It is people who build machines to work for them. But will there be any work left over for humans when machines become intelligent? One thing is clear: the working environment will change. Looking back at the industrial revolution can give us a hint as to how.
Now it’s Go. First the chess world champion Garry Kasparov loses a game to the computer Deep Blue in 1996. Top participants in the TV quiz show Jeopardy are beaten by IBM’s supercomputer Watson in 2011. And as if that wasn’t enough: we’ve now been outsmarted by artificial intelligence in Go. This Japanese board game, though little-known in the West, is famed in Japan for its vast complexity. But on March 15, 2016 the 33-year-old Go grandmaster Lee Sedol lost to a computer by 4-1.
It’s hard to escape the conclusion: if machines can defeat us at board games, it won’t be long before they can subjugate us in real life.
This, at least, is a scenario that preoccupies both theoretical physicist Steven Hawking and Elon Musk, founder of Tesla and Space-X. They fear that machines will soon be the equals of Homo sapiens; indeed, that they will be able to surpass humans intellectually – a condition known as “technological singularity”. Hawking and Musk see no place for humanity in this future: we will have made ourselves superfluous. If Mark Zuckerberg is to be believed, the end of human superiority may only lie five to ten years from now.
This actually counts as a modest prognosis. It was the logician John McCarthy who first discussed the notion of singularity in 1955. He suggested that if ten good people put their heads together for two months, the problem of artificial intelligence could be solved within a summer holiday.
If two months had indeed sufficed, the world we live in today would be very different indeed. Production would already be completely automated—as would be the entire economy. Contrary to those fears expressed by Hawking and Musk, we would not, however, have been rendered completely superfluous as a result. Humans remain the ultimate end of all economic activity, something our profit-oriented economic system would do well to recall from time to time. But how does someone earn their keep when they are no longer economically productive? What remains for him or her in a world full of intelligent machines?
And when will we have reached this point? McCarthy’s plan for the conquest of artificial intelligence is, after all, 70 years old. Even if current progress in the field of AI is explosive, there are key fields within which the distance between computers and human intelligence has hardly shrunk.
Without doubt, IBM’s Watson’s abilities are impressive, but it remains the case that operations carried out by a supercomputer ultimately have nothing to do with human thought. If one looks more closely at Watson’s process design, it becomes clear that the computer doesn’t even need to understand the questions put to it in order to generate its answers. Watson employs a syntactic routine to deconstruct texts and extract their central terms before searching its database for synonyms. Using the resulting word clusters, it then searches a 100 gigabyte lexical library for similar accumulations of terms, filtering out the most common sentence constructions from the entries returned by its query. Using these, it generates its answers. This makes Watson excel at answering questions on Jeopardy, but still it remains within the realm of classic data processing. It bears no similarity to the semantic understanding on which human thought is based. People make spontaneous associations; they do not continuously check their memories for certain constellations of words.
This exemplifies the fact that in areas of genuinely human ability, such as understanding the world around them, machines are still not at home; conversely, humans will never outperform a simple pocket calculator in the realms of logic, calculation, and data-processing.
Thus, despite our fears, we need not worry about being overtaken by AI at any point soon. It has already lapped us, at least in the fields for which its architecture is designed: formal operations. But the triumph of the algorithm will, at least for now, have to remain limited to this domain. Even accelerating increases in processing power can do little to change this fact. Only a fundamentally new technology, capable of operating associatively rather than formally, would have a chance at developing a truly human-like artificial intelligence. Yet there is no sign of such a development on the horizon.
Instead, we should free AI research from the unreasonable demand that it reproduce or simulate human intelligence. Because man and machine posses different competencies, it makes more sense to work towards a division of labour in which human and artificial intelligence complement each other.
To achieve this, we will need to create precisely tailored niches for machines to fill. Winning at board games is maybe exciting, but ultimately meaningless. It is only humans who can say what is meaningful. Delivering meaning to machines will become one of the central tasks people will tackle in the future.
“Claims processor” sounds abstract, but it already exists as a job. Contrary to the cliché, tax advisors or insurance claims processors do not simply follow a rote formula. Instead, their job is to translate real-world occurrences into the formal language of contracts, laws and software. It is exactly this work of formalising meaningful associations and contexts that will continue to belong to the tasks for which humans are indispensable.
On the other hand, jobs in fields where computers can apply their unsurpassable precision and speed are at risk, especially in logistics and middle management.
Both “Uber” and Amazon’s “Mechanical Turk” (MTurk) already operate according to this principle. They replace middle management with software that brings customers into immediate contact with those providing services for which they are willing to pay. Uber connects drivers and passengers, while MTurk recruits human workers for tasks that cannot yet be performed by computers, such as translation or data entry. However, because of their lack of political organization, click workers and Uber drivers are wholly at the mercy of the cold economic calculus of algorithms. The Uber algorithm sets alone, without human intervention, the price of a journey based purely on the present state of the local Uber market. If not programming or marketing new technologies, humans are reduced to precarious jobs involving menial labour and little prospect for advancement.
With the introduction of algorithms into economic competition, every margin will be automatically whittled down to an absolute minimum, including money spent on wages. In a competition between algorithms, no quarter can be given, and no room is left for profit.
When it comes to way returns are distributed, automation also fits poorly with our current economic system. If the production of goods can be disaggregated into a series of logistical processes consisting of manufacture and resource management, then it is only a matter of time before it is nearly completely automated. This would spare workers all manner of hard manual labour, but presents us with three important questions:
1 – How will automatically generated economic yields be distributed?
2 – How will we still be able to earn money?
3 – Why should we have to pay for a commodity produced without the input of human labour using money we have earned ourselves?
None of these questions can be satisfactorily answered within the free-market economic paradigm that currently reigns.
One thing is clear: people’s livelihood can no longer remain tied to their economic productivity.
This is not, in fact, a new development. In the industrial age, machines have inserted themselves ever more thoroughly between human labour and its products, alienating people from production to an ever-greater degree. In response to their being rendered replaceable as the operators of machines, workers began to organize. They brought about the modern labour movement and the introduction of social insurance programs. Since the Industrial Revolution, workers’ pay has been increasingly decoupled from the economic productivity of their work, right up until the present and the minimum social benefits guaranteed by the state.
A logical next step would be to introduce an unconditional basic income, which would finally complete the detachment of livelihood from economic work. From a cultural-historical perspective, this actually seems plausible. The question is, will it solve the social problems presented by automation? Will people of their own volition seek meaningful work if their income is already guaranteed?
It seems clear that for most people work is more than just making money. Work is part of one’s identity, one’s sense of self-worth, and place in society. It is a basic human need.
The good news is that there will still be enough work to be done in an automated world. As long as there are problems, people will need to work to solve them. The difference is that in future, these problems will lie in domains not subject to automation. In culture, education or social work, for example. Indeed, precisely those areas of work that, because of their lack of profitability, are organized by the public sector.
The biggest challenge for the automated society will thus consist of turning all the problems that computers can’t solve into meaningful and motivating work for humanity.
Latest posts by Mads Pankow (see all)
- On robots and class struggle: Are we being replaced by machines? - 11. July 2017