Monday, May 25, 2015

Growing old in the age of machine learning

One of these two people will never get old 

"I customarily killed old women. I used to kill my aunts [classificatory aunts] when they were still moving [alive]. ... I would step on them, then they all died, there by the big river. ... I didn't use to wait until they were completely dead to bury them. When they were still moving, I would break them [their backs or necks]. .. I wouldn't care for old women; all by myself, I would stick them [with his bow]"

I found this chilling testimony in the book "The World until Yesterday" by Jared Diamond, in the chapter "The treatment of old people: cherish, abandon or kill ?".This a testimony from an Aché Indian man, given to anthropologists Kim Hill and A. Magdalena Hurtado.

The Aché are a nomadic people subsisting on wild forest resources. Amongst several such nomadic tribes, encouraged suicide or even ritualized murder of old people has been noted by anthropologists. This might horrify us.  But as I was reading this book, I was left wondering if our modern society is any better. This is because specific cultural traits that characterize nomadic societies are increasingly getting common in our current modern society and these changes are being amplified by technology. These include unsteady systems of production, seasonal downturns from economic surplus to starvation, fierce competition for territory, and limited use of experience that comes with old age.  The underlying catch-phrase to denote these changes is "technological disruption". But its practical effect is that our economies are resembling more those of nomadic societies than those of steady systems of production, such as based on agriculture.  So with increased automation and robotization of society, will we humans be trampled as useless old people by robots ? Or will it be some fellow humans, equipped with new technologies, that trample on the rest of humans as futile competition ?

There is currently a scare in popular media about artificial intelligence (AI) and how that threatens our future as humans. I will present more discussion on this below. These scare stories about AI are nothing new and enter our popular culture in periodic waves. However, this time, there is something quite dramatic happening in many applied areas of information sciences as computing power and storage capacities of machines have crossed a threshold. As a computer scientist, I enjoy the success of this research and have the palpable sense of big changes coming. But as a regular person on the street, I believe this important discussion is distorted into extreme caricatures.

Philosophers and pundits see intelligent machines as something akin to God, with immense power to do good or bad. A more sensible view would be to consider them as artificial life forms, subject to Darwinian laws of natural selection and adaptation. Whether we can share a common environment of  successful coexistence with these intelligent machines is a question that is best answered by experts in ecological biology, not philosophy. Just as there are many kinds of living organisms, there are many kinds of intelligent machines. Some are designed explicitly to alleviate the physical and mental handicaps of people. For example, new sensors and camera systems are being developed that can help the blind and partially sighted people to perceive the world. New bionic devices can help the physically handicapped people to move autonomously.  But other  intelligent machines and software systems (connected to the "cloud" data servers on the internet) have different objectives, not necessarily aligned to human betterment. Their intelligence is also very context-specific, suited to various ecological niches in our market economy. In light of all these aspects, we need to adopt a biological / ecological mindset when thinking about machine intelligence.

More specifically, experts in human anthropology have studied a wide range of human societies on how they cope with respect to different environmental conditions and competition. I find it unfortunate that we don't tap into their expertise to the problem of how human societies will adapt to increased automation. It is obvious that this process affects different people in different ways. It can be argued that old people will be affected in a very different manner to young people. In this aspect, it is interesting to observe how the treatment of old people varies across different societies.

In one extreme, we have nomadic tribal societies like the Aché. Other examples of nomadic tribes murdering old people include the Kaulong people of New Britain, the Chukchi people of Arctic Russia,  and the inhabitants of the Banks Islands in the South Pacific. In many other nomadic tribes, old people are abandoned when the tribe shifts camp. These examples include the Lapps of northern Scandinavia, the San of the Kalahari desert, the Omaha and Kutenai Indians of North America, as well as the Aché Indians mentioned earlier. A more benign way of killing old people, practiced by far more human societies, is ignoring them i.e, letting them to starve, to wander off or to die under neglect. This is reported among the Inuit of the Arctic, the Hopi of the North American deserts, the Witoto of tropical South America, and the Aboriginal Australians.

