Abstract
“Will Big Data and artificial intelligence (AI) finally make it possible for humans to do what many have longed to do—predict the future?”
Will Big Data and artificial intelligence (AI) finally make it possible for humans to do what many have longed to do—predict the future?
Almost everyone initially engages with futures studies wanting to predict the future of the world around them so that they can then learn how to adapt to and succeed in it. If they are only interested in predicting their own personal future, then they probably consult astrologers, tarot card readers, the I Ching, or their grandmothers.
I entered the futures field in the mid-1960s fresh from a PhD in political science that emphasized quantitative methods. I was especially interested in “judicial behavior” and the relationship between the ideologies of judges and their judicial decisions (Dator 1967, 1969) During this time, I attended and then taught in three extensive summer seminars on Mathematical Applications in Political Science, directed by Joseph Bernd (Bernd 1965, 1966, 1969).
One of the first projects I began on coming to the Department of Political Science of the University of Hawaii in 1969 was called “HAFDAM—Hawaii Alternative Futures Decision-Aiding Model.” The idea was to develop a computer model that Hawaii state legislators could use to consider possible consequences of alternative versions of legislation and enact the legislation that seemed to produce the results they intended. The model was to be based on the assumptions held by each legislator of how the world works, and not upon my views. The project got sidelined and eventually abandoned because of many factors that drew our attention elsewhere, none of them related to absurdity the concept itself.
Nonetheless, everyone serious about the futures field soon learns that the future cannot be predicted (i.e., “accurately and clearly foretold”). So they either give up and try to muddle through; or they pretend—or convince themselves—that they can predict the future anyway; or they learn—as I did—how to develop and use “alternative futures” (Dator 2009), constantly scanning the horizon accordingly, missing some changes/continuities while discovering others so that they may seem to be “wise before the event” (Scientific Council for Government Policy 1997).
There is also the issue of time frame. While everyone is interested in short-run futures (tomorrow to a few years hence), and so many futurists focus on short-run trends and events, others insist that the rate of social and environmental change is so rapid and the impact of technologies and events so long-lasting that it is necessary to focus on twenty to fifty years ahead to make wise decisions in the present. In the horse and buggy days of old, you (and your horse) could pretty much concern yourself with the familiar pathway immediately ahead. Now, in the era of speedy automotive and post-automotive transportation, the driver (human or mechanical) needs to look very many meters ahead and around, and react instantly to distant signals to steer or brake safely. Thus, some kind of narrowly focused trend extrapolation might have some validity in the very short run, but (depending on what is being measured and how) less, little, or no validity in the longer run.
Over the past fifty or so years of the existence of futures studies and forecasting, many theories and methods have been developed, tested, and used to overcome these challenges. There has been much discussion of theories and methods of forecasting and design in World Futures Review (WFR) and the other futures journals such as Futures, Technological Forecasting and Social Change, Foresight, and Journal of Futures Studies. And yet, most futurists still understand they cannot predict the future in any precise and useful way.
However, there has been considerable optimism recently that so-called Big Data and advances in machine learning and even in AI are already making possible what has heretofore been impossible. So as Professor Guillermina Benavides Rincón, the person responsible for conceiving and organizing this Special Issue of WFR, makes clear in her editorial, we thought it would be a good idea to see what impact these developments have made—or might make, or should make, or should not make—on the theory and practice of futures studies. We sent out the call and received responses that resulted in the papers in this Special Issue of WFR, as she described in her editorial.
I have a long track record of interest in robots and AI (Dator 2019a), and generally take what might be called a “strong view” of AI: namely, anything a human can do an AI can do (and often is already doing now or surely will be able to do eventually). When coupled with facilitating servomechanisms, this means that AI can physically manipulate things as well as engage in “thinking” and decision-making in manners equal to and/or vastly surpassing humans.
Of course there are many skeptics to this view, and indeed even expert opinion varies widely (Brockman 2019). However, long ago, David Miller, a professor at the International Space University in Strasbourg, France (and the University of Oklahoma, not in France), taught me that AI is a constantly moving target, always in the future, never achieved in spite of truly amazing things that toasters, watches, automobiles, airplanes, and scores of other machines have in fact learned to do. So, Professor Miller decided, “Artificial Intelligence is whatever machine’s can’t do yet.” Once a machine can do it, then it becomes just an unremarkable aspect of our environment, and we focus on other things machines can’t do yet, but that humans can do, perhaps to hold fast to the belief that humans are still the masters of terrestrial intelligence.
I later learned that Miller’s dicta may have been based consciously or unconsciously on the following statement by Larry Tesler: Tesler’s Theorem (ca. 1970). My formulation of what others have since called the AI Effect. As commonly quoted: Artificial Intelligence is whatever hasn’t been done yet. What I actually said was: Intelligence is whatever machines haven’t done yet. Many people define humanity partly by our allegedly unique intelligence. Whatever a machine—or an animal—can do must (those people say) be something other than intelligence. (Tesler 1970)
Indeed, the list of things that once only humans (or humans and other animals) could do that are now done by “machines” extends well back into human history, prehistory, and evolution, though growing rapidly now. I find the following steps in human–machine interaction to be instructive, and think we are nearing the last of the steps listed here.
