While highlighting current AI limitations, it is still being argued that the development of increasingly cognitive AI systems will continue in the coming decades, possibly even at an accelerated rate. If so, the increasing cognitive ability of these systems will allow such systems, inclusive of robotic extensions, to operate in a more autonomous fashion. Of course, there is also an inference in autonomous action that might suggest that humanity may not always be in full control or understand what actions are being carried out or to what purpose.
As has been argued elsewhere, AI does not have to be ‘strong AGI’ to start to trigger profound changes in human society, although the transition towards this ‘brave new world’ will undoubtedly be subject to many social, economic and political constraints and caveats. As such, developments may take longer than some futurists are predicting. However, before jumping into any further speculation about the future of AI and robotic automation, it might be worth quickly reflecting on how humanity has achieved its dominant position in the natural world, especially in the last 500 years or so. We might readily accept that much of the initial progress can be attributed to the industrial revolution and the development of the steam-engine followed by an increasing use of fossil fuels, which was then extended by the widespread use of electricity. As such, we might reasonably conclude that the expansion of the human ecosystem was in no short measure dependent on energy. Within this framework of increasing energy usage, more and more people have been elevated above mere survival, although this change required many people to abandon the traditional rural workplace to live and work in urbanised cities. As such, we might perceive a price associated with this technical progress, which resulted in people living in ever-larger urban structures in ever-greater concentrations.
Note: In 2016, it was estimated that 54% of the global population live in urbanised areas of some description. By 2030, urbanisation is projected to house 60% of the global population, where 1 in 3 people will live in cities with a population of over 500,000. In 2016, there were 512 cities with populations in excess of 1 million inhabitants, which by 2030 is projected to increase to over 650 cities. Likewise, in 2016, there were 31 cities with populations in excess of 10 million, which is projected to rise to 41 by 2030.
Today, we might readily understand how dependent we have all become on energy, but possibly less aware of our direct and indirect dependency on computer developments, which are now necessary to support the complexity of day-to-day operations in most modern cities. As such, the lives of 60% of the global population, projected to be living in cities by 2030, may now be dependent to some degree on both energy and computers. While the idea of a ‘cognitive evolution’ has been linked to the start of the European Renaissance, some 500 years ago, it appears that this cognitive evolution has accelerated exponentially with the development of computer systems over the last 50 years. However, we might realise that the potential for further developments in cognitive AI may lead to even more profound change than most of us may care to accept.
How might we summarise the potential impact of AI and robotic automation?
From a historical perspective, domesticated farm animals started to replace the need for some human manual labour thousands of years ago. However, this replacement of human manual labour became more obvious during the industrial revolution with the invention of the steam-engine, which subsequently led to increasing mechanisation that started to reduce the requirement for manual labour even further. For example, the graph blow shows how employment in US farming fell from 90% in 1800 to just 2.6% in 2000.
Likewise, before the invention of machines like the spinning jenny , cotton weaving was a cottage industry, in the literal sense of the word, where weavers worked in their own cottages to produce cloth. These jobs were effectively destroyed by mechanisation when powered looms were first developed in 1784 and developed over the next 50 years, such that the jobs of even skilled weavers were also replaced. As such, degrees of automation that can replace humans in the workplace is not necessarily a completely new phenomenon, although we might realise that AI and robotic automation may have far more serious implications on global employment in the future. For AI automation has the potential to affect a much wider profile of the global population across the entire spectrum of employment, i.e. blue to white collar professions. As such, the question that everybody might now need to consider is:
How quickly might AI and robotic automation affect me?
Today, we might recognise that we are already living in a world where even the current generation of AI and robotic automation may already perform a range of routine tasks, i.e. jobs, better and more cheaply than most, if not all humans. As such, it is not unreasonable to assume that these systems will simply become capable of doing evermore tasks, i.e. jobs, as AI developments continue into the 21st century. Even in terms of the generalised cognitive AI architecture previously outlined, we might recognise that deep-learning neural networks are becoming increasingly capable of processing images, e.g. face-recognition, as well as converting human voice-to-text and written text-to-voice. As such, cognitive AI is already well on the way to supporting a human-like interface that may exceed any original expectations of the Turing Test. Of course, if this interface is then backed up by expert systems using a machine learning paradigm, these systems might be capable of accessing and processing more information in seconds than a human person in an entire life-time. On this basis, we might begin to recognise the scope of the danger that humanity may have to confront in terms of future employment.
But why would humanity allow this to happen?