Indeed, old people are of little value in nomadic tribes subsisting on seasonal produce. These societies also have little resources to spare to take care of them.  The situation only changes when the tribes become sedentary, produce agricultural surplus, and have place for new occupations where the old people become an asset with their long life experiences, instead of being a liability due to their reduced mobility and foraging skills.

In well-established societies where the old people's lives are no longer at risk, an opposite tendency evolves, where property rights become concentrated in old people. For example, in the Confucian Chinese, southern Italians, and Mexican households, all the economic authority is vested in the "patriarch", who is the family's oldest living male. In ancient Greece, old people were effective rulers of the society in a gerontocracy. Not only the choicest land property, but also the fertile young women become the property of these old men. As women become mortgaged into relationships with men of much older age, they no longer have the opportunities to develop their intellectual faculties and engage as equals in a society. Instead, they get shunted into a purdah and become mere objects of family honor. Young men and boys become the junior partners in a relationship with older men, which often also turns into sexual subjugation. These social phenomena have been observed in a multitude of societies - Greece in the classical period, the middle east in the middle ages etc.

Thus, we see that there are two extremes with respect to how old people are treated by human societies. This is still a crude one-dimensional approximation. Importantly, the treatment of old men and old women differ in considerable ways.  But it is still illuminating to pose the question as to which direction our modern society is evolving.

When we think of old age, we typically think of physical deterioration - the wrinkles on the face, the weakening muscles, or the graying hair. But more than this, we share a deeper dread about old age,  that we will become irrelevant to the society.  It is this feeling of isolation that makes old age unbearable and kills people. Susan Pinker, in her book "Village effect", studied village societies in Sardinia where old men and women have long and fulfilling lives with complete social engagement in their neighborhoods.  She argues that this social engagement is missing in North American societies, as people have become more mobile and do not have friends and family in the same city they live in.  I think the social malaise is deeper, as our modern societies are increasingly resembling nomadic societies, instead of settled village societies.  It is in this sense that we need to approach how automation affects the process of us humans growing old: Will it accelerate us towards it or will it completely liberate us from it and achieve immortality ?

At the outset, the second premise may sound even nonsensical to a novice reader. But there is a belief amongst technological optimists - Ray Kurzweil being the most famous exponent, about the singularity, which is a point in the near future when machines become more intelligent than humans. Singularity optimists believe that humans will merge with machines - nanobots cleansing our bodily organs, brains wired up to the internet, cell and gene rejuvenation by software updates etc. They hope that this will translate into vastly improved health and opportunities, a hope expressed most vividly by Peter Diamandis in his book "Abundance". This hope of technological hybridization extends even to immortality as the architecture of the brain is uploaded to the cloud.

I look at this tribe of technological optimists as the followers of any other religion, framing their beliefs as a matter of faith. In reality, there is not much evidence to show for a technological rapture moment. But the elite of the software industry take this seriously.  Google is closely involved with the Singularity university of Kurzweil.  Just as there are singularity optimists, there are singularity pessimists. Bill Joy, co-creator of  the Java programming language, wrote an influential essay in 2000 titled "Why the future doesn't need us?".  Philosophers such as Nick Bostrom worry about "existential risk", where super-human intelligence might annihilate 100% of humanity. This is taken seriously by the likes of Elon Musk.  Nick Bostrom gives an analogy about chimpanzees, who are genetically very close to humans, but who depend completely on the grace of humans for their survival. So he argues, the survival of us humans will be at the grace of super-human machines. However, in order to do great harm, machines need not be "intelligent",  "conscious", or even "malevolent".  A rogue AI with all the intelligence of a flu virus can do great harm. This may cause collapse of economies or even the death of vast numbers of humans. But this does not qualify as "existential risk", as at least a small percentage of humans and human institutions will still be alive and  functional. To compare, even climate change, one of the most severe threats facing us humans, is also not an existential risk.  In contrast, a super-human AI is considered capable of destroying 100% of humanity.  A comparable risk is an asteroid impact that destroys our inhabitable earth. Why are the elite of the software industry, like Elon Musk, fixated on such existential risks ?

