On AI and Robotics

Modelling human health

Alongside applications that involve using AI and robotic technologies to improve medical practices, the systems are also being deployed in research settings to improve our understanding of the human body. Models of human physiology and biological processes enable the acquisition of new insights into health, and the effectiveness of medical interventions.

For example, researchers at The University of Manchester Advanced Processor Technologies (APT) Research Group are assisting in the development of a new computer architecture inspired by the design and workings of the brain in the SpiNNaker project (an acronym for Spiking Neural Network architecture). They aim to build a synthetic model of the human brain from a million embedded Acorn RISC Machines (known as ARM chips) to build a massively parallel computing platform.

The SpiNNaker architecture is made up of nodes communicating simple messages, known as spikes, that are inherently unreliable. This innovative approach is enabling researchers to discover powerful new principles of massively parallel computation.

The finalised model will be able to demonstrate around 1% of the capacity of a human brain (which has approximately the same computing power as the brains of ten mice) and is built in a different manner to traditional supercomputers due to the biological basis of the design. There are many challenges to overcome when modelling complex structures that occur in the human body, particularly the brain, as shown in this video:

It is this very structural complexity that will enable the SpiNNaker project to imitate aspects of brain function. This will enable simulated ‘drugs’ to be applied to the model so their effects may be studied, resulting in a model that is both based on and contributes to modern neuroscience.

One of the principal researchers on the project Professor Steve Furber, co-designer of the ARM chip, and ICL Professor of Computer Engineering at The University of Manchester. He explains the importance of initiatives such as the SpiNNaker project:

"Medicine is already a very active area for computer modelling research, from designer drugs (where modelling is at molecular level) through to aphasia rehabilitation (using computer models to determine which therapy is most likely to be effective for a given patient). "

"Medicine has moved a long way from empirical experiments with random drug-candidate chemicals towards disease understanding, in which computer models play a role, leading to drug specification and design. Of course, there is still a lot of empirical testing required, with unexpected outcomes such as Viagra, which was originally designed for soft-muscle treatments. It was rather ineffective for the intended use, but then some interesting side-effects were reported… "

The enormous hardware and software challenges faced by the SpiNNaker team also show the difficulty involved with creating truly strong AI. Professor Furber explains that although our brains run on approximately the same power as a small lightbulb, current technology is nowhere near as efficient for building artificial alternatives;

As a discipline healthcare delivery is closely aligned with research, and AI and robotics are finding several applications in this area. In addition, the technologies are also being used to improve aspects of the research process itself.

One of the major issues debated in this area is how it may be possible to legally protect knowledge developed by a machine. Patent law in most countries usually refers to discoveries by humans, and certain adjustments may be needed to account for the knowledge developed by machine – though this is yet to be tested to any great extent in court.

This is a challenge not just in research, but also in the business operations of the future. As a recent PwC report states:

Although AI is likely to bring significant new capabilities to businesses, companies will need understand how to protect the value that newly developed knowledge will bring. In addition, in regulatory terms the learning capabilities of AI systems is also a challenge. In developing effective regulation there is fundamental problem with any deployed system that continues to learn, as its future behaviour may be hard, if not impossible, to predict.

Policymakers and regulators will need to be careful to develop oversight rules that balance ensuring the benefits of a system’s learning abilities are exploited (i.e. that there is flexibility for it to learn, test and deploy new strategies), with protecting against the negative effects of future unpredictable behaviour as a result of this learning (i.e. it should be possible to limit the system’s flexibility to implement what it has learned when necessary).

In order to facilitate this, policymakers should proactively engage with experts and try to understand as much as possible about how such technologies will work in order to create better regulation. And these challenges will soon be faced by regulators across many sectors; not just in research, but also in industry, as intelligent systems continue to revolutionise working environments.

AI and robotics – promises and risk management

By Professor Vasco Castela

Sooner or later we will build machines with human-level intelligence. There is no compelling reason to believe that this is impossible; that we cannot create a machine that both behaves intelligently, and even has a consciousness, along with the emotions that come with it. After all, we already know that machines can think and feel, for humans clearly are (biological) machines, as 18th century physician and philosopher La Mettrie pointed out. It would be strange indeed if it turned out that intelligence and consciousness can arise by natural selection alone, but cannot ever be built by human beings, who can harness the power of both design and natural selection, with the help of computer models that implement evolutionary processes.