Historically, it has been shown that automation, in almost any form, has led to an increase in productivity, which in turn has supported further economic growth. However, any implied growth in the various sectors of the national economy has also to be weighed against the government’s balance sheets, e.g. welfare costs of the unemployed. So while earlier forms of automation did lead to large-scale unemployment in certain industries, new industries invariably offset and even outweighed any earlier job losses over time. However, whether this will be the case in respect to AI and robotic automation might now have to be seriously questioned. This said, many businesses may simply have to pursue the cost and productivity benefits being offered by AI automation, not necessarily as a matter of choice, but rather as a matter of survival in a competitive world.
What timescales are involved in the AI-led industrial revolution?
Given the description of cognitive AI outlined, automation may be both incremental and transitional, but where a continuously expanding process comes to affect the very nature of employment over many decades to come. For example, even partial automation in the early stages might result in an increase in part-time employment, while longer term predictions suggest that between 25-50% of all current jobs may be at risk to AI and robotic automation by 2050. As such, it is possible that the impact on employment, and incomes, may start to be felt within a much shorter timeframe and only get worse over time for many people. How this might affect different economies may depend on the demographics of the population, e.g. if the overall population is in decline, then the working population may also be smaller causing a shift in the age demographics of the population, such that automation may be seen as a benefit. Of course, such a benefit might not be so obvious in many developing economies, where population growth is still a factor, which might then simply increase the pressure of migration back from the developing economies towards the developed economies. However, even now, the incentive for such migrations may not result in any meaningful employment, which automation may only make worse, such that many people may simply migrate as a means of survival in the hope of accessing the welfare systems of the developed economies.
What is the initial scope of robotic automation?
It is highly probable that robotic automation will continue along historic lines in all manufacturing industries. However, if robotic automation can continue to reduce the cost of human employees as a total of overall costs, manufacturing that was once outsourced to the developing economies may start to return to the developed economies as a matter of preference. Likewise, the combination of AI and robotic automation may come to revolutionise the idea of ‘just-in-time manufacturing’, such that the costs can be more easily adjusted to product demand, which may then be adapted to higher degrees of bespoke manufacturing to best meet both market and individual demands. In this respect, manufacturing industries represent an environment where robotic process automation (RPA) can be most easily applied to analyse specific and repetitive tasks as well as providing an environment where human activity can be restricted for safety reasons. Of course, adding increased cognitive AI abilities to robotic systems will allow them to be deployed in ever-wider environments.
What jobs are ultimately at risk from both AI and robotic automation?
The pace and extent of this form of intelligent automation will vary by geography and economy over time. However, it is probable that jobs in the developed economies will be the first to be affected if businesses, both in manufacturing and services, come under increasing pressure to reduce costs in order to remain competitive, especially when operating within global markets. We might attempt some general quantification of the types of jobs at risk by both AI and robotic automation in terms of the next chart, although almost any job might be subject to some degree of automation, such that full-time employment may be at risk in almost all sectors.
Despite the suggestion of the chart above, it is not clear that robotic developments will threaten all manual and skilled blue-collar jobs outside certain controlled environment. For example, domestic plumbers have to deal with a multitude of jobs and different physical environments, which may be impractical for any robot to carry out in the near future. However, in contrast, some highly skilled professional jobs, such as medical general practitioners, might be more seriously at risk than suggested as cognitive AI might be far superior in the process of a technical diagnosis, while becoming increasingly capable of providing a human-like interface, inclusive of facial recognition of emotions and a potential ability to converse in multiple languages.
Note: Since 1950, health care spending in the UK has increased dramatically in real terms. In the post-war period, there has been a sharp rise in public health care spending as a % of GDP, from 3% in 1960 to 7.8% in 2010. The US spends more than $3 trillion a year on healthcare, equating to nearly $10,000 per person. Overall spending rose 5.8% in 2015, which was faster than the pace of inflation or wage growth. Clearly, either politicians reset the public expectation for ever improving healthcare, which is never popular, or they must find new ways to dramatically reduce costs.
In general, it seems likely that we will see increasing numbers of jobs being subject to varying degrees of AI automation, not only to reduce cost, but because AI may be better at doing many of these jobs. As such, human involvement may be reduced in scope to ever fewer tasks that AI is not initially capable of doing well or are simply retained as contingency back-up for when things go wrong. However, such predictions have to also be put into some wider context that may slow the rate of AI automation in practice, which we might summarise as follows:
- Much of the sophistication of AI and robotic automation being outlined
has still to be developed, although it might reasonably be argued that
the discussion has been based on a realistic extrapolation of current
technology and not just science fiction.