A Possible Continuum of Human–AI Interactions
1. AI offers no assistance: humans make all decision and actions, or 2. AI offers a complete set of decision/action alternatives, or 3. Narrows the selection down to a few, or 4. Suggests one alternative, and 5. Executes that suggestion if the human approves, or 6. Allows the human a restricted time to veto before automatic execution, or 7. Executes automatically, then necessarily informs the human, or 8. Informs the human only if asked, or 9. Informs the human only if the AI decides to, or 10. AI decides everything and acts autonomously, ignoring the human, or 11. AI changes goals set by the human and does something else. (Based on Sheraton and Verplank 1978, as adapted by Sim et al. 2008, and by Dator 2012. Step 11 was added by Jerry Glenn on the APF list, March 22, 2019)
The implications of this are profound and numerous. For all of my career as a futurist, I have been intrigued by the possibility of AI and robots allowing humans to return to their early condition where nature was so abundant that “work” was not necessary (Sahlins 1972); that “work” is an invention of agricultural civilizations, made rigid and formalized during the industrial era and the development of the “science” of economics and the ideology of capitalism.
One of my first public declarations of this was in a talk I gave in 1970 before a Joint Session of the Hawaii State Legislature. There I said, in part:
I refer of course to advances in automation and cybernation which are eroding our time-honored notions about the priority of production problems over those of distribution of goods, and the relative value of work vs. leisure, of discipline vs. freedom, of responsibility vs. non-responsibility, and, in sum, those values and institutions of our present society which make the worth of a human being depend upon the social significance of that person’s labor. We are moving very rapidly towards a situation where it will be a privilege to work, not an obligation; where only a very small proportion of the population will be engaged in labor. How rapidly we get to that state, and with what upheaval, depends largely on what we do, or don’t do now. Factories have already discovered that by automating, they are not only relieving themselves of laborers, but also of purchasers of their products as well. Yet, our obsolete economic structures, and their supporting value systems, assume that goods are scarce, and the only labor available is human labor. Thus goods should be distributed on the basis of the social value of human labor. But what happens when goods are produced by machines alone, and there is no one “employed?” How are goods distributed then? In addition, if human value (and supporting institutions) are based on the assumption that since human work is necessary for survival, all institutions must be geared to forcing people to work and to deriving ultimate satisfaction from their work, what is going to happen when mechanical labor replaces human labor? What are we going to do then? But problems attendant to the replacement of manual labor by machines are the “easy” part of the cybernetic revolution. The “hard” part is that which faces us when we come to realize that all other functions which humans perform now can, and probably will, be taken over by machines. (Dator 2019b)
Since that time, as machines have become smarter and smarter, assisting and often out-stripping humans in one enterprise after another, I have been incessantly singing the song of guaranteed “full unemployment,” and the urgency of turning our full attention to imagining and designing how people can peacefully live lives of meaning and identity in such a world without “working”—as humans did for many thousands of years before agriculture, as Sahlins described.
But we are here talking about something precursor to and much more modest than “full unemployment.” We are interested in how “machines” might help, alter, or transform the work that futurists have traditionally done. And in preparing this volume, we had a big surprise. In spite of fact that many futurists are as fascinated by robotics and AI in various scenarios as I have been, I was not able to find many futurists who are pioneers in the AI field, and even fewer who seem to be contemplating, learning, or using AI or Big Data as aids in their work. This is strange, it seems to me (Kastrinos 2018; Rotha et al. 2019; Slaughter 2018a, 2018b, 2018c; Zhang et al. 2019).
It is of course highly likely that we overlooked many futurists who are AI pioneers, but in spite of personally inviting scores of likely contributors, not many people responded to our call for papers.
Moreover, when I surveyed the vast field of Big Data and AI itself, I saw that almost all of the many forecasts using Big Data and AI are by people who appear to be unaware of the futures field, expressing more confidence in their machine’s predictions—and less concern about the ethical and related issues of their undertaking—than I think are justified or healthy:
On January 9, the World Health Organization notified the public of Speed matters during an outbreak, and tight-lipped Chinese officials do not have a good track record of sharing information about diseases, air pollution, or natural disasters. But public health officials at WHO and the CDC have to rely on these very same health officials for their own disease monitoring. So maybe an AI can get there faster. “We know that governments may not be relied upon to provide information in a timely fashion,” says Kamran Khan, BlueDot’s founder and CEO. “We can pick up news of possible outbreaks, little murmurs or forums or blogs of indications of some kind of unusual events going on.” (Niiler 2020)
See also Hahn 2019, DellaVigna et al. 2019, Senior et al. 2020, and Agrawal et al. 2019.
One among the many concerns involved in relying on Big Data and AI for futures analysis that bothers me is that the assumptions built into the underlying algorithms contain biases that are often hidden to the creators as well as to subsequent users. Horrifying examples of this are widespread in the popular as well as academic literature. This concern is discussed by several papers in this issue of WFR.
A second concern is that while there is no doubt that Big Data and AI often enable researchers to perceive historical patterns of data even clearer than traditional statistical methods do, since they rely on past (even the very immediate past), and on very large amounts of data, researchers have difficulty identifying novel, emerging issues that interrupt old trends and start new ones. Emerging issues analysis (Dator 2018) is the heart and soul of futures work compared with typical forecasting, and I found no awareness of that in the AI/Big Data literature, as the Blue Dot example quoted above demonstrates, writing as though looking for early indicators was a great novel insight the author just discovered.
A corollary of this is that Big Data and AI techniques are probably more successful at “predicting” very short-run behavior and developments, while increasingly misleading if they are used to try to predict long-run futures.
A final concern, among many others that might be mentioned, is that since most futures work is done in the absence of an underlying theory of social and environmental continuity and change, our notions of cause and effect, and policies made upon those notions, may be overdetermined by the sheer weight of the Big Data that may obscure rather than illuminate the most important underlying causes. Lack of guiding theories is still the big Achilles heel of futures studies, I believe, and Big Data may encourage many of us to continue to ignore our throbbing appendage.
In any event, please thank the authors of the papers here for being intrepid pioneers, taking first steps toward exploring whether and how Big Data and AI might find a place in futures studies, and in helping futures studies mature in ways it has not been able to so far.