Using machine learning to discover innovations that can boost child development
This article was written in collaboration with Karlis Kanders, Laurie Smith, and Natalie Lai
In the UK, children from lower-income backgrounds tend to enter school with a lower level of attainment compared to their peers. Nesta’s Fairer Start mission is dedicated to narrowing the outcome gap between children growing up in disadvantage and the national average.
When supporting early childhood development, the earlier intervention takes place — whether that be for special educational needs, or facilitating improvements in the home environment — the better that child’s chances of excelling at school.
Alongside finding ways to improve the home learning environment and helping local authorities to innovate with their services, we also want to consider the role of emerging technologies. The prevalent methods of assessing child development, although accessible, are largely low-tech and come with inherent limitations.
Could trends such as recent advances in artificial intelligence and growing venture funding for products and services aimed at parents and children pave the way for novel, impactful innovations for supporting child development?
Our new project aims to identify emerging technologies that could be used to detect or anticipate young children’s developmental needs in a more timely and nuanced fashion than is currently possible.
For example, AI-powered speech recognition has shown promise in early detection of dyslexia.
Using our data-driven horizon scanning approach, known as Innovation Sweet Spots, we will analyse various datasets capturing research, investment and public discourse trends around innovations relevant to child development.
This will allow us to make more informed calls about which innovations are close to a tipping point where they can have impact in the real world.
Casting the net for innovations
Applying our Innovation Sweet Spots approach, we will prioritise getting a bigger-picture view of the research and development landscape, initially focusing on trends in academic research publishing and patenting.
We expect that research publishing data will primarily reflect trends in fundamental research, highlighting innovations that will be able to make an impact further in the future. For this purpose, we will use OpenAlex, a large, open-source dataset of publication data, including research paper summaries and metadata such as the authors’ affiliations and number of citations.
Meanwhile, patent data might cover more mature innovations that are closer to commercialisation. Here we will explore openly available datasets like Google Patents and Logic Mill.
Analysing the signals
The analysis will utilise natural language processing and will have three main parts: exploring the data and defining an innovation taxonomy; curating the data using supervised machine learning; and performing descriptive analysis of the trends.
The innovation taxonomy will allow us to label texts as being related to different areas of either child development (eg, physical health or literacy) or technologies and innovations. We will aim to characterise technologies and innovations rather loosely so as to be open-minded about what we may find — these might include things like speech recognition, wearable devices or phone apps, but there may also be process or social innovations like checklists.
Creating this taxonomy will first involve exploring a subset of the research and development data, which we’ll select using a broad set of keywords related to child development. We’ll then use unsupervised machine learning methods such as clustering to identify more specific themes in the publication and patent data, which we can build upon to define our taxonomy.
We will then use this to create training data and train a supervised machine learning model to help us reliably distinguish research on different aspects of development and different types of innovation. We will consider different approaches for supervised learning such as text classification or named-entity recognition. Importantly, we will also be able to use the trained models to label new data in the future, allowing us to eventually monitor trends in this space.
Assessing the risks and impacts of innovations
Besides the trends analysis, it will be equally important to consider the range of impacts these innovations could have.
We are mindful of the potential risks related to this area of innovation. The increasing variety of smart technologies in the home has caused a UK Parliamentary select committee to issue a warning about how such tech may be exploited by abusers. Many parents will also be uneasy with the potential scale of data that smart technologies may collect from their children, as well as with the idea of their child’s development being measured automatically rather than by expert practitioners.
Moreover, there is a need for sensitivity and empathy when assessing development, rather than placing children in boxes that are too narrow — do we trust technology to make these potentially life-changing judgement calls?
Finally, we will also need to consider whether the technologies and innovations highlighted by this research work well for children from minoritised communities, and whether they carry any risks of further exacerbating existing inequalities in early childhood education.
What’s next?
Once we’ve made the first step of surfacing innovation trends, we hope to continue this work by exploring the attitudes of caregivers towards the promising innovations we identify and by conducting a more in-depth analysis of the potential impacts and future consequences of those innovations.
At its most ambitious, this project could contribute to a captivating vision of a more precise and timely approach to support child development, leading to ideas on how we could transform our current development screening and management procedures.
We’ll be publishing updates about our work here and on Nesta’s website. If any part of this sounds interesting to you, please get in touch.
We’re keen to hear from a diverse range of voices, particularly:
- Experts in early years development
- Those who have worked with the mentioned datasets in data science project
- Parents and caregivers
Thank you to Tom Symons and Faizal Farook for reviewing and editing this article.
