How To Eat an Elephant?
Taking one step at a time to solve wicked problems using AI
In december 2019, a number of strange pneumonia cases were reported to the WHO by the Chinese authorities. By mid-2020 over 10m cases have been identified worldwide, half a million people died, and the end is not in sight. On top of this wave of disease, social unrest has been mounting in multiple countries over crucial systemic issues like racism and social inequality.
This double drama made it easy for us to forget that until the pandemic started, we were worried sick about bushfires in Australia, the possibility of World War 3 starting over a row between the US and Iran, and major protests in Hong Kong.
One disaster after another, mankind seems to take a beating. And so does our planet. If we zoom out, the picture of a planet in distress comes to mind. Global warming hangs over us like a looming threat. Its a whole set of problems for which no easy way out exists. Even the proposed solutions to the problems pose problems in themselves. These types of problems have a name: wicked problems.
It is clear that the wicked problems of our times have outgrown and outpaced the capabilities of us mere mortals alone. We need help. Yet it will not come from above. We float on a big rock in the middle of light years of solitude.
We can get help from within, though. Applied Artificial Intelligence can help us to tame the complexity that wicked problems pose. Through machine learning and deep learning, problem solving tasks that we deemed impossible before, are now within our grasp.
Yet we should not perceive AI as a silver bullet, or a catch all if you will, for the huge challenges we face. AI when applied right, is working hand in hand with humans, in a step-by-step approach, to inch closer towards a better future continuously.
A wise saying from Francis of Assisi teaches us how to eat an elephant: one bite at a time. That’s also how applied AI works. We take small bites out of a problem, until its manageable enough to rid it altogether.
This article will lead you through a series of examples in which artificial intelligence has offered a practical way forward, one bite at a time, without major side effects.
To free up land to feed the world, and to provide building materials, trees are chopped down faster than they can regrow. In 2020 alone, this has resulted in over 14.5m hectares of cut or burned forest. The effects are devastating. In the next 25 years, 28.000 species will have gone extinct due to deforestation. By 2030, we might only have 10% of the original rainforest cover left. And in terms of climate change, deforestation is responsible for 15% of global carbon emissions.¹ A swath of primary rainforest fulfills too many essential roles for our worldwide ecosystem to let it just disappear.
Yet this problem is wicked as can be. While laws exist to prevent illegal logging, enforcement is near impossible with current techniques. Many acts of illegal deforestation are also performed by those at the bottom rungs of the societal ladder, as a bare necessity to get by. And on top of that, powerful multinational corporations have a bigger stake in maintaining the status quo, than in stopping deforestation altogether.
So where does one start? Since before we can do anything, we need to know where it happens. AI has proven to be of help here. By choosing a specific area of forest, and bringing together multiple sets of public data, illegal deforestation becomes predictable. That is exactly what AI startup Hemisphere has done together with the University of Wageningen. Future acts of logging become visible through hidden patterns in data, like the fact that logging usually takes place once an inroad has been built into a primary rainforest, or when certain permits have been requested. By putting these facts into a deep learning model, it can advise areas forest rangers should pay special attention to. At least now, enforcement can take place, bad actors can be named and shamed, and root causes can be tackled where they matter most.
Hemisphere’s model has already helped rangers in Borneo to predict the locations of 40hm3 of illegal logging, and promises to do much more.
No problem is more wicked at this moment than Global Warming, or rather Global Heating in some areas of the world. At the time of writing of this article, the Arctic is on fire. Unmatched drought has made tinder boxes out of the Siberian Taiga Forests, while the permafrost of the Tundra is melting, releasing unknown quantities of methane into the atmosphere. These emissions in themselves exacerbate the greenhouse effect, in turn leading to even bigger problems down the road.
Facing such a disastrous vicious cycle, one might be utterly gutted and unmotivated to keep striving for positive change. It’s a true elephant in the room. Again, applied AI, specifically Deep Learning, can help to eat the metaphorical elephant. Not with huge silver bullet solutions, but by gradually taking away the source of the problem.
In the Netherlands, Deep Learning models are applied to speed up the transition from gas heating to CO2 neutral households. For example Feenstra, a energy service provider, first focuses on improving energy efficiency of traditional gas heaters, while offering solar energy installations, and then helping households to find the perfect heat pump solution.²
For this transition, Feenstra relies heavily on data and Deep Learning models. As we speak, 15.000 households with rental heaters are monitored 24/7 through sensors. With the accumulated data from these sensors, Feenstra can predict exactly when a heater will become less efficient, dispatching a service mechanic before inefficiencies arise.
