Artificial intelligence is gaining traction in areas of social responsibility. From climate change to social polarization to epidemics, humankind has been seeking new solutions to old but persistent problems. From a technological point of view, the amount of daily data produced in the digital universe now allows for state-of-the-art approaches, which may lead to innovative solutions in these underserved areas.
AI for social good turned into a reality for us at Tryolabs after we collaborated with an NGO to improve upon how African lions are tracked, which helps with species preservation. We will go into more detail on that timely case, especially as wildlife conservation faces the immense challenges posed by devastating megafires threatening the lives of millions of animals in historic ways.
Further, we’ll cover some applications of AI for social good in fields such as education, urban planning, healthcare, and wildlife conservation (where we’ll discuss our journey with lion tracking). These are all subjects that have been widely examined with technology, as their profound global impact only grows.
These examples serve as the foundation for our search to figure out how to do more with AI for social good. This led us to build a framework for socially responsible contributions. A collaborative approach is a major part of these solutions, since a primary constraint on tech investment in this domain is cost. Therefore, we encourage you to share your project, research or idea with us, no matter the stage you are at, as we are very interested in any new AI implementation for social good!
With that being said, let’s now dive into the examples! ✨
How predictive analytics is helping in cancer treatment
One of many relatable AI applications for social good pertains to the healthcare sector, as it may personally affect us all. Cancer causes 1 in 6 deaths globally, an estimated 9.6 million people died in 2018, and over 300.000 cases are diagnosed each year. As the disease keeps spreading, AI approaches for healthcare are quickly establishing revolutionary solutions in this space.
Predictive analytics techniques are being applied to healthcare scenarios, in order to evaluate clinical data and anticipate future trends. One of the main advantages of predictive analytics in this field is in improving the accuracy of diagnosis and treatment.
PathAI, a technology provider for pathology laboratories, found a strong correlation between AI-powered and manual qualification of one particular protein across various tumor cells. The PD-L1 protein keeps immune cells from attacking non-harmful cells in the body, and since some cancer cells contain high amounts of this protein, they can deceive the immune system and avoid detection as harmful agents. Thus, they are not attacked. By precisely identifying that protein, health professionals have a greater understanding, which in turn gives them the ability to determine if a patient is more likely to develop cancer or not.
This AI pathology interpretation in protein qualification can help predict patient outcomes more quickly, and with greater precision.
An AI model prediction was evaluated and compared with manual assessments by a network of pathologists to discover whether the platform performed consistently. Researchers trained a model with more than 250,000 pathologist-provided annotations, which eventually performed successfully at the level of a certified pathologist, as stated in the press release. This finding may save vital time and resources by forecasting the probability that cancer cells will spread in a patient.
Deep learning applied to improve sanitary conditions
Also related to livelihood and health conditions, over a billion people in the world are living in urban areas lacking basic sanitation services, water, and/or electricity. It is expected that as the global population grows by the millions, 1 in 4 people on the planet will live in an urban settlement by 2030, without access to essential services.
Bangalore is one of the most crowded cities in India. Home to more than 8 million people, around 8% of the city’s population lives in slums. This reality inspired deep learning research that strives to segment and detect those geographical movements.
In this study, researchers found that the first step in rehabilitating crowded areas is by mapping and monitoring field dynamics. Previously, those tasks were carried out manually by human annotators and consumed a vast amount of time and effort. The focus of this application was to automate an inefficient process used to identify changes in satellite images. Doing so will make it easier to monitor how those geographical areas evolve. The study explored the potential of fully convolutional networks (FCNs) to analyze the temporal dynamics of small clusters of temporary slums using very high resolution (VHR) imagery.
A deep learning approach was used to segment and map individual slums in satellite imagery and a change detection and monitoring system was created.
The results showed that between 2012 and 2016, slum areas changed trajectory; many disappeared, but almost the same number showed up in new areas. Some of the findings revealed that slums didn’t increase in size, but were displaced after short periods to vacant urban areas.
This new information about how slums vary over time can aid in long-term urban planning and management, and help determine the best way to provide the affected populations with the essential services they currently lack.
Making education accessible and personalized
Education is another essential aspect of human development, and around 617 million children don’t reach minimum competency levels in reading and mathematics. Of those, 265 million are entirely out of school.
