Least-First Machine Learning: 15th February, 7.45-9pm (GMT)
Why is Dignity having this conversation now?
Machine learning technologies are becoming increasingly pervasive, such that we regularly, inadvertently encounter and engage them. Everyday online tools such as search engines, product recommendations and sentence completion in our emails are all powered by machine learning. Other useful digital services such as face detection, spam filters, fraud detection and route planning depend on machine learning, where machines filter information and decisions on our behalf. As part of Dignity’s digital technology vision, we could see cases where we might consider adopting machine learning technologies. For example, machine transcription of voice recordings from meetings and conversations could help compile minutes, stories and learning resources. Additionally, machine translation could help us to get resources to more people and communities, in their own native languages, faster.
However, the relationship between machine learning and alleviating poverty, part of our core mission, remains challenging. Firstly, much of the groundwork of gathering and classifying training data for machine learning models is often outsourced to poorer nations, which could lead to unfair pay and exploitation, if not properly regulated. Secondly, people can unknowingly provide personal data such as their voices, connections and habits, to machine learning training and processing without receiving any compensation. Finally, reliance on machine learning services could present minimum requirements on networking and devices that make them unaffordable for people accessing our resources.
Our reliance on machine learning technologies is hard to ignore but increasingly easy to overlook as they disappear into the background, becoming indistinguishable from human decision-making and action. By shining the Least-First spotlight on machine learning at this point, we will position ourselves to step forward in a way that aligns best with our core aims.
Scroll down to read more about machine learning and Least-First design…
What is Machine Learning?
Machine Learning is an approach to “teaching” computers how to solve problems with examples, instead of programmers writing the instructions. This doesn’t exclude programmers or coding from practice, it means that programmers instead write code for how to learn to solve a problem rather than how to solve a designated problem. This might seem difficult to conceptualise but consider the way we generally learn as humans: we don’t need to see all the cars in the world to recognise an individual vehicle as a car. After we have been taught by our parents or teachers with enough varied examples and images of cars, we carry on recognising cars of various shapes, sizes, colours and brands. Machine learning scientists have found methods of mimicking this ability using large, complex networks of statistical calculations, where each node in the network has a responsibility for observing a unit from a complex input (e.g. a pixel from an image, a word from a sentence, a sample from a sound file) and determining if it is relevant or not. The collective calculation of relevance is a classification of the overall, complex input. As an example, a Machine Translation model is been presented with examples of words in one language (e.g. Bemba) in different contexts mapped to translations in a target language (e.g. English). Once trained with sufficient examples, the machine translation model can evaluate, classify and translate new sentences and text that it had not previously been presented. Other types of machine learning include:
- Machine Vision: map images in the world to text labels
- Speech to Text: map spoken word to written text
- Text to Speech: create synthetic spoken word from written
- Summarisation: condense a large body of text into a short, coherent report
- Game Playing: enabling a machine to make seemingly intelligent game-playing choices (e.g. chess, GO, video games) and compete with human
- Autonomous Vehicles: enabling cars to navigate roads and road signs without human intervention
Machine learning continues to progress given advances in computational processing speed, storage and networking. Machine learning services and applications are becoming increasingly commonplace and integrated in many aspects of life. Our reliance on these technologies is hard to ignore but increasingly easy to overlook as they disappear into the background, becoming indistinguishable from human decision-making and action.
Why is Least-First Design relevant to Machine Learning?
The field of machine learning ethics continues to develop and typically refers to diversity of training data, transparency of algorithms, operational safety, fairness of decisions made by machines and clear justice for cases of unfairness and harm. Applying Least-First Design to machine learning compliments work in ethics, by reviewing the contexts within which machine learning approaches are selected, the nature of socioeconomic bias in training data and thought to the human-machine interaction required when machine learning systems are in training and operation. There are known initiatives where machine learning approaches are being actively applied to alleviating poverty and providing insight into agriculture and resource management in smaller economies. Least-First Design could help to motivate more of these opportunities, by allowing communities with the least access to wealth and resources to express their needs from automation and machine assistance.
Dignity exists to help people know Jesus and to fight poverty together. We have been asking the question, “How can we better use technology to increase speed and decrease costs of reaching, partnering with and delivering publications to people on the outskirts of society?” Least-first design has become a key component of our emerging discoveries. To further our explorations we are inviting interested and interesting voices to join the conversation through Least-First X.
The United Nation’s 2030 Agenda for Sustainable Development includes the eradication of poverty. Still we see a growing divide between wealthy and desperate, while the global technology market is predicted to continue accrual by trillions of dollars annually.
We believe it is time to rethink the design, development and delivery of technology solutions globally, if we seriously want to eradicate poverty and economic exploitation from our world. The principle of least-first design offers a framework for identifying and committing to ways of bringing benefit to the poorest in society from the onset of technology project or solution planning. It brings the needs of the least in society into the centre of ideation and emphasises that it continues as a matter of urgency, even as technology organisations and teams proceed through discovery, design, development and delivery. Least-First Design requires servant leadership and sacrifice of our natural inclination to pursue progress at the expense of others, especially those with the least access to wealth, infrastructure and education.
The Least-First X series is a gathering of technology professionals, enthusiasts, influencers and consumers, interested in contributing to conversation and action that challenge the technology industry to consider and rethink its impact on the poorest in society. During each Least-First X Session we will have an invited speaker to talk about a project or product in planning, design, development or delivery phases, highlighting the key technologies and capabilities. There is then a reflection on the three discovery questions of least-first design, pertaining to technology theme X:
1. Who benefits the most?
2. Who benefits least?
3. Who is most disadvantaged?
The discussion then looks at the 5 key expressions of Least-First Design, in order to discover ways that benefit could be delivered to those identified as either receiving least benefit or most disadvantage from the technology under discussion:
1. EMPOWER: inclusion in a participative design and development process of technology as partners, such that their views, needs, concerns, ideas and expectations are reflected and represented.
2. EDUCATE: ways in which relevant education resources could be developed and delivered.
3. EQUIP: chances to exchange useful skills, equipment and tools that make technology more accessible and affordable.
4. ENABLE: derive solutions to actual problems identified with measurable impact.
5. ENACT: actively partner with community-led initiatives and enterprise that may incorporate or develop technology through a least-first process.
The plan is to have an ongoing series of technology themes presented under the Least-First X banner, with the possibility for repeats at any time. The selected technology themes will be based on either emerging or existing technologies, as the Least-First X approach allows both forward and retrospective thinking.
Join the team
Dignity is a rich tapestry of people each bringing their different skills to the bigger picture. Are you interested in adding your skills into the mix?
If so, we would love to learn more about you so that we can connect you with like minded people and opportunities. Together we can support more people as they get to know Jesus and fight poverty. Tell us a little about where your skills lie and how you’d like to be involved.