new ways of achieving a mission

Like many other technologies, AI has the potential to be harnessed by charities and civil society organisations to help them further their mission. The number of examples of this in practice so far remains relatively small, but those that are out there give a sense of the future potential.

Making existing information more accessible

Perhaps the most prominent example of AI in a charity context so far is the use of chatbots. These are AI-powered, text-based conversational interfaces; which most of us will by now be familiar with through their widespread use in commercial customer service settings. These are being harnessed by a number of charities to provide services of various kinds.

For example, Arthritis Research UK has partnered with IBM Watson to develop a bespoke ‘virtual personal assistant’ that can provide information and advice to people living with arthritis. Other charities are tapping into more “off-the-shelf” versions of the technology by piggy-backing on existing platforms: WaterAid, for instance, has developed a chatbot aimed at awareness-raising and engagement that can be accessed through Facebook.

There is a potential long-term financial incentive here, in that charities might be able to reduce costs significantly if they are able to provide services via chatbots 24 hours a day, 365 days a year, rather than having to employ large numbers of staff. However, as with any examples of automation, this may prove controversial and raises questions about the responsibilities of organisations towards their human employees in the future. Cost-cutting is not the only reason to consider utilising chatbots though - there are also potential advantages for service users and supporters. One is that advice can be tailored based on the user’s stated needs. In addition, the time it takes to find relevant information significantly reduced. For advice on things like acute mental health issues, where there is a premium on people being able to access the help and advice they need at crisis moments (which might well be in the middle of the night), this could offer a real advantage over human operator-delivered advice services.

One step beyond site-specific text-based chatbots there are conversational assistants. Many of us will now be aware of these more generalised voice-operated interfaces as a result of their presence on our mobile phones or, increasingly, in our homes in the form of devices like Amazon’s Alexa or Google’s Home. One immediate possibility this offers is to make the internet (and, as a result, other technologies) much more accessible to certain groups. This is most obvious in the case of those with visual impairments. However, it could also benefit those who have no particular physical disability but are simply less comfortable with technology; because conversational interfaces mirror more closely the ways in which people are used to getting information offline.
 

Could we see an end to language barriers?

Another area in which some charities have been able to harness the existing application of AI is language translation. Many of us will have used Google Translate or another text-based translation tool at one point or another. Despite the obvious power of these tools, they have until now been seen as somewhat inaccurate and not a valid replacement for human translation services. But that may be about to change: we are starting to see new AI-based translation tools that not only match human performance, they outstrip it. It is only a matter of time until AI translation is the standard.

The crucial thing is that this does not only apply to written translation: the development of powerful speech recognition and natural language-processing algorithms has now made it possible to achieve highly accurate voice translation in real time, and this is a product that is already available commercially. This could be of huge benefit for charities that deliver services, in terms of ensuring they are as accessible as possible.

For example, The Children’s Society reported in 2017 that it was using Microsoft’s Translator app to facilitate some of its interactions with refugee and migrant young people. In addition to reducing the cost and administrative burden of having to procure professional translators, charity workers who used the app also reported that in some cases it had made the interaction easier, as the young people were more comfortable talking about sensitive and difficult issues without an unfamiliar third party in the room.

To see where things are likely to go next, it is worth noting that in late 2017 Google announced the launch of its new Pixel Buds: in-ear headphones that are able to use AI to live-translate 40 languages directly into the ear of the user.  As devices like these become more widely available, and the technology becomes embedded in other products and services, it may well be that we will see an effective end to language barriers in the foreseeable future.

An introduction to AI Terminology

   

Algorithm

A set of step-by-step instructions for performing given tasks. These can be very simple (e.g. “If it is Tuesday, put the bin out”), but they can also be incredibly complex (e.g. Google’s PageRank algorithm).

Artificial Neural Networks (ANNs)

ANNs are learning models based on the ways in which networks of neurons function in the brains of animals. Neural Networks have long been one of the major fields of study within Machine Learning.