One of the most popular trending AI applications for education has been the adoption of voice assistants in the classroom. This enables children to interact with academic material and assistance, without the necessity for the physical presence of a teacher. It also opens new doors in the educational sector. Additionally, these tools are accessible to the blind and low vision communities, which constitutes yet another application outside of the strictly educational context.
SeeingAI is a renowned project by Microsoft. It’s a free app that narrates the world for those who are vision impaired. Using VoiceOver, it provides information about who and what are around the person. Defined as an “intelligent camera app”, it has numerous functions, from reading short texts and documents to scanning products and identifying people, currency, colors, and more.
This mobile app combines different AI techniques and processes together, such as optical character recognition, a barcode scanner, and object recognition.
SeeingAI can scan a barcode and identify the associated product, as well as fetch related details like directions or ingredients. What’s more, the app can describe people in terms of their presumed gender, perceived emotional status, or estimated age.
Regarding personalization in learning, we want to mention Plan Ceibal. UNESCO has featured it in its paper on Artificial Intelligence in education as “probably the most advanced state agency devoted to digital education from the region,” specifically for its PAM solution (Mathematics Adaptive Platform).
Plan Ceibal was created in our home country, Uruguay, to support educational policies with technology. Currently, every child in the public education system receives a computer for personal use with a free Internet connection at school, per the program. One of Plan Ceibal’s latest initiatives is PAM.
PAM is a tool that provides personalized feedback based on each student’s skill level via an analysis of the student’s experiences.
The platform identifies areas of improvement for each student and suggests exercises and activities accordingly. With over 25,000 step-by-step exercises and 2,800 feedback patterns, it equips teachers with an adaptive mathematics content approach that provides automatic reporting and can assign follow-up tasks no matter the pace of learning or the level. You can find more details on Plan Ceibal’s vision of AI in education on their latest paper.
Our experience with AI techniques in wildlife conservation
Recently, we had the opportunity to collaborate with LINC - Lion Identification Network of Collaborators - a community of conservationists and researchers who seek to preserve the African lion population.
It is expected that 38% of the species as we know of will be extinct by 2100. In fact, just a few days ago, the first animal extinction of the new calendar year was confirmed: the Chinese paddlefish. This threat applies to Africa too, as Panthera leo has lost over 40% of their natural terrain over the last 20 years. The loss of land has forced animals to roam more extensively across a very fragmented landscape, which difficults the tracking efforts.
Several AI applications for wildlife conservation include computer vision solutions as their primary thrust. In particular, image recognition and classification processes have had significant positive animal preservation results. Our approach was along the same lines, as we wanted to further develop the non-invasive solutions used in LINC to help monitor lions, one that moved away from manual identification to an automated system.
Historically conservationists identify lions by manually analyzing precise whisker patterns in each animal. Facial whiskers do not change frequently throughout an animal’s life, and this makes it possible to recognize each individual animal over time, by observing their whisker patterns alone. This thorough examination was performed by conservationists, who had to compare whisker photographs with a predefined grid and isolate the unique design that matched the specific animal being analyzed from a database of over 400 lions.
We became involved in this project in order to help further improve and develop these automated solutions.
The tool chains consist of computer vision and pattern recognition algorithms that can automatically perform the different methods for lion identification: face and whisker id. This reduces time spent and the number of human resources required allowing conservationists to viably use large data sets in their work.
Soon, we will be sharing the complete case study involving our participation in LINC’s project. If you’d like further technical details, you can always give us a shout.
To sum it all up
The abundance of data and deteriorating climate conditions make the case for harnessing AI for social good. New social initiatives relying on deep learning or machine learning techniques to address issues are thriving across a range of industries. Computer vision and predictive analytics approaches may help us ensure a better environment for everyone some day.
From predictive analytics in healthcare, to tech-based alternatives in education and image recognition tools in other industries, there is a broad scope to how artificial intelligence is being applied for social good.
We got inspired by a large number of other cases, like Google Bolo also in AI for education, or Guillermo Sapiro’s research in autism that we keep coming back since the last Khipu. It wasn’t easy to narrow it done to just these few, so you surely have preferred cases that didn’t make it to the list.
We are extremely eager to explore new AI projects for social good. If you know of any, or want to share your research, please leave a message below! Or if you’re interested in taking the next step with your idea or ongoing project, let us know. Let’s collaborate!
Like what you read?
Subscribe to our newsletter and get updates on Deep Learning, NLP, Computer Vision & Python.