- Another important factor, regarding the rate of adoption of AI
automation is the possibility of increasing development and deployment
costs when compared against the possibility of falling human costs,
especially if subject to an increasing supply of the unemployed in a
falling jobs market.
- The details of each business case may need to provide clear benefits,
possibly both economic and political in scope. While reduced employee
cost might be an obvious factor, businesses might also be attracted
to the fact that automation might better support product quality on
a 24/7 basis.
- Finally, political regulation may be imposed on the rate and scope of AI automation adoption due to growing public concerns. However, this option might prove to be difficult for any government, in isolation, if a business sector is dominated by multinational corporations who can simply move their operations to another country if governments attempt to impose too much regulation on the market.
If we assume the technology hype surrounding AI is now justified and that there is a real-world business case for adopting AI and robotic automation, the issue for governments may become increasingly problematic, both in terms of public demand for job security and income protection plus the potential impact of increasing unemployment welfare costs on the nation’s balance sheet. For governments may, on one hand, have to encourage investment and provide market incentive that supports economic growth predicated on AI productivity and further technology innovation; while, on the other hand, attempting to evolve policies that help its population and institutions to adapt to fundamental changes in the nature of long-term employment. Such change will most likely have to include a major rethink of the governments education strategy as well as a radical change in welfare support with additional safety nets for those simply unable to adapt to the brave new world that confronts them.
Note: Today, many governments and think-tanks are considering the idea of a basic income , also called an income guarantee, citizen's income, unconditional income, universal income, living stipend. Irrespective of the name, it typically describes a form of welfare payment in which all citizens of ‘working’ age receive a regular and liveable ‘wage’ from the government, even though there is no requirement to work or even to look for work. However, many question how governments could afford to support such systems unless they are a net beneficiary of considerable GDP growth, possibly from AI and robotic automation.
As has been argued at several points, not everybody will be a beneficiary of AI and robotic developments. In fact, it might even be suggested that in a world in which AI and robotics could come to replace much of humanity in the workplace, the inequality of the haves and have-nots being outlined may only get worse. Of course, it does not necessarily have to be this way and possibly this brave new world may not appear as quickly as some futurists are predicting, but the idea that the potential benefits associated with AI and robotics will make everybody’s lives better is possibly overly naïve, such that the dangers have to be highlighted. However, many will contest the apparent negativity being suggested by arguing that AI and robotic automation will not lead to increased unemployment over the longer term, as other jobs will simply emerge surrounding new industries enabled by future technical developments.
Note: In 2016, the International Labour Organization predicted global unemployment would rise by about 2.3 million to 199.4 million, and that 1.1 million will be added to the global count in 2017, taking joblessness to more than 200 million for the first time on record. This is at a time when the global population is increasing by over 75 million every year of which we might assume some 20-30% would ideally be employed.
While this discussion has acknowledged that developments in AI and robotics holds out the promise of substantial benefits to business and society, it has also been highlighted that such developments come with the potential for malicious misuse and unintended consequences. Today, we might already highlight this danger in the growing developments within the theatre of cyber-warfare, where participants are increasingly both state-sponsored and criminal. Today, daily battles are constantly being fought in which various aspect of the digital infrastructure, i.e. social, economic and political, are being attacked and defended. In future, it is not unreasonable to assume that these battles will increasingly involve AI directed action, which can utilise millions of data points, such that it may eventually endanger the growing idea of an Internet of Things. This concern is self-evident in the following quote by Derek Manky, a global security strategist:
“In the coming year we expect to see malware designed with adaptive, success-based learning to improve the success and efficacy of attacks. This new generation of malware will be situation-aware, meaning that it will understand the environment it is in and make calculated decisions about what to do next. In many ways, it will begin to behave like a human attacker: performing reconnaissance, identifying targets, choosing methods of attack, and intelligently evading detection.”
Today, few who use the Internet can be unaware of the increasing danger of crafted emails that contain malware that seeks to gain or destroy personal information. In the future, AI with a human natural-language interface will have the ability to extend malware attacks onto the voice interface of smart phones. Others believe that next-generation AI is already in developments which may come to restrict the general use of the Internet, both in terms of on-line shopping and as a valuable source of information. These developments include use of ‘polymorphic malware’ that can infect part of the network, then change or automatically deletes itself, such that it cannot be traced back to any source. So while the 2017 Asilomar conference may believe it has defined 23 principles underpinning the beneficial development of AI, others may simply escalate a cyber-AI arms race in pursuit of economic or political gain. Unfortunately, in this respect, AI and robotic developments may simply reflect all the problems that might be associated with the ‘human condition’.