Machines For Good: The Human Face Of Artificial Intelligence
There has been a notable debate about the potential for artificial intelligence (AI) to have a detrimental impact on society, from being a threat to our data privacy to destroying employment. Whether these concerns prove legitimate or not, AI and more specifically machine learning can be a force for good for the social sector.
Understanding AI
Before focusing on how AI and machine learning will impact the social sector, it is useful to understand what exactly is meant by AI and machine learning. True Artificial Intelligence, for example computer responses being indistinguishable from human ones, is not quite with us yet for most complex scenarios. The term AI is however often used to signify the use of machine learning. In its most basic form machine learning builds on a computer’s ability to recognise complex patterns. When faced with a massive amount of data and a wide variety of variables, computers can help us spot useful patterns.
Consider the simple scatter graph, it compares two different variables for as many points as we can reasonably fit on a page. You may be able to picture scaling this up to three dimensions (Figure below) but how would you compare 4, 5 or even more variables? Computers are not constrained by our three-dimensional lives, neither do they become exhausted, forgetful or just plain bored when faced with millions of data points. This enables them to find patterns that we cannot see and make predictions.
As you may have experienced or read, machine learning techniques are now being extensively deployed in the corporate environment. Machine learning takes over when the excessive quantity and/or complexity of variables means that traditional, simple rules-based systems are unable to cope and start to break down.
Machine learning techniques are being used by businesses in five key ways:
Applied to the social sector, these technologies can help improve performance and drive meaningful change in the world. A great deal of time in the social sector is spent reacting to problems. The predictive analytics generated by machine learning algorithms could identify early warning triggers allowing a move to more impactful, cost effective, preventative models.
The potential of machine learning is deeply related to the potential of data because data is the material that computers are learning from. The good news is that data is becoming increasingly abundant in the social sector as more people, organisations and devices become digitally connected, from the growing number of ‘smart homes’ to the digitalisation of national health records. We will explore in a subsequent article how to get the most out of your data.
We have identified three distinct but related areas where machine learning can meaningfully support public, private and voluntary organisations involved in the social sector across education, employment, care, health, justice and housing.
Improving social impact by identifying what works (and doesn’t work) and delivering relevant preventative and support services
Improving operational efficiency and user experience by automating processes and offering a simpler, more intuitive access to information and services
Increasing income generation by better understanding client and supporter behaviours and propensities
Improving social impact
The social sector can use technology to find and connect with more people who need support, understand communities and issues on a deeper level, predict outcomes to inform preventative and support activities, monitor risks, and even measure the outcomes and long-term impact delivered.
This would help organisations alleviate the considerable administrative burden on their staff, free up time for more critical tasks, improve decision-making, and deliver better, faster services.
Identifying people in need of support
Most organisations involved in the social sector will develop an explicit or implicit view of how their action is having an impact on the specific issue that they are trying to address. A logic model or theory of change are often used to articulate how the organisation is working towards achieving specific outcomes and long-term impact. The causal links between activities, outputs, outcomes and intended impact are often derived from the experience of delivery teams. Machine learning techniques can provide teams with a deeper level of evidence so that they can better differentiate which activities have the greatest impact and redesign their services accordingly. In addition, through increased data sharing between agencies and improvement in data quality, teams will be able to harness larger amounts of data to develop meaningful service and sector insights. They will be able to recognise patterns within a community or particular cause, predict future outcomes and initiate preventative actions.
For example, educational institutions can recognise patterns within a student’s journey, so that teachers and advisers can proactively reach out to students who exhibit risks of failing or dropping out before it actually happens. In a different sector, an NGO may use mobile phone data to identify the specific movements and number of refugees coming into different countries to predict where their settlement will be and pre-emptively send the appropriate level of aid and supplies to the right location.
Case Study – Children at risk for neglect and abuse
London councils are using data analytics to predict which children are at risk of neglect and abuse, allowing them to act before crisis occurs and prevent more costly, extensive interventions. Using data from multiple agencies, they were able to identify children not previously known to the local authority and to help social workers to intervene early.
Benefits of this pilot included:
Increased identification of Troubled Families (TF). One Local Authority has already identified almost 400 additional families to receive support through their TF programme.
Savings of circa £122k from increased efficiency in TF teams.
Improved access to multi-agency data, leading to increased efficiency in safeguarding teams, equating to circa £148K of savings.
Total cost avoidance estimated over £700K
Other examples include:
A suicide prevention charity uses voice and key phrase recognition to identify and prioritise support to people with a higher risk of committing suicide.