Human-level AI will likely exist, but not this year, and perhaps not this century. Since the 1950s, there have always been experts claiming this is just around the corner, but it is likely that we have really just started, and that there is much to be done. A common metaphor used to describe the situation is that we’re making good progress climbing a tree on our way to the moon – meaning both that we’re still very far from our goal, and that although we are advancing, there is no guarantee that we can keep going along our current path, and may have to try something altogether different.

New approaches to AI and robotics are paying off

Even though we still very far from achieving human-level AI, that doesn’t mean that we haven’t had much progress in the field. Interestingly, much of the progress achieved in robotics in the last few decades is connected to becoming more humble in our aims, with many researchers abandoning the goal of directly aiming for human-level AI. This has happened in two ways. First, instead of aiming to build a mind, which involves making massive assumptions on how we think, we are now trying to build a learning brain. We have been moving from strictly symbolic computer programming, where we directly and explicitly issue orders to the machine (in effect simulating conscious reasoning), to the increased use of artificial neural networks, where the machine does a lot of its own learning, and the programmer does not even fully understand why it does what it does. Secondly, we have moved to aiming for insect or animal-level cognitive skills, both as an end in itself and as a basis for higher cognition.

The idea that we should start with a foundation of basic bodily cognitive skills such as navigation and obstacle avoidance - of the kind we find in insects and animals - before we can consider aiming for human-level general problem solving, was outlined in a 1990 paper by MIT’s Rodney Brooks entitled Elephants Can’t Play Chess. Rodney Brooks later co-founded robotics company iRobot, where he developed, among other products, the autonomous vacuum cleaner Roomba. The navigational skills and speed of these relatively common household robots may seem ordinary now, but their performance was unthinkable in the early 1980s, where autonomous robots took hours to achieve basic tasks.

Crucially, the burst of progress in robotics that started in the 90s had little to do with Moore’s law of ever faster computing speeds, and it was more connected with the fact that this new approach to robotics involved the building of mechanical insects that didn’t bother with understanding or mapping the world, but merely responded to it as appropriate. It turned out that a clever artificial insect could accomplish much more than an extremely crude artificial human.

Recent advances by companies such as Boston Dynamics (e.g. the Spot Mini and Atlas models) demonstrate the next steps in these advanced physical capabilities, and although they cannot chat or solve any formal problems, their uncanny cognitive sophistication allows them to successfully navigate the real and unpredictable physical world, while older robots could only get around in synthetic environments, known as “toy worlds”, which were specifically designed for them, with high contrast lighting, and with shapes that have sharp edges and bright colours.

It’s not just in robotics that progress is accelerating. Neural networks can detect patterns – something traditional computers are very bad at –, and they can do it better and of course much faster than human beings. We have been connecting them to large databases, which they can mine in order to find correlations that were previously unseen and invisible to us. Feed them information on our buying habits, and they can predict better than a psychologist, and better than ourselves, what we are likely to buy next, either as a group or even as individuals. This is true for any other patterns of human behaviour. Neural networks can be connected to CCTV cameras, and now easily outperform human beings in facial recognition, not only in speed but also in accuracy, being able to spot an individual in a crowd of thousands in a split second. Before neural networks, it was possible to track someone 24/7, if you really wanted to, but it took large resources and a massive team of human specialists. Soon the technology will be in place to make it possible to track everyone, all of the time.

Developments in AI are starting to have a powerful disruptive effect in how we live and will live in the short to medium term future. There is good and bad, here. On the one hand we will see vastly increased economic efficiencies leading to overall greater wealth, in what will likely be a new industrial revolution. AI will take on jobs that are dirty, dangerous, or dull, and thus save lives and prevent human suffering. Some jobs will inevitably disappear, but history has shown us so far that new technologies have always end up creating more jobs overall than those they eliminate.

On the other hand, wars may become politically more attractive than they are now, as it will be easier to justify the loss of autonomous drones over the loss of soldiers’ lives, and we’re of course already seeing this with current remote-controlled drones. This can lead to more frequent and wider spread military action around the world by any countries with the resources to deploy these machines. As AI systems replace not just physical but increasingly also cognitively complex human labour, the rate of job loss is likely to be faster, for a while, than the rate at which the job market can adjust by creating new and different occupations. In addition, increasingly powerful data mining capabilities will start to have a very worrying impact on privacy.

