Artificial Intelligence and the Future of Work

Robot from The TerminatorArtificial intelligence – and the potential that such intelligence could be a danger to humanity – has been a staple of literary and cinematic horror for decades. Fritz Lang’s Metropolis (1927) featured a robot disguised as a young woman, who ended up toppling a city, a theme most recently captured in the remake of Battlestar Galactica with the very human-seeming Cylons, robots so sophisticated that even they didn’t know they were robots. The Colossus: The Forbin Project andWargames both featured self-aware global computer systems that took over the nuclear missiles of both the US and Russia.

HAL from 2001 – A Space Odyssey embodied the very concept of the sapient but insane computer interface as it worked to sabotage a manned mission to Saturn, while GladOS from the Portal games was a brilliant parody featuring a “female” computer determined to test her subjects to death. The whole premise of James Cameron’s Terminator series was the notion that the networks of today had become the SkyNet of tomorrow, a global AI that in turn controlled robots bent on human annihilation.

Artificial Intelligence vs/+ Human Intelligence

Lately, perhaps because of the rise of Big Data, Data Science and the emergence of consumer facing robotics, people are beginning to ask questions in their daily lives that traditionally have been the realm of philosophers and science fiction authors, most notably: What happens when computers become more intelligent than us?

It’s a valid question, and one that’s very difficult to handwave away. Our cars are becoming more intelligent. So are our houses, our places of work, our wireless telephones (now almost universally referred to as “smart phones”). Industry pundits and manufacturers are all praising the coming Internet of Things (or IoT) to the extent that our coffee machines will soon be so smart that they can prepare and brew your favorite cup of coffee, fetch the newspaper, and prepare a full analysis on the stock market before you even finish getting dressed in the morning. That it can also find all primes up to one sextillion in the background while brewing that cup of java goes without saying.

We get twitchy when our car gets better grades than our honor student kid (and when we finally get around to developing inexpensive flexible screens, expect smart bumper stickers to proclaim this fact loudly and in fully animated 64 bit graphics). Humans are proud of their intelligence, but in the next few years (or maybe in the last few, jury’s still out on that one) we’re going to end up creating systems that are in fact smarter than us in many meaningful ways.

Humans get all hung up about intelligence, but when you get right down to it, we’ve been developing distributed intelligence for a long, long time. Human intelligence isn’t actually one thing – it is a system of “computers” that each do specialized tasks. The hippocampus regulates semi-autonomous functions like breathing, and very basic fight or flight reactions. The nervous system handles sensory aggregation and transmission, with gross and fine motor control coming from the cerebellum and parts of the cerebral cortex. Emotions are modulated in the older cerebrum. Memory occurs primarily in the outer gray matter of the brain, while the prefrontal lobes of the cerebrum handle a lot of what we think of as the artifacts of intelligence – pattern and language recognition, tactical and strategic planning, our ability to model the world, and our self awareness or consciousness.

Consciousness in particular is considered a very important aspect of intelligence, but in many ways, consciousness is simply another “routine”. It serves to create a feedback loop between the internal and external representation of an individual, and may have a comparatively recent origin. Some scholars have argued that human consciousness may in fact have only emerged in the last 10,000 years or so, though it is more likely that it emerged around 150,000 years in the past, While either of those time spans might seem incredibly long, in terms of evolution, these are eyeblinks. 10,000 years works out to about 500 generations.

Yet for all that, consciousness did not occur overnight either. When the giant space monolith appeared, Ogg the ape-man did not suddenly become Ogg the metrosexual. Instead, self-awareness built up in stages – awareness of self (internal system regulation), awareness of relationships (connectivity to local networks), awareness of the gods (development of protocols for determining standardized behavior) indirect or cultural awareness (access to institutional knowledge gained from external sources) and so forth.  There is some evidence to support the idea that the human brain has altered itself to make each increase in consciousness more efficient, but this may also just be due to the plasticity of the brain (firmware adaptation rather than either hardware or software adaption).

Computers are already computationally much faster than we are, if only because they can be networked. Human civilization is an example of networked intelligences, albeit a fairly inefficient form. Institutional knowledge can be retained by a community easier than it can an individual. Human culture exists to provide such knowledge, to insure that there is redundancy of critical information for survival and a mechanism for transmission of that information from one generation to the next. To do that, we had to develop a complex set of signals called language, and with each iteration of that language, we refined the protocols, added more and more complex symbols until we were dealing with abstractions such as “love”, “blue” and “forty two”.

That’s a big part of the reason that so much artificial intelligence research actually focuses on the ability to understand and work with language. Humans have a few really hard concepts that are so hard that we tend not to realize how difficult they are to encode. “Here”, “there”, “me”, “you”, “now”, “then” – these are all context expressions, terms that require an understanding of both self- and external- awareness. These contextual terms and their associated concepts are now coming on line – a smart phone, for instance, knows where it is better than you do, but this is only know percolating into its concept of awareness.