Deep Neural Networks (DNNs)

DNNs are ANNs with one or more hidden layers between the input and output layer. This enables the representation of more sophisticated relationships between data, which has made it possible to use unstructured data in a way that was previously not possible. DNNs are behind many of the high-profile successes of AI in recent years, such as improvements in image recognition or game-playing.

Human Computer Interaction

Human–Computer Interaction (commonly referred to as HCI) researches the design and use of computer technology, focused on the interfaces between people (users) and computers.

Machine Learning (ML)

A branch of AI that allows computer systems to learn directly from examples, data and experience. The crucial difference between this and traditional approaches to AI is that you do not need to know all the “rules” up front in order to create effective algorithms. You can simply set goals and allow the algorithms to learn for themselves through trial and error and self modification.

Natural Language Generation

NLG technologies transform data into written or spoken language. For example this is used by services like Apple’s Siri or as seen in films like Space Odyssey. This is the technology most people think of first when they imagine interacting with an AI platform.  NLG can also be behind marketing content such as push notifications, emails, and some short-form articles that readers may not be aware are artificially generated.

Reinforcement Learning

A form of unsupervised learning in which the system is given a metric of success (a ‘reward signal’) against which to measure efforts to perform a set task. The algorithm learns and adapts over time to maximise its own reward signal. This is the method often used to teach ML systems how to play games (e.g. arcade games, where the reward signal is simply the points scoring system of the game itself).

Supervised Learning

A form of ML in which those programming the system already ‘know the answers’ (i.e. the correct pairs of inputs and outputs), so the data can be labelled. Hence there is direct human involvement in “teaching” the algorithm.

Unsupervised Learning

UL is a form of ML in which systems are trained to find patterns in sets of data where we don’t necessarily have a set of “right answers”, either because we cannot determine it or because there is no “right answer” to the task in hand. Often the outputs of these systems involve ‘clustering’ data together based on perceived patterns.

Data Analysis & New information

Other potential charitable or social good applications of AI focus less on improving accessibility and experience for service users, and more on harnessing the power of machine learning algorithms to analyse data at unprecedented scale and speed in order to develop new insights.

Medical Research

Google DeepMind, for instance, has a partnership with Moorfields Eye Hospital which seeks to apply the company’s deep learning algorithms to the large numbers of Optical Coherence Tomography (OCT) scans taken each year in order to develop more effective diagnostic tools. Researchers have also highlighted the further potential there is for applying machine learning to large sets of medical images as a means of getting earlier and more accurate diagnoses in future.

A key part of the strength of ML systems in this context is that they do not have their rules fixed from the outset and are able to adapt and ‘learn’. This means that they have been able, in some cases, to identify imaging features that are predictive and can help to form a diagnosis; yet have never been picked up on by human medical experts. As an early sign of future developments in this area, Cancer Research UK has issued a challenge prize for projects that can extend beyond image analysis and apply AI to a wider range of potential predictive data to improve cancer diagnoses.

Another area of medical research in which AI is being applied and charities are playing a role is around drug discovery. Traditionally, finding new drug treatments is enormously expensive and time-consuming, as it involves a large element of predictive guesswork, which then has to be tested rigorously through experimental trials. The potential benefit of using AI is that it enables accurate modelling of things like protein folding, which plays a huge part in the efficacy of many drugs and which was previously very difficult to predict. This means that it is possible to estimate the likely effectiveness of a proposed drug much more accurately, and thus reduce the time and cost of focusing in on the most promising candidates. The charity LifeArc has partnered with the University of Cambridge on a new project along these lines, which seeks to identify and validate new drug targets in immuno-oncology and respiratory diseases using machine learning.

Environment and conservation

An intriguing suggestion for how AI could play a much larger role in conservation in the future is that of “algorithmic wilderness”. This is the idea that entire ecosystems and habitats could be managed by algorithms with the ability to regulate conditions and deploy autonomous drones for maintenance, seed-spreading etc., so that they could be maintained as pristine, effectively ‘wilderness’ areas ─ free from the need for human intervention.

