skills and a dramatic red uction in bbor requirements per unit of output and probably also in the aggregate. An increasing proportion of lhe adult population may thus be excluded from the labor-force of the future and without any sources of income, unless we implement a massive onslaught on computer-illi teracy and begin to formulate realistic strategies for the large-scale distribution of income other than by the paycheck . I will confine my rema.rks to the options for the distribution and redistribution of income, particularly in the context of a developed economy.

The tc)tal national income earned by labor plus pruperty-type income accruing both to individuals and to businesses must (by definition) cover national expenditures for personal consumption and new business investment. A given bill of consumption and investment goods can be produced either in the existing way, or, alternatively, using new types of capital and significantly less labor. (Studies of technological change lead us to expec t declining labor to output ratios and increasing capital to bbor ratios. While the measurement of capital is problematic, capital to output ratios can be expected to fall once a certain level of industrialization is reached.) The unit-cost saYings obtainable by adopting the new technology will be distributed between labor and capital in the form of increasing (real) wage and/or profit rates. If the economy could grow indefinitely, then increased automation would make it possible to increase real wage and profi t. rates and keep everyone employed, as in the pa.~ l. There arc good reasons, however, why it may not be realistic to count on this outcome. ·

Declining employment opportunities could be accompanied by a sufficient rise in real wage rates to assure that labor's share of income was maintained or even increased. Then a significant distributional role could be played by mechanisms such as decreasing work hours or tho work lifetime to spread around existing jobs and thus increase leisure while maintaining income. Another opt.ion is moving toward larger households, within which income would be shared with nonworking · family · members. These alternatives naturally have different social implications.

If a significant part of the benefits of technological chaugc accru e to capital, some of this property-type income could be more >\idely distributed than presently is the ease if a ll or most workers owned, or received part of the return on, the capital they employ. This is achieved , for example, in the case of the farm family, which both owns its capital and provides its labor. In the future this could take the form of extensive industrial and service sector employee profit-participation or ownership schemes. Such an approach has the potential advantage of being dt>cent.ralized in its implementation, but reaches only t.hose who have jobs.

Some redistribution is, of course, already effected by ceulra.J governments through the taxation and transfer systems, and the scale of such operations could, and no doubt will have to, be expanded as the number of unemployed and impoverished grows. These programs will assume both familiar and new forms, like larger welfare programs on the one hand and negative income tax sc·hemes on lhe other. In these cases, the provision of income is dissociated from work.

Gon, rnrncuts can also effect transfers by organizing large-scale work programs like the maintenance of infrastructure or care and companionship for various categories of individuals (e.g., childcare), not to mention upgrading and updating the public educational curricula to teach computer literacy along with reading and writing. Such programs would have to be conceived as pennancnt measures, more like the public education syst.em than the New Deal. In this way jobs could be created and income redistributed through the provision of useful , even essential, services, financed by the taxation of those income shares that have benefited the most from automation.

In the future we will no doubt see increased reliance upon all of these mechanisms. The relative equality of the distribution of income that results and the smoothness with which redistribution is carried out will depend in large part on the quality of the policy dialogues now Nnerging. My colleagues and I are pursuing, in our rcs<'ar<·h on t.he changing structure of the U.S. economy, t.bc det.ai!Pd, concr<'te analysis of the types and numbers of jobs t.hat will be displaced and the new ones that will be required over the next few decades, under alternative scenarios about the nature and the pace of technological chauge in each sector of the economy. Such studies are a ncce~sary input to the formulation of realistic employment and income policies.

AHTif'ICJ.AL INTELLIGENCE AS A IIUYfANIZING FORCE

Margaret A. Boden

The quantitative unit of beauty, it has been said, should be the · rnilli-Hel en · -- the amount of beauty required to launch exactly one ship. The unit of happiness might be t.he · Polyneson": the amount of happiness enjoyed by a Sont.h S<'a Island er in one hour, at Pacific temperature and prPssure. But why? Why did European explorers compare Tahiti to t.he lost paradise of Eden, and why do man~· Westerners today envy the people of the South Seas'/ Climate and physical beauty are only part. of the answer: at least. as important is the Polynesian way of life, which seems more congenial -- even more human -- than our own.

On the one hand, Tahitians do not have to spend their days workin g in field or factory, since their island is so fertile thai. ripe mangoes (still) lie uneaten under the waysirle trees. On the other hand, they have developed a cul ture in which personal relations are supreme, in which famili es arc extended, sexual life is adventurous, and social ocrn.sions are marked by the subtle rituals of the dance. These facts are connected, for Tahiti's mango trees provide another precious resource besides food. Danci ng, offering the rup of · kava. , · playing with childrrn , sharing a fri end 's joys and anxieties, or exploring one's own-- all these social skills take time.

AI could be the Westerner 's mango tree. Its contribution to our food, shelter, and manufactured goods, and to the tunning of our a.dministrative bmeaucraeies, can free us not only from drudgery but for humanity. It will lead to an increased number of ·service" jf'Jbs -- in the caring professions, education, c·raft, sport, and entertainment. Such jobs are human rather than inhuman , giving satisfaction not only to those for whom the service is provided, but also to those who provide it. And because even these jobs will very likely not be fulltime, people both in and out of work will have time to devote to each other, which today they do not enjoy. Friendship could become a living art again.