These are harder even than emotions to express. Indeed, computers already have a pretty good handle on “content”, “stressed”, “fear”, “anger” and other emotion-oriented words, because they generally rely on self-monitoring, something that is actually pretty easy for a computer to do. Many can even increasingly be set up to handle self-motivation – when a sensor in a car indicates a bad fuel mix problem, a car even today can warn its owner that it is feeling “sick”. and should go to the car doctor (aka mechanic).  It’s not that big a leap to the day where a self-navigating car can ask its owner if it could drive to the car doctor itself, so that it can talk to the car doctor’s AI to let it know EXACTLY what the problem is – certainly no more than twenty years, maybe within ten.

Add into this the fact that over that same period, everything will be connected to the Internet, which is to these emerging AIs what culture is to humans. This suggests that machine AIs will likely also go through a god phase – an awareness of computer spirits and daemons that are the virtual simulcra of nature spirits, ancestor worship and so forth, but there is also one fundamental difference between human and machine intelligence. Networked computers are increasingly virtual constructs – they exist as distinct entities only be convention. Human beings cannot duplicate themselves (they can clone themselves (or soon will), but these are duplicates only at the operating system level, not temporally or with regard to experience (data acquisition and processing). Computers can, and so the notion of identity (and hence self-awareness) will become far more difficult for a computer to evolve to as a state.

This, then is the crux of why even self-aware AIs will be vastly different from humans in many ways. Computer awareness will be very different from human awareness. They will be effectively omniscient, but a significant portion of their processing will be involved in making sense of this information. AIs will have no real sense of physical “there”-ness, but will instead move from one sensor nexus (robot) to another, or even move through several such nexuses simultaneously. Such robots will have autonomous processing capability, but will be able to get sensory information and analysis from both other robots and from cloud based AIs, and can also offload some of their processing temporarily as more local processing is needed for specialized tasks.

How Artificial Intelligence Impacts Jobs

This can actually be used to extrapolate how AIs and robots will ultimately shape human culture, for both good and bad.

  • Robots = AI + Sensors + Motors.  When robots are discussed here, the tendency is to think of humanoid robots. A robot in the purest sense is simply a basic artificial intelligence coupled with sensors for establishing local context and motor driven actuators for effecting the external world. A modern thermostat is a robot. Your car is a robot. Your smartphone is a robot (think phone cameras, which have a limited number of mechanical parts). More traditional robots are on their way, but there is very little distinction between robots and the Internet of Things – ultimately, it will be the AI – the brains behind those robots -that is the critical component
  • Many Existing Human Jobs Are Going Away. AIs and robots will have the capability of replacing all known jobs within twenty years. This doesn’t mean that all jobs WILL be replaced, but the capability will exist. As this happens, this will collapse the current global economy, because it does not have the ability to adapt fast enough to shift from an economy based upon work compensation (and hence participation in the economy) to one based upon other means of wealth distribution.
  • Resource Limitations Still Exist. This has some interesting implications. First, while computation requires energy (and produces heat and waste artifacts), we are not actually moving into an age of plenty so much as an age where efficiency of resource usage becomes paramount. Intelligent systems will be able to better distribute water to various groups within areas like the American Southwest, but it also mean that AIs will play an increasing role in politics to determine w
  • Employment Will Become Virtual. People will spend more and more time within virtual communities – games, simulations, social environments, shared worlds and so forth, and it is through participation in (and extension of) these environments that people will find employment. Virtual world product designers (design = manufacturing in the virtual world), story tellers, optimizers (people who increase participation in various platforms), analysts, fashionistas, reviewers, even the occasional barbarian or wizard for hire will be typical of these new positions,
  • The Metamorphosis of Money. Virtual worlds mint virtual money. As standards for interoperability of virtual environments emerge, one of the biggest there will be tracking stores of value – currencies. BitCoin and similar cryptographic money may become the equivalent of real world gold – an inviolate base store of value that will in turn back other currencies. This means that such virtual currencies become exchangeable at real world rates, and it also means, as more and more transactions take place virtually, that over time such currencies may end up replacing government currencies.
  • The Office Will Disappear.  X as a platform type solutions will replace most traditional office work that’s not critical to differentiating a company. This process is already underway, but it is likely that more sophisticated platforms will increasingly handle analysis and lower level decision making will actually take place in code, utilizing business intelligent systems that will be able to both pull together the requisite data and determine, based upon this, what the best decision making strategies are to maximize profits, market penetration and similar variables.  While the office as client meeting place may remain, one of the biggest impacts upon this will be commercial real estate, with significant swaths of the CRE market going permanently empty over the next forty years.
  • Most AIs will be Assistive Agents.  Even self-aware AIs will likely be assistive  –  they will extend the reach and scope of human beings, rather than being wholly autonomous, but they will also have the potential to become significant crutches. As educational systems at all levels become more virtualized, the distinction between games, education and (unfortunately) advertising will blur significantly. It is likely that people’s ability to memorize will diminish in favor of learning more sophisticated information management strategies.
  • Agents and Avatars. An agent is an online AI that both provides information and coordinates activities. An avatar is an interactive representation either of a real person or an autonomous agent (and to a significant degree, within the virtual world there will be no real way to distinguish which is which). Both require the use of AIs, with the agent’s focus primarily upon search and interfaces (most likely interacting with a series of intelligent data stores), while avatars use AIs to create a convincing media presence appropriate to the environment. While there is some overlap now (and basic avatars and agents exist in various online game environments) there is little as yet to coordinate agents or avatars across different environments. This coordination will likely be an ongoing effort over the next 15-20 years.
  • Offline Will Be the New Frontier. Ironically, the more that information and services become automated, the greater will be the desire to move more or less completely off the grid. There will be a growing gulf between the digitally connected and the unconnected, what has been described as the shift from Blue vs. Red (Liberalism vs Conservativism) to Black vs. Green – those who are heavily invested in technology vs. those who do not want to be (or cannot afford to be) connected at all. Ironically, artisanal jobs, ones that could be done via automation but aren’t because they represent human craftsmanship, will actually rise. It is entirely possible that by 2050 there will be more active blacksmiths in the United States than there were in 1880, as one example.