The deeper concerns

Privacy issues get plenty of coverage these days, but the rise of autonomous weapons and the possible consequences of short and medium term job loss do not seem to be getting sufficient attention. Autonomous weapons are already in use by the military, even though they are not currently allowed to make kill decisions. Initially the focus was almost exclusively on the straightforward replacement of a person with a machine, while sticking to conventional designs, such as the fighter jet or tank. The UK’s Taranis or the US’s X-47B are good examples of autonomous planes. The X-47B has already shown it is fully capable of taking off and landing on an aircraft carrier autonomously and with more precision than a human pilot ever could. The game changer, however, may be the recent trend towards developing swarms of hundreds of expendable cheap drones that work together in coordinated groups, performing detailed surveillance tasks and overwhelming defense systems. Adding kill decision-making capabilities to current drones could be trivial from a technical point of view, depending on how accurate we want their decisions to be. This technology cannot continue to be unregulated, and there are good arguments for a complete ban.

The video slaughterbots, part of a campaign against lethal autonomous robots, shows how this is a pressing issue. Shot in the form of a sci-fi dystopia, it shows the very real power of these machines when governed by AI. Most importantly, although the particular drones in the movie don’t exist just yet, that’s only because no one has chosen to build them. The technology to make them autonomous is cheap and easy to obtain right now, by a rogue state, a terrorist group, or possibly a single individual. The movie shows autonomous drone swarms acquiring and destroying subjects autonomously, based on a database of political targets, without the orchestrator of the hit being traceable.

In addition to these concerns, there is also anxiety around ensuring that the benefits of these new technologies are fairly distributed. The increased wealth that can result from higher productivity will of course not benefit everyone, so there may be rising inequality, unless policymakers are prepared to take action. The issues that Karl Marx worried about (the risks arising from capitalists owning the means of production) will be amplified, as it can be argued that capitalists will now also own the mechanised labour. However, worst case scenarios will always get more attention in the press, so we need to be careful not to panic unnecessarily, and it is too early to say whether there will be a permanent rise in unemployment.

Preparing for opportunities and impacts

The advent of new technologies has been eliminating some jobs (and creating more than those that are lost) for a long time, so in a sense this is not a new phenomenon. It could be argued, however, that AI is different. It is more flexible than other technologies, and so it will hit more industries and displace jobs faster than new jobs can be created or workers can be retrained, although it is more likely that parts of each job will be done by an AI, rather than the entire job. There will be some notable exceptions, such as driving jobs, which could potentially be wiped out in the next twenty years or so.

Policymakers must start to seriously consider what solutions can be developed in order to deal with the temporary job losses that do seem inevitable. Some possible approaches that have been discussed are:

• Taxing robots for their labour (i.e. as they replace human labour) – This seems to be a popular idea in some circles, although governments have never taxed other productivity tools (whether they are hammers, laptops or more efficient marketing strategies) for the labour they displace. It seems unlikely in the competitive environment that is the global economy that any government would choose to handicap its economy like this. What is the intention, other than increasing corporate tax, and calling it something else?
• Increasing corporate tax everywhere – This is a possible solution if there is sufficient political will globally. This could only be carried out with global coordination (e.g. at WTO level), if at all possible, to avoid a loss of competitiveness if it happens at the level of a single state. If left to each government, no one has an incentive to do it. Possible but unlikely exceptions are very large economies (e.g. the US), with large enough internal markets that enable them to endure trade wars and remove themselves, to some degree, from exposure to global competition, choosing to live by different rules using a range of protectionist measures. It is clear that increasing tax revenue is a very difficult problem.
• Investing in skills – In addition to enabling greater social equality, any increased tax revenue that could be obtained could partly go towards government-sponsored retraining programmes. Up-skilling workers at a greater rate than ever before is vital to ensuring a robust and productive workforce, and helping more people to take part in the economies of the future. Policymakers will need to keep abreast of technological changes to make sure that such programmes are up-to-date and can genuinely prepare workers.

While we must be careful not to overstate the pace of development in AI – let’s not worry about a robot takeover just yet - there is certainly a potential for these AI-powered machines to radically transform the way we live. There are both great opportunities to live better, and worst case scenarios to prepare for that are far too real not to take seriously.