Strange as it may seem, AI could foster a renewal of the ritual clement in Western life. Rituals (like the Pacific · kava · ceremony) are often scorned by us as · timewasting, · because of their apparent lack of function and rationality. But the common commitment to a specific social nexus is itself an important function, providing a human meaning to life distinct from breadand-buttl'r issues. Many rituals in our society have died be<":o use their context-defining functions have been taken over by mass-employment: one's workplace is, among ot her things , an arena for ritual enactment. Jobs provide a soe ird focus and context, a recognized network of social relat ions (sorne formally la.beled, others not) within which ind i1 idunls can seek their niche. With decreasing emphas is on a person's job, these functions will need to be fnifilled in other ways . This is not to say that old rituals will be revived, for many of them symbolized and reinforced particular social structures that have lost favour . Nor should one expect that postindustrial rituals will n.ttain the baroque degree of complication that some preindust.rin.i ecremonies did: there will, after all, be more alll·rna.tives available for filling our time. But, for all our impatience wit.h anachronistic absurdities, we should not deny the human satisfactions of etiquette and ritual.

S<'x-roles might be profoundly affected by Al-grounded changes in the distribution and division of labour. The iu t n'ase in out.-of-job time (to call it "leisure· time has some misleading connotations) will benefit more men than women, given that women today tend to have more of thi~ t.ime available. The new jobs in the service sector (which lws traditionally been seen as a largely female pre~Nve ) will :Jtt.ra.ct men, who will thus have increased occupat.ion1tl opportunities for ·caring· as well as · problem-solving. · Combined with the increase in time for domestic pursuits, this could foster a re-evaluation of the emotional dimension in the masculine role. Western men are discouraged from expressing emotion in all but a very limited family circle (and even there do not have the emot ional freedom that women do). With changed job­ patterns and increased ·social time,· they may achieve an emotional librration; this, in turn , could lead to shifts in family life and other social groupings and patterns of comrnu nica.t ion.

A re-evaluation of emotional and personal relations could also be encouraged by the likely contrast between computers and human beings, assuming that Al-programs in general will be unemotional systems. People will probably want (as they clearly do today) to draw a line between human beings and computers, and will argue (again , as t.hey do today) that they themselves are emotional whereas artificial systems are not. I do not believe -- as do most of the general public -- that computer-systems, in principle, could not even simulate emotion. But I do doubt that seemingly ·emotional· features will be built into the Al-programs used by public institutions and individual citizens. This view is not shared by the social psychologist, Neil Frude, whose book The lnl imale Afachine ( 1983) reminds us of the universal human tendency to anthropomorphism, and predicts wid c~pread sales of · friends-off-the-shelf. · These include cuddly, weeping, friendly robots akin to the inflatable dolls sold by sex-shops, and sympathetic conversationalists, religious advisers, and marriage­ counselors. I see this prospect as highly threatening: such quasi-emotional Al-products would be deeply dehumanizing, for they would lessen our reliance on, and concern for , our fellow humans. I find comfort, therefore, in judging Frude's predictions to be technologically and commercially implausible.

Even tho~e who deny that AI can acknowledge emotions arc fort\'d t.o admit that it ha.~ helped to bring the mind back into psychology. to make ideas theoretically r('~pcctahlc again. Behaviourism is a dead horse, and I shall not. flog it her<'; but it is relevant to note that AI was a main cause of it s demise. In this sense, AI has already had a humanizing influence.

So(·ial critics who complain of the alienation endemic to industrial ~ociet. y are commonly derided as unrealistic sentimentalists, or -- at best-- plucky idealists doomed to ~; cratc· h a living by a self-defeating self-sufficiency. This derision has two sources. One is the impossibility of producing modern goods and services (such as medicine) ou a pn·industrial economic ba.~e. The other is the Protestant. work ethic, which suggests that solving problems a.nd doillg things is the essence of a fulfilling life, personal concerns being mere irresponsible self-indulgence. This as tringent vi \'W may have been essential to the rise of capitalism, but it is becoming increasingly inapp ropri :lf.e now. Information t.echnology, while helping to provide the goods and services on which we have come to rrly, will undermine the Protestant ethic and make way for a different. view of life and society. AI can be a humanizing force , for it offers us the chance of developing a more Polynesian culture.

PROBLEMS OF TRA.t'\'SITION

Dennis Chamot

The nature of work in many industries is being transformed through the development and introduction of new technologies based upon computers and related equipment. Unlike previous periods of workplace automation, the changes being made today affect much more , than prod uct.ion industries and production workers. Office work and service industries are also becoming heavily dependent upon electronic systems. While many of the bendits of these changes have already begun to be felt -- greater productivity; new and better services -- the problems are only beginning to surface.

In the past, productivity improvements brought about by new technologies were often confined to particular industries, or more often, certain specific functions within production facilities . In the long run , businesses expanded and jobs were crented (although there may have been severe short-term dislocations). Usually other growing areas were available to absorb excess labor during the transition. This de facto piecemeal approach, coupled witti" shifts into growing economic sectors, permitted both overall economic growth and employment growth.

Today, all sectors are embracing computer technologies -- manufacturing, private services, government operations, agriculture. When the world recovers from the current severe recession, we may well be faced with economic growth fueled by higher capacity equipment, but without the high rate of job growth experienced in the past.

All this having been said, most people need to work. The primary questions for the next several decades relate to the pot~· nt.i al for long-term imbalance between demand for job~ nnJ the supply of meaningful, satisfying, and economically rewarding jobs. For well into the foreseeable future , most people will still derive a major part of their income from their jobs. Even in the most economiwlly advanced societies, and in the most terhnologiplly advanced industries, people will be required for planning, economic analysis, oversight, research and development., design, sales, technical support , customer service, and so on. For these people, computerized systems represent powerful tools for assisting them in the performance of their work . The problems df'five from the increas ing ability of these systems to take over much of the work now performed by manufacturing, clerical, and service workers.

Solutions will not be simple. Elements leading towards acceptable solul.ions must include human considerations in the direction and pace of change; shortening individual work time to spread available jobs around more widely; and creation of new products and services both to create addi.tional jobs and to meet. social needs.