If I want to be uncharitable, I can say that it is because it is a risk that concerns them personally. Even if 99% of the humanity gets destroyed, the elite of the industry will most likely survive. But existential risks like a hypothetical super-human AI are a different story. But I don't think it is a conscious bias. It may simply be due to the lack of a compelling alternative narrative on the risks of machine intelligence.

In fact, even short term trends about automation in the current age don't look distinctly beneficial to humans. One of the most vocal authors on the debilitating effects of automation on our economies is Martin Ford, whose new book "The rise of the robots" argues how intelligent machines may be shrinking our economies and raising unemployment to obscene levels.

While discussing about optimists such as Kurzweil and Diamandis, Martin Ford says,

 "In general, technology optimists tend to underestimate the impact inequality. They don't think enough about what this means for 90% of the people. Of course, these are extra-ordinary people. Ray Kurzweil probably does not hang out with average typical people. He is living in an elitist tower."

Ford argues for a guaranteed minimum income for all humans.  This may be an improvement, but a minimal income may not necessarily mean an opportunity to rise up.  A cautionary tale is what befell aboriginal peoples when they were put into reservations by colonialist powers. With their traditional ways of life destroyed and without any hope for future, most of these aboriginal people succumbed to alcoholism and other addictions. It is quite possible that such a predicament may befall a vast majority of the human race.  Here is another testimony from Jared Diamond himself, from his earlier book "Guns, Germs and Steel"

"As a teenager, I spent the summer of 1956 in Montana, working for an elderly farmer named Fred Hirschy. Born in Switzerland, Fred had come to southwestern Montana as a teenager in the 1890s and proceeded to develop one of the first farms in the area. At the time of his arrival, much of the original Native American population of hunter-gatherers was still living there. Among the farmhands, there was a member of the Blackfoot Indian tribe named Levi, who behaved very differently from the coarse miners - being polite, gentle, responsible, sober, and well-spoken. He was the first Indian with whom I spent much time, and I came to admire him. It was a shocking disappointment to me when, one Sunday morning, Levi too staggered in drunk and cursing after a Saturday-night binge. Among his curses, one stood out in my memory.  

-"Damn you, Fred Hirschy, and damn the ship that bought you from Switzerland!"      
It poignantly brought home to me the Indians' perspective on what I, like other white schoolchildren, had been taught to view as the heroic conquest of the American west. Fred Hirschy's family was proud of him, as a pioneer farmer who had succeeded under difficult conditions. But Levi's tribe of hunters and famous warriors had been robbed of its lands by the immigrant white farmers." 

When people living in traditional societies make contact with modernity and avail themselves of modern appliances, healthcare and state-supported institutions, their material well-being undoubtedly gets better. However, this improved material well-being does not always translate into more fulfilling lives, greater social engagement, and most importantly, into greater hope and opportunities. As the modern society races past like a super-fast train, traditional people often feel left behind. They don't find the necessary means or skills to contribute to a changing society. Consequently, a lot of them suffer from addiction (this is also an under-current in the above passage by Diamond). Drugs and alcohol take a huge toll on these unfortunate people.

With the increasing success of machine learning algorithms, the whole of humanity is under a similarly grave threat of becoming obsolete and losing hope for a meaningful engagement in society. The scale of this threat is inconceivable for even highly intelligent and educated people. So we argue about what kind of education we can give for young graduates in the universities to "compete against" the machines. The short and simple answer is none. Nada. Zilch. There is nothing that a machine cannot do when  trained on sufficient amount of data. The first jobs to get automated will be those where data is readily available or those where the salaries are at a premium: "expert systems" were first developed in the 1980s for automating medical guidance. Essentially, old people (experts with experience, but unable to learn new skills) will be automated first. Highly skilled engineers in many developed countries are already facing severe risks of losing their jobs as the pace of technological change is getting faster than the speed at which they can learn new skills. Thus, the allegory of the Aché Indian man murdering his aunts, related in the beginning, is applicable figuratively, though thankfully not literally, to the current situation in technological employment. It has to be noted that to be replaced by a machine, the task does not need to be completely automated, but only partially automated and thus shipped off to a human with far fewer skills, and thus willing to work for much cheaper.  An analogous situation has developed in the creative sectors, where the vast majority of artists, journalists and musicians are being pushed out of professional employment and surviving on the margins of the organized economy. Intermittent work opportunities force the majority to adopt a "technomadic" lifestyle with few or zero social benefits. This is particularly true of work in the visual effects (VFX) industry.