Professor Vasco Castela teaches ethics and philosophy of AI at the British Columbia Institute of Technology. He holds a Master’s in Philosophy of Cognitive Science (Sussex) and a PhD in Ethics (Manchester).

Many of the technological advancements discussed by Professor Castela have the potential to bring both benefits and negative consequences. For example, AI systems coupled with CCTV technology could positively contribute to society in a number of ways, but may also have severe implications for individual privacy.

The ability to intelligently monitor huge volumes of real-time video data would assist law enforcement to respond faster. A system able to analyse vast numbers of video feeds and alert officials to developing situations would be very valuable in a crisis. AI surveillance would also make it easier to gather evidence used to prosecute criminals (or for any other purpose). Intelligent systems that have the ability to analyse and understand the contents of a video (beyond the obvious metadata, such as timestamps) would make it easier to search for footage featuring a certain person, vehicle, act, location or other characteristic.

However, the potential impact on privacy is clear. Although CCTV today is widespread only a small fraction of footage can be actively monitored in real-time. Were AI to be used this fraction could increase significantly, and it may be possible to more easily track an individual using both real-time and historical footage. Policymakers are increasingly grappling with regulations that define how technology companies can store and use data on user activity online, where it can be easily tracked. If AI and CCTV combined enable the ability to track a person's activity in the real world, it is important that the use of this data is also regulated effectively.

Fairer working arrangements for a digital world

By Professor Tony Dundon

The explosion in new forms of business model enabled by innovative technologies and digital platforms raise questions about the ‘future of work and employment’. These concerns centre not just on how to make activities more efficient, profitable or productive, but also relate to public policy debates about whether we are witnessing the end of conventional employment patterns for large sections of society and, if so, what alternative arrangement may be available.

While debates and policy preferences differ widely (from those worried about the consequences of widespread work displacement to those who welcome labour being substituted by automation) it is widely recognised that technology in its digitalised form is driving change. This section questions the extent of digital platform modes of employment and discusses how genuinely ‘new’ such arrangements are.

The longevity of work design systems

The key principles for the structure of work underpinning digital platforms are not necessarily that new. Many of the organisational features about digitalisation and work actually mirror some of the structural systems of early capitalism. Key features include:

• Work tasks that are performed on-demand, similar to zero hours contracts. Workers often spend time ‘hanging around’ or ‘searching for gigs,’ and this is time spent without remuneration making ‘platform work’ precarious, insecure and unpredictable.
• Pay is often casual, based on a piece-rate or payment-by-results system.
• Workers typically have to supply their own equipment (examples are a bike for Deliveroo workers, a car for Uber drivers and a computer for performing online tasks for many worker segments).
• Intermediaries are often used for connecting workers with end-users. In previous periods such intermediaries would have been known as gangmasters, local market recruiters etc.

These are issues of power, control and coordination which mean there are risks. Such working arrangements can make for more unstable, unpredictable employment regimes that have wider societal spin-offs affecting welfare, security and insecurity, social issues and economic wellbeing of regions, areas and occupations.

Is technological change inevitable?

When viewing the pace of its progress and development it can seem as if technology - and its digital platform additions – is somehow neutral, or that change is inevitably technologically determined. It is not.

The type of technologies used for different tasks, and their subsequent effects on work and employment regimes, are shaped by the decisions of stakeholders who typically occupy positions of power and authority, and who control resource allocations. When employers use cheap labour instead of more expensive - and potentially more productive - technology it is often because they are choosing lower labour costs over higher infrastructural investments. When it comes to digital work platforms and new technologies (whether considering Uber, robotics or automation in manufacturing), the primary motives are increasing value, efficiency and profit. Rarely is the social good or societal gain part of the strategic rationale.

Significantly, under digital platform work the intermediary relationships are governed by technology; namely the app or computer interface that supports the new business model. Often instead of empowering the worker such intermediary platforms result in an uncertain contractual status as to who is a worker, employee, freelance agent or those genuinely self-employed. The net effect is the dilution of legal rights and exclusion of protections for large and growing sections of the labour force.

This also results in a change to psychological contracts and social exchange relationships; people now interact and communicate via technology and no longer in person. One consequence can be a lack of opportunities for social development and for employee voices to be heard.