A carers organisation enables patients to remain at home instead of at hospital while ensuring their safety through a combination of Internet of Things (IoT) devices and carer support workers ready to act.
New York City has reduced the number of fires in the city by identifying buildings most at risk and making targeted improvements
Machine learning can also help to identify abuse of social systems leveraging the techniques developed to identify banking fraud. In the social sector these same techniques are starting to be applied to identify scams and frauds to make sure that services are reaching those truly in need. Recent estimates show that fraud represents a national cost of £2.1 billion to councils every year, including £133 million in council tax discount fraud. It also offers opportunities to get ahead of any potentially scandalous misconduct issues.
Personalising services
By helping to understand individual needs, machine learning solutions enable a more personalised service and better matching between stakeholders. It also helps organisations to better tailor services to their clients’ needs and expected outcomes. This is of course critical to support the successful development of the personalisation agenda which started over 10 years ago in the UK. Personalisation means starting with the person as an individual with their own set of strengths, preferences and aspirations. It means putting them at the centre of the process of identifying their needs and making choices about how and when they are supported to live their lives. Machine learning techniques can help individuals to better assess the likely consequences of their choices and therefore make more informed decisions. In the education sector, pupils can be informed of the best pathway for the career that they are contemplating.
An interesting application is in the adoption matching process between children and adopting families. With 8% of adoptions at ‘breaking point’ and a further 21% facing major difficulties, understanding what makes a good match could save already overstretched resources and prevent the trauma of a child being rehomed multiple times. In addition, pre and post adoption support services can be better tailored to the individual needs of the children and their adopting families.
Improving operational efficiency and user experience
AI and machine learning should prove a very effective tool in improving the efficiency of services and the user experience of clients and beneficiaries.
One core area is process automation. AI is likely to have wide application in the processing of applications and submissions, including for tax, benefits, visas, passports, and other Government licences. Many of these applications from citizens can be processed more quickly, and abuse of the system spotted more accurately, using AI. Machine learning algorithms could be used to triage the risk associated with different applications. The very low risk applications can be processed automatically, leaving caseworkers to focus on the more difficult applications and on applicants who need more assistance. The time it takes to process applications fall, the associated costs would also fall, and the system would become more secure.
Within charities, front line workers are often spending a significant portion of their time performing analysis and administrative tasks. This can be as high as 70% in social care. Automated chatbots and AI can help them to reduce that burden and free up their time so they can focus on interacting with the service users. These strategies can also be targeted at operational issues within the social sector such as recruiting the right people for roles, optimising the allocation of support workers or identifying the best operational locations.
Additionally, many organisations are starting to use AI solutions to improve user experience by reducing waiting times, engaging with clients across digital and social channels, and using chatbots to offer 24/7 support (something they could not afford to do if they were to use a traditional call centre or face-to-face model).
Other applications include:
Faster and cost-effective services using automated decision processing for applications, referrals, etc.
Automated performance and impact reporting
Resource allocation / demand matching
Asset maintenance optimisation
Language translation
Increasing income generation by better understanding supporters’ behaviours
Whether income comes from individual donations, grant or funding applications or fundraising from campaigns, understanding any causality between the methodology, timing and phrases used and the impact on income is essential to increasing future funding. Organisations that can master lifetime value modelling and supporter segmentation through machine learning solutions will be a step ahead of the others. This is critical in a post-GDPR world where the acquisition of new supporters and donors is likely to be more expensive and time-consuming.
Similarly, predictive algorithms can identify when contracts or donations are at risk of cancellation or non-renewal, providing opportunities for early intervention. Machine learning algorithms can also be used to identify the types of intervention which are most likely to work by training it with previous history and keeping it up to date with changes in policy, decision makers and strategy. This can lead to a greater retention of donors or contracts, reducing the need for costly investigations of new revenue streams.
Current applications include:
Identification and segmentation of customers/supporters
Supporter engagement strategy
Lifetime value models for donors
Pricing optimisation for goods and services
In summary
An independent report commissioned by the government estimates that AI could add an additional £630 billion to the UK economy by 2035 . There is a real opportunity for AI and machine learning techniques to drive sustainable change in the social sector and organisations must now invest to create their own AI strategy. Doing so requires informed and structured imaginative thinking to identify how machine learning can be used to meaningfully support public, private and voluntary organisations in the long run.
In our next article we will explain how to diagnose your organisation and eco-system to identify and prioritise machine learning opportunities in a structured and efficient manner.