In a next step in reducing greenhouse emissions, its essential to make heat pumps as attractive as possible for as large a group of households as possible. One barrier for consumers to adopt heat pumps is the fear of losing comfort. A barrier that can be overcome by using AI. When combining the real life heating characteristics of a home with preferences of the consumer, house characteristics and weather data, a model can predict what CO2-neutral heating solution is ideal. This not only means that we can pinpoint what solution to use, in order for the consumer to achieve equal or better comfort levels than with gas. It also means that this pinpointing can be done at scale, and cheaply. A big driver of the price of heat pumps is the sales process needed to convince the consumer. By automating the advice part of this sales process, costs can be lowered substantially.
Its not a catch-all solution, but its scalable to every household in the world, substantially helping to speed up the energy transition.
No article like this can be concluded with a mention of the biggest pandemic in 100 years. COVID-19 is the disruptor of a lifetime. As a pandemic, it impacts all 7 billion inhabitants of our planet. Solving the puzzle of the epidemic, so we can all go towards a new and better normal, is key to get our focus on issues like Global Warming again.
While a large chunk of the challenge that COVID-19 poses is tackled by traditional medicine, and civic cooperation through social distancing and washing hands, the scale of the problem stretches the resources and resilience of our societies. Help from a scalable resource would be a welcome addition.
Its becoming a cliche, but AI can be our partner in fighting this pandemic as well. The three biggest priorities in solving the crisis are 1) successful identifying and tracking those who are contagious 2) adequate medical help to prevent people from getting seriously ill, and treating those who do become hospitalized, and 3) achieving herd immunity through a vaccine. For all these priorities AI offers tangible solutions.
Successful identification of contagious COVID-patients can be done in multiple ways. Through an app tracking symptoms through questionnaires in large parts of the population, and by combining multiple open data sources about movement, events, restaurant reservations, PCR test results, and doctor visits to predict the spread of outbreaks.³ In medical triage, detection can be done by deploying AI to visually diagnose COVID symptoms in chest X-rays. Another, more granular, way of tracking the spread through AI, is by using wearables from Garmin, Apple and FitBit to identify typical combinations of symptoms. These symptoms are a rising resting heart rate, temperature and breathing frequency, deteriorating sleep, and lowered blood oxygen. With 70m smart watches being sold annually worldwide, this can provide a solid sample to predict infection patterns in parts of the population.⁴
For adequate medical help for those who need it, deep learning can be used for two main purposes: predicting which patients are most at risk of becoming seriously ill, and identifying the right treatment. By combining the genetic and medical information of patients who become seriously ill, and unleashing a Deep Learning model to detect patterns, we can predict patients at risk. Once a patient is admitted to hospital, a genetic test and medical record check can quickly help to show what path the disease will take in the patient. This can help focussing resources on those who need it most. The process to design treatments is already being sped up by using neural networks to rapidly test medication, delivery methods and dosage.⁵
These same methods for rapid testing are also being used to solve the third priority: a vaccine. Most pharmaceutical companies and research institutes have by now openly shared their data sets and AI tools to more effectively battle the pandemic together. For example, the Allen Institute of AI has partnered with several entities to set up the COVID-19 Open Research Dataset (CORD-19).⁶ By combining datasets and tools, AI is used to predict mutation patterns to ensure that we develop the right vaccine, but also to identify compounds that prevent the virus from binding to human cells.
AI and its methods, like advanced analytics, machine and deep learning, are no futuristic tools anymore. The techniques can be applied to almost any hard problem nowadays. The most important takeaway is that it is not a silver bullet for the elephant in the room, but a partner of society to tackle problems one bite at a time. While taking these steps, and using open source tools, the AI tools are getting smarter and can be used for bigger problems, or entirely different problems. This not only fosters a better future, it also offers attractive revenue models for businesses. Models that we will dive into in the next article (so don’t forget to follow us). By using AI we supercharge humans for positive change, letting business, society and the environment flourish.
At Hemisphere, we are helping organizations to use AI for a positive impact on the bottom line and society. Together with them, we create scalable solutions to wicked problems. We already assisted KLM, Vattenfall, Feenstra, the Dutch National Police, Deloitte and Epcor with our approach to AI. We offer a streamlined innovation process that focuses on building a working solution fast. We understand AI, and we understand business. Our seasoned experts possess all major AI skills to translate your ideas into actual solutions. This enables rapid testing, improving, and value creation. Want to know more? Lets have a chat! Contact email@example.com.