This reality is still some way off, and some might think that it represents a depressing, technocratic vision of a future in which we have simply given up on trying to address the damaging effect we as humans are having on our planet. However, whatever the ethical debates about whether this particular approach is desirable; it does highlight some of the future potential for applying AI to complex systems of all kinds.

Data for scientific breakthroughs

A final application of ML that could benefit charities right now is its use to make scientific research more efficient. For instance, in 2015, the Allen Institute for Artificial Intelligence (founded by Microsoft co-founder Paul Allen), launched Semantic Scholar, a ‘smart’ search engine for scientific journal articles that uses machine learning, image recognition and natural language processing  to enable scientists to identify to most relevant papers and the links between them. Likewise, in 2017, Mark Zuckerberg and Priscilla Chan’s philanthropic vehicle, the Chan-Zuckerberg Initiative (CZI), purchased a commercial startup called Meta, which has developed an AI system that does a similar job of helping scientists navigate, read and prioritize the millions of academic papers in existence.
elderly lady holding cup case study image 580 260 unsplash claudia van zyl

ML tackles Parkinson's disease

  • Parkinson’s UK

Parkinson’s UK has partnered with Benevolent AI to explore the application of Machine Learning data on miniscule body movements captured by smart phones, which could lead to the development of new early warning indicators.

Read about intelligent research for Parkinson's
herd of elephants africa 580 260

The Air Shepherd fights poaching in Africa

  • The Lindbergh Foundation

Air Shepherd is an unmanned aerial drone that patrols conservation areas and record footage. The Foundation has worked with Neurala to apply deep learning algorithms, aiming to teach drones how to recognise poachers.

Visit Air Shepherd

Challenges for AI for Good

We are only beginning to explore the opportunities for charities to harness AI to deliver their missions more effectively. However, there are going to be major challenges when it comes to realising this potential. For instance:

Skills

It is unlikely that many (if any) charities will have skills in-house when it comes to developing AI systems. A recent report on charity digital skills in the UK found that 73 per cent of organisations reported having low to very low skills in AI. This isn’t at all surprising: widespread use of the technology is still at a relatively early stage, so we probably should not expect charities to have put resources into employing people with expertise in machine learning or similar at this point.

Furthermore, reports show that the huge interest in AI currently is outstripping the supply of people with the required skills; which means that a talent war is being waged by many of the big companies looking to use AI to get a competitive advantage. In this climate, the chances of charities being able to compete successfully to employ AI specialists - even if they wanted to - seem very small.

Risk appetite

New technologies (often quite rightly) are seen as high-risk Given that most charities and civil society organisations operate with very limited resources, it is going to be extremely difficult for them to justify making speculative gambles of the kind required to be genuine early adopters of a technology like AI. The challenge, then, is how to mitigate or reduce this risk.

One possible solution lies in developing stronger partnerships between civil society and the tech sector, so that the responsibility for shouldering innovation risk can be distributed more appropriately. There is a lot more work to be done by governments, technologies companies and civil society organisations to make this happen in practice.

Leadership

Leadership ─ both at a senior management and a trustee level ─ is absolutely vital when it comes to CSOs engaging with new technology. This is not necessarily about leaders having specific technology skills, but rather about them understanding key trends and drivers and articulating a clear vision for how their organisation can adapt to the new challenges and opportunities they bring. In the UK Charity Digital Skills 2018 report, 77 per cent of respondents said that they would like to see their leadership team develop “a clear understanding of what digital could achieve”, while 63 per cent said they would like to see them develop “understanding of trends and how they affect your charity”.

There may be cultural issues (at least in the short term) as those in leadership roles are likely to be older and thus less conversant with technology that many “digitally native” younger professionals. This is particularly true of trustees: research published by the Charity Commission for England & Wales in 2017 found that the average age of trustees was 55-64 (and this increased to 65-74 in the smallest charities, i.e. those with annual turnover of £10K or less).