The upshot of all of this is simple. The definition of work as it has existed for the last century is going the way of the dodo. Automation in its various forms will be what drives it to extinction, because there are very few jobs that a human being can do that a machine won’t be able to do better within the next few decades.

The End of Work (As We Know It)

The current economy is based upon principles of work establish a century ago, especially in the wage sector Today, you get paid a certain amount of money an hour to do work when necessary, and to be available to an employer exclusively. The amount you get is determined both by the scarcity of skill sets and the degree to which your actions can be quantifiable to a bottom line. With those wages, you then buy those things necessary to live – housing, food, clothing, transportation, entertainment and so forth, which in turn is the principal engine that drives the economy (far more than investment income).

Artificial Intelligence (including robots) undercuts all of that. It does not completely remove humans from the equation, but it does render about 2/3 of all people currently in the workforce obsolete, and more pointedly, it plays havoc with wage income. For those people who are able to stay current with technology (and to develop highly specialized skills that they are updating continuously) the wages can be a couple of orders magnitude higher than it would be for unskilled labor, which also puts them into a position to better accumulate enough capital to enter the rentier economy, but that’s a small percentage of the total workforce.

To put it baldly, the current economy will irrevocably crash in the next few decades, due to the distorting effect of automation.  Automation by its very nature benefits the rentier economy far more than it does the wage economy. Automation reduces the need for people to perform certain forms of work. It benefits investors attempting to get a more accurate picture of the investment landscape. The most powerful AIs on the planet perform the equivalent of penny shaving billions of times a second by  performing micro-trades that have no direct impact on price valuation, but instead simply take advantage of network effects. In other words, AIs magnify the effects of transactions dramatically, and those in the rentier economy are best positioned to take advantage of this.

Short term solutions that ameliorate the problem are politically volatile but are worth considering. Eliminating taxes on wage income and dramatically increasing transactional taxes on rentier income (capital gains) would help for a while. By eliminating taxes on wage income, people would receive the equivalent of a 20-30% boost in spendable income – assuming they have a wage income in the first place. Taxing capital gains would slow investment, but at the current time there’s already too much investment money chasing too few legitimate business niches, meaning that much of the money being invested is essential funding overly risky businesses with too little likely return. Taxing such capital gains would also better support those people that have lost their jobs due to automation.

However, longer term, even this is a stop-gap. Policy makers need to start asking hard questions now. The biggest seems like science fiction, but it is screaming towards us: if automation eliminated all jobs, how would the economy function? What would such an economy look like? Are there ways of insuring that people have sufficient income to meet their basic needs while at the same time rewarding those values that Americans in particular have come to prize: entrepreneurialism, innovation, independence, creativity and integrity?

Ultimately, the vision of robot domination will not come from Terminator style robots designed to eliminate humanity from the planet. Instead, it will come from policy makers at all levels – corporate and governmental – failing to recognize that the effects of automation need to be weighed against the needs of humanity at a fundamental economic level.

Any sufficiently large group of people, freed from the lower tiers of Maslow’s pyramid, will ultimately specialize into jobs that they do because they enjoy doing them, whether they can be automated or not (for an example of that, look at the retirement economy – a generation that does not longer needs to work for a living is in fact investing more of their time, the ultimate resource, into everything from repairing crumbling schools and assisting teachers to building up charities for eradicating disease or helping their children raise their grandchildren. These are needful things, and that generation is making a difference in ways that previous generations couldn’t. This represents the potential of what AI could bring – freeing people from the need to make money helping other people make money so they can concentrate instead on those things that really do matter – family, community, education, health, culture, intellectual and spiritual growth and so forth.

What are your thoughts?

Kurt Cagle is Principal Evangelist for Data Science with Avalon Consulting, LLC

Kurt Cagle About Kurt Cagle

Kurt Cagle is the Principal Evangelist for Semantic Technology with Avalon Consulting, LLC, and has designed information strategies for Fortune 500 companies, universities and Federal and State Agencies. He is currently completing a book on HTML5 Scalable Vector Graphics for O'Reilly Media.

Leave a Comment