Updated: Nov 20, 2020
Many suppliers describe AI and Machine Learning as digital gold. Sprinkle a little onto your data and watch it transform your business in ways you could never imagine. Despite the promises, many organisations derive no meaningful benefits from these new and emerging technologies.
There are many reasons for this; some are unique to a particular business or industry sector. Often, the lack of sufficient and appropriate data will be a significant constraint. For example, in education, we need to consider our students' wellbeing and their academic progress. We must understand the balance between accuracy and fairness for decision-making algorithms. We should view AI as a complementary addition to 'more traditional' teaching methods - rather than a replacement.
The individual elements combine to form a data strategy. A plan that describes the people, actions and assets to achieve a future, data related, goal. The outcomes enable you to make better decisions faster by leveraging the data at your disposal.
The challenges of a pandemic are speeding up automation and remote learning. Making mistakes is inevitable, organisations should see them as an opportunity and a way to learn - not a reason to assign blame. We are in uncharted waters, and as Albert Einstein once said:
"Anyone who has never made a mistake has never tried anything new."
Successful AI adoption requires creative thinking, strong collaboration, and complete transparency. An ethical framework promotes trust and generates confidence among both the providers and the consumers of AI services.
Please click the link below if you prefer to read this article as a pdf.
Preparing for the future
In an uncertain world, agility is essential. The ability to think creatively, try something and adapt quickly has never been more critical. IBM suggests that 120 million people will need retraining over the next three years because of technological evolution. Academic establishments, training companies, online learning and employee mentoring will all play a part.
Established forms of education may change to accommodate a more dynamic world. Many people take the typical path from secondary school to college to university—a three or four-year degree course and possibly a post-graduate diploma afterwards.
As progress accelerates, agility becomes increasingly essential. There is no longer the luxury of creating static, multi-year curricula. As knowledge expands, we must specialise more - consuming information continually with tailored subjects and shorter timescales. Emphasis is shifting from knowledge accumulation to knowledge flows. Degrees that take multiple years to complete are not always appropriate, and continuous learning is becoming a preference for many people.
A complete digital transformation is unnecessary; smaller, incremental improvements add value over time. Realistic outcomes and quantifiable progress help us overcome the frustrations that can occur when progress is perhaps slower than we might expect. Arthur Ashe, a former Wimbledon and U.S. Open tennis champion, once said,
"Start where you are. Use what you have. Do what you can".
Advice as relevant in the world of technology as it is on a tennis court. An iterative and incremental approach delivers continual and sustainable results. Feedback from students and staff is your most valuable asset - early and frequent updates promote the ability to do this.
Vocational versus academic expertise
Many people worry that AI leads to mass unemployment. Many receive little comfort by the promise of retraining and switching to a different career. Not everyone has the ability or inclination to start over as a pure mathematician or an AI engineer.
There are many opportunities to consider. As AI services become more democratised, organisations can focus on integrating them into their working practices. If we do not incorporate the insights that AI offers with our business processes and decision-making workflows, then nobody benefits.
Teaching vocational skills is an increasing responsibility for employers and learning institutions. As technology continues to play a role in everything we do, the goals of vocational education are to:
Introduce people to AI and related technologies with the ability to observe and improve the way they integrate with existing business processes.
Create artisans and technicians who are enterprising, innovative and self-reliant.
Teach people to apply technical knowledge to address environmental problems and challenges.
These vocational roles exist alongside technical engineers and software developers. Strong collaboration between them forms valuable feedback loops where AI services and business operations continually benefit from each other.
The whole is greater than the sum of the parts
While there are many elements to higher education, we focus on three broad areas here. AI can help to address challenges in each of these areas.
Teaching staff have many responsibilities beyond educating and supporting students. Everyday tasks include creating teaching schedules, assessing student performance, and designing subject curricula. Much of this administrative work is repetitive, time-consuming, and an essential part of any educational institution.
If we can automate some of these activities, educators can focus on what they well - collaborating with their students. In this way, everyone benefits from the combination of people and technology.
Research by the Organisation for Economic Co-operation and Development (OECD) shows there is a strong correlation between teacher-student relationships and academic achievement. Teachers that show qualities such as empathy, warmth and encouragement have a direct link to improved student achievement. Spending more time together is the key to building effective relationships.
A report by Public Health England describes how the health and wellbeing of students contribute to their ability to achieve their full academic potential. Further, the universities UK website urges universities to make mental health a strategic priority.