Technology erodes standard employment arrangements

Digitalisation now engenders ever more ‘bogus self-employed’ and ‘sham contracts’ (in the US for example the proportion of workers with alternative work arrangements has been estimated to have grown by 66% from 2005-2015). In doing so it is encouraging employers to dodge long-established statutory rights that a society values as good for its citizens, and this is having an effect at a global level.

Contemporary issues of precarious recruitment practices resonate more with Victorian or Dickensian era insecurities resulting from temporary and peripheral labour on short term hire contracts. Today, instead of workers standing at the dock gates waiting for the foreman to select a few workers for the day, employees now log on to a digital app in the hope of a new gig or digital task.

In such arrangements risk and cost are no longer the responsibilities of the employer or of large corporations but are instead placed on the individual worker. In addition, a key difference between the working arrangements of today compared to those of previous eras is that payment is regulated by the intermediary, and in some cases workers are not paid until the final end-user is satisfied or even after the intermediary has been paid.

Technology may uplift some skills, but it also contributes to employment precariousness, with a culture of long and unsocial hours, where labour can be considered a cheaper alternative to technological investments by firms.

Policy issues and options

In order to develop effective and fairer working arrangements, policy positions that meet the challenges of new business model may need to be considered; for example:

• Debates are required on how to regulate new work arrangements that appear to outstrip the traditional legal employment contract which prevail in most advanced countries. While companies will continue to innovate and disrupt traditional business models and working arrangements, strong foundational regulatory approaches that can adapt to the effects of modern technology can help employers develop sustainable fairer employment options.
• In the UK the Taylor Review appears to not go far enough in its recommendations for changing the nature of working practices. The Review recommends renaming platform workers and those in other non-traditional arrangements as ‘dependent contractors,’ but is unclear on how this will help them to avoid precarious, bogus self-employment situations. As the Review states, more work is needed to clarify the definition between dependent contractors and those who are genuinely self-employed.
• Better regulation and employment citizenship standards are currently on the agenda for transnational policy bodies such as the International Labour Organization (ILO), the Organisation for Economic Co-operation and Development (OECD) and the European Union (EU). This focus is welcome and will hopefully contribute to developing fairer, more relevant working structures in the future. Policymakers and other stakeholders dealing with these issues are encouraged to incorporate and engage with the transnational consultations underway and seek to develop aligned policies in this international context, particularly as the UK prepares for an uncertain Brexit landscape and new global trading challenges in the years ahead.
• Businesses should consider promoting sustainable employment and emphasising human rights in working arrangements in order to develop social as well as commercial value from robotics and digitalisation. This is particularly true for multinational corporations operating across supply chain networks in less developed parts of the world.

In addition to ensuring investments in AI are sensible and profitable, many contend that they should also be fairly distributed.

Dr Barbara Ribeiro of the Manchester Institute of Innovation Research explains that because investment into AI will come at a cost for tax-payers, policymakers should try to ensure that the benefits are fairly distributed.

The essay below details approaches that policymakers and industry stakeholders can use to make this a reality.

Ensuring social justice in AI development

By Dr Barbara Ribeiro

AI is inherently biased and can be discriminatory, both intentionally and unintentionally, meaning more should be done to ensure its democratic and socially responsible development.

AI technologies rely on big data and the use of algorithms, which influence decision-making in public life and on matters such as social welfare, public safety and urban planning. In these ‘data-driven’ decision-making processes some social groups may be excluded, either because they lack access to devices necessary to participate or because the selected datasets do not consider the needs, preferences and interests of marginalised and disadvantaged people.

Recent concerns around data discrimination in financial services and employment opportunities for example have motivated the US government agency the Federal Trade Commission (FTC) to develop a number of best practice recommendations for firms. In the UK, regulations for data analytics will need to accommodate the emerging risks associated with potential discrimination as AI technologies are increasingly embedded in society (and as new potentially discriminatory practices emerge).

With support from the government, an independent body could bring together a diverse range of voices from citizens (not necessarily their representatives!) and aim to ensure that the knowledge, values and concerns of civil society reach policymakers. A ‘stewardship body’ for data governance has recently been debated by the British Academy and the Royal Society, and recommendations have been put forward that such a body prioritises transparency, accountability, and open and inclusive approaches.

To help ensure more representative public policies, Dr Ribeiro also offers recommendations for policymakers in three areas:

How organisations can develop more representative AI public policies.

How local governments can ensure AI development incorporates greater social justice.

How to carry out and implement such processes.