Leadership ─ both at a senior management and a trustee level ─ is absolutely vital when it comes to CSOs engaging with new technology.

A cause for concern

A survey conducted as part of that research also showed that only 21 per cent of respondents felt they currently had sufficient representation of digital skills on their board. The report concluded that “whilst it is clear that trustees have begun to embrace the onset of the digital revolution, it is equally clear from these results that they require further training and development support if they are to be fully integrated to discussion at board level.” We should also bear in mind that this primarily refers to digital skills in the sense of using social media and the internet, and doesn’t really touch on the level of skills and understanding when it comes to emerging technologies like AI. Given the degree to which AI could affect charities and the communities they serve in coming years, this may be cause for concern.

In addition to addressing immediate leadership challenges when it comes to technology in the charity sector, it is also important to look to the future. That means supporting wider awareness and education so that the potential leaders of tomorrow, for whom it may come more naturally to engage with technology issues, have the skills and understanding they need to make the sector as a whole more technologically adept.

Investment

Another challenge that most CSOs will face is finding the resources to invest in new technology. Even if an organisation’s leadership is on board and they are comfortable with the risks, they may simply not have the money required to do anything about it. There is a role here for grantmakers such as trusts and foundations to provide financial support to organisations that can provide a compelling case for investing in experimenting with new technology.

Companies could also play a key role: many organisations across a wide range of industry sectors are experimenting with technologies such as AI, and there is a case for saying that they could extend this into some of their CSR or corporate philanthropy activity by helping charities to develop applications of the same technology to social and environmental use cases. Not only would this potentially solve the investment challenge for charities, it could also allow them to tap into the additional value of the corporate’s expertise by getting pro bono technical help where possible. Furthermore, there is probably a reasonable element of enlightened self-interest on the part of companies in taking such an approach: a technology such as AI is still so nascent that developing any meaningful use cases is likely to bring insight that will be of use when it comes to commercial innovation.

Data

For machine learning of any kind to work you need large amounts of clean, usable data on which to train the algorithms. This is likely to present a major barrier to charities in terms of harnessing AI, as the sector faces huge challenges in terms of both quantity and quality when it comes to data.

In terms of using ML to determine patterns in data that can be used to design new interventions, we have seen that there are some fascinating examples of this happening already. Many of these are focussed on healthcare so far, which may be a reflection of the fact that in this area it is possible to use an existing body of medical records. However, this also raise difficult questions about data ownership and usage, as Google DeepMind and the Royal Free Hospital found when they were censured in 2017 for breaching data privacy laws by sharing medical information inappropriately.

More recently, the huge furore over the illegal sharing of millions of Facebook users’ data with Cambridge Analytica, and the role that may have played in the 2016 US Presidential election, has brought the issue of data misuse to the heart of mainstream debate. The ongoing tension between people’s desire to get the benefits of technologies like AI and their growing unease at the way their personal information is used is likely to present challenges for charities wishing to apply ML in other areas in the future.

As well as data on the “demand” side of the charity equation (i.e. data that enables us to understand needs better), there is also a question about the “supply” side (i.e. data that enables us to assess which organisations and interventions are most effective at addressing given needs). We will look at this in more detail in section 3 in the context of “philgorithms”, but it is worth noting here that we are nowhere near having enough data on social impact at this point. Although there are initiatives seeking to drive more consistency in impact measurement, there is still no real consensus about the best way of doing this, and it is far from universally agreed that measuring impact is even a good idea. It may be that the potential benefits of harnessing AI offer an additional incentive to drive the impact measurement agenda forward, but there is a long way to go in any case.
     

< Introduction: Why is AI such a big deal right now?

Part 3: Impact on organisations and funding >

Charities Aid Foundation © | Registered Charity Number 268369
25 Kings Hill Avenue, Kings Hill, West Malling, Kent ME19 4TA
10 St. Bride Street, London EC4A 4AD
Telephone: 03000 123 000