Many factors affect a student's wellbeing. Problems can go unnoticed by everyone except those who are suffering. When issues do become visible, it is often by proxies. Someone's grades not being as good as you expect or their attention and collaboration seeming lower than it was. AI can help to identify indicators that predict and detect potential problems. Is a student's course load is too high? Are there personal problems that someone is struggling to overcome? Are there financial commitments that people are struggling to meet?
If we can surface these problems earlier, then addressing them becomes a little easier. When people do not feel well, their grades suffer. Lower grades can create anxiety – there are storing ties between wellbeing and academic performance.
Another area where AI can play a significant role is with personalised learning. Everyone has unique preferences and learning styles. Understanding and optimising at scale is difficult without AI. Machine learning algorithms can help to plan a learner's journey with targeted recommendations, focusing on specific content and setting goals that align with individual needs.
Systems can show where a learner may skip specific modules, allowing them to focus more on areas of increased benefit. There will be subjects where a student can learn online by themselves.
This reduces the effort educators need to spend on analysing progress. Instead, teachers have more time to produce course content and more time available for their students. Continual monitoring can optimise the balance between online learning and direct teacher-student tuition. This results in an environment where everyone works better together.
Be careful what you ask for
AI has the potential to provide significant benefits in the education sector. Still, the challenge is not purely technical - we need an ethical framework to guide us. The potential for discrimination and bias is never far away, and we must consider how to address this risk. Algorithms cannot apply judgement... or common sense; we need humans to help here. For example, ask a robot to avoid hitting its surrounding walls, and it may decide that standing still is the best course of action.
There is an unavoidable trade-off between fairness and accuracy - with no absolute right position on the line that connects them. Our goal may still be to optimise a student's ability to learn - with the added constraint that everyone has an equal opportunity to do so.
The diagram shows a blue boundary called the Pareto Frontier. We can eliminate any point that does not lie on this line because we can improve the fairness, or accuracy, or both by moving to a point on the line. However, the Pareto Frontier does not help us decide which point on the line to choose. That choice comes down to the relative importance of accuracy and fairness, given the context of what we are analysing.
The balance between fairness and accuracy depends on context. For example, for medical diagnoses, choosing accuracy over fairness may provide 'better' results. When processing university applicants, stakeholders may prefer a greater emphasis on a fair and ethical approach. When we try to maximise accuracy across multiple populations, the majority contributes more to the overall outcomes. As a result, people in majority groups benefit more, and this may not be what you intend to happen.
The Pareto Frontier gives us a way to measure the trade-off between accuracy and fairness. The need for human judgement and ethics is still an essential part of the process - primarily where qualitative factors exist.
Nothing is anonymous
Protecting people's privacy is a fundamental responsibility for anyone handling data. The importance of privacy within education – especially for younger learners – is paramount. Less data almost always means less accuracy – this is another trade-off that we must manage. For example, if we change someone's age from 22 to a range between 20 and 30, then at least two things happen:
We reduce (although not necessarily remove) the ability to identify someone by using age as a contributing factor.
We most likely lower the accuracy of a data processing algorithm that relies on age as an input feature. Or at least increase the level of granularity with which we can classify people within our datasets.
Privacy is not free of charge: the bigger your privacy constraints, the more significant the impact on the value of your data models. Be sure to consider defensive and offensive elements when building AI and data assisted services. The diagram below shows a comparison between data defence and data offence.
The concept of differential privacy provides a way to balance privacy and accuracy. It states that the risk of harm or exposure increases by only a little amount because of using individuals' data. Of course, the question is: what does 'by only a little' mean? This is another judgement call that stakeholders need to make.
Information can exist across the records of many people. So, something that you cannot infer from a single record becomes apparent when you have multiple records available. These are essential considerations that form part of an overall data strategy.
A little about Objectivity
Objectivity is a values-driven IT outsourcing partner creating Win-Win outcomes for our clients and stakeholders. We cut through today's complexity and use whatever technology it takes to get you where you want to be. In short, we work in partnership with you to help you solve your business problems.
As a mature organisation, our ethical framework guides everything we do. We are socially engaged and always willing to help. Our goal is to help our clients grow in a focused and sustainable way. We have an agile and people-oriented philosophy. For our clients, this means we combine flexible resourcing with effective working relationships.
Want to know more? I'd be delighted to understand your thoughts on AI in Education. What key issues do you feel could be helped through the use of AI?
Are you getting the best outcomes from your data? Please click below to find out more.