Thus, it is the "cool jobs" that are being lost first to automation by software. There is not as much financial incentive to automate "crappy jobs". It is said that robots will automate jobs with the 3 D's - Dirty, Dangerous or Drudging (Repetitive). This is no longer the priority for technology on the market. The software guru and investor Marc Andreessen famously said "Software is eating the world". A more accurate saying would be "Software is eating old people" (By that, I mean people with skills and experience acquired over a lifetime, and with limited budget of time and resources to change). But as the cost of machine learning gets lower and lower, and data is collected from every single human individual (sometimes surreptitiously by spying), even simpler tasks will get automated.  In effect, every human being will have a virtual copy of himself on the cloud, eager to perform all the tasks he is capable of, but better and vastly cheaper.  As machine learning continues to get cheaper and better, the effective age of replacement by a machine will sink lower, ultimately replacing even young people. How do we create hope for humanity in such a society ?  I think this is a more urgent question than worrying about the risks of a super-human AI.

Another interesting author to consider is Jaron Lanier, who specifically analyzes creative industries and the devastating effect that the internet had on musicians, artists etc. He argues for intellectual property protection and micro-payments built into the very structure of content consumption on the internet, as first devised by hypertext pioneer Ted Nelson. Property rights are an important means of protecting the interests of old people, and they arose first in agricultural societies. The desire for stronger property rights, especially on intellectual property, is greater in older economies and in countries with greater numbers of old people. But taken to the extreme, property rights stifle innovation and make the young subservient to the old.  An even greater risk is that property rights don't necessarily encourage production, but merely function as a rent of passage through constrained distribution channels. This will stifle the economy. We need to create a hopeful society for both young and old humans, but with a clear understanding that nobody can ever be younger than a machine.

I don't have a solution myself. I don't know if any of those proposed solutions will work, but I agree with Ford and Lanier that we do have a grave problem ahead of us, as the age of machines need not necessarily be a good age for us humans. I am not talking about a future mythological moment where machines overtake human intelligence. I am talking about our regular computers, information processing on the internet, and machine learning applications already possible today. As a computer scientist, I believe this is an existential question that we need to face, especially those of us working in data science and machine learning applications. A gifted engineer known as Thomas Midgley had once single-handedly managed to almost destroy earth's ecosystem, by creating and spreading CFCs that ate the Ozone layer. I fear that we machine learning researchers can be unwitting successors to Midgley, if we don't pay attention.


BBC Interview of Jared Diamond at the Royal Institution (go to 22:40 for comments on treatment of elderly people)

TED talk by Peter Diamandis, about his optimistic vision for the future:

Introduction to "The World until yesterday" by VSauce, which also talks about Thomas Midgley:

Nick Bostrom talking about existential risk. Please note how the interests of future humans (progeny of the 1% of human survivors) are compared against those of the majority who perish. Existential risk for humanity is a serious issue, but thinking about this has to be tempered with a notion of democracy:

Susan Pinker on her book "The village effect", explaining how engagement in the society is essential for old people to have long fulfilling lives: 

Martin ford on his books "Rise of the Robots: Technology and the Threat of a Jobless Future" and  "The Lights in the Tunnel: Automation, Accelerating Technology and the Economy of the Future"

Jaron Lanier on his book "Who owns the future ?".

Dedication: This post is dedicated to the memory of John and Alicia Nash, who died recently in a tragic accident.