PREDICTIVE ANALYTICS & MACHINE LEARNING: WHAT ARE THE REAL PROBLEMS WE NEED TO SOLVE?

What a time to be alive. As I write this, an electric sports car is hurtling through space, having just beamed selfies down to earth. The very same cars can even autonomously drive, park, and occasionally rear end other vehicles. My relatively low budget smartphone I have just clumsily juggled while typing this (and subsequently broke) has over 1300 times more processing power than the computer that landed Apollo 11 on the moon (we’re apparently still refining how to make crack resistant phone screens though). Every day, we generate at least 2.5 Quintilian bytes of data on the internet – that’s almost 600 million Encyclopaedia Britannicas, for those of us old enough to know what those are. There is even this robotic dog that can freakishly open doors and escape:

A postman's worst nightmare, courtesy of Boston Dynamics

Technology moves fast, and if renowned futurist and computer scientist Ray Kurzweil is correct, will only get faster as we approach “technological singularity” (runaway technological growth). He states "the pace of change will be so astonishingly quick that we won't be able to keep up, unless we enhance our own intelligence by merging with the intelligent machines we are creating".

There is now no doubt that the heart (or brain) of these technologies, Artificial Intelligence (including Machine Learning and Deep Learning), is heralding a new chapter for human existence: another industrial revolution that will bring into question what most of us are doing with our lives. However, just like any technological revolution in the past, jobs will change rather than be destroyed. Just as the switchboard operators, street lamplighters and pinsetters have made way for more elegant technologies, new interesting careers have developed. Even just in the last 10 years new and highly paid jobs have developed, such as app developers, driverless car engineers, and drone operators (we’ll explore this more in another blog post).

Pinsetters from a bygone era (Photo Lewis Wickes Hine, 1910)

However, is technology really solving the problems we need to solve today? We know that techies fall in love with the technology, and billionaires love to play with their toys, rather than taking time to understand real problems faced by real people. Is the deluge of data being generated by social media really a benefit to anyone besides start-up founders and venture capitalists? Silicon Valley is at the heart of solving frivolous first world problems such as who’s going to deliver that pizza, or who’s going to drive me home fastest after a night out. Surely, with all the capability and computing power we have globally, it is possible to solve some of the more complex problems plaguing the lower income third world countries. In this post we explore some of the problems that the world faces and where AI can be part of the solution.

(Note for ease of reference, I use the abbreviated term AI for Artificial Intelligence, although most of these examples make specific use of Machine Learning, or Deep Learning which are subsets of Artificial Intelligence)

Challenges

When it comes to implementing technology, developing countries have the significant challenges of an unreliable infrastructure. Whether it is a consistent power supply, data connectivity or just the sheer distances and rugged terrain, engineers and data scientists need to have a very different approach to solving these issues than if it was for the yuppies of London or San Francisco. That being said, developing countries are ripe for innovation, and technologies in areas such as mobile payments (e.g. M-Pesa) have taken off, partly due to such a strong need for a practical and simple solution. Software that runs in the cloud, or even simpler versions of Machine Learning built for low-powered mobile devices (e.g. “Tensorflow Lite”), are such approaches to overcome these barriers.

We look at three areas which I believe are vitally important to third world countries, and some examples of where AI can make a big difference.

Smallholder Farming

Developing countries are highly dependent on agriculture. Smallholder or subsistence farming accounts for a large proportion of a country’s livelihoods and sustenance, and unfortunately can be very susceptible to disaster. Artificial Intelligence can help in many areas to help alleviate these issues:

• It can help with increasing yields. By analysing soils, terrains and weather patterns, predictive recommendations can be made on what to plant where and when. This will also feed into national development policies and subsidies.
• Much of the above idea can be automated by using drones driven by AI, from the analysis stage by taking detailed aerial photographs of the land, all the way to even physically planting crops.
• Crop disease can be devastating. By taking photos of crops and running them through a Convolutional Neural Network, an algorithm can identify disease and recommend treatment before it spreads to a wider region and decimates a season’s crop.
• Better predictions of natural catastrophes like droughts or floods mean that resources from NGOs and charities can be deployed much more efficiently with forward planning.
• Agriculture insurance can be highly valuable to a smallholder farmer. If disease, flood or drought were to hit, this insurance would make a huge difference to last out the season until the next harvest. The advanced analysis and prediction techniques will be able to better price and hedge these risks, ultimately giving a better deal to the farmer and making the insurance offering more sustainable. 

Drone Photography and Analysis, Source: Aerial 9 Media

Healthcare

Access to reliable, affordable healthcare is a critical global issue. In developing countries, access to very simple preventative medicine and techniques can make a significant difference, especially in countries still ravaged by epidemics from malaria, HIV and Ebola.

Financing healthcare in any country, let alone developing countries, will always be problematic and therefore keeping costs as low as possible is an imperative. Automation will be key in such areas:

• Diagnosis – by asking some simple questions and even analysing images of a person, automated diagnosis is becoming very accurate. This will ensure a patient with a relatively minor issue receives the level of care that is necessary from a nurse or health worker, freeing up doctors and specialists to focus on the more serious cases. This can all be done from very remote areas with the use of a mobile phone prior to a visit from a mobile clinic, speeding up the process and increasing efficiency.
• At a more macro level, outbreaks can be predicted more accurately in advance, so that resources are deployed more effectively preventing larger impacts. There are models already predicting which animal species are more likely to be harbouring viruses that may infect humans. Such “Zoonoses” include Ebola, HIV, and many strains of flu.
• Accessibility to health education – AI can tailor health education to individuals by placing emphasis on key topics for that region by utilising outbreak predictions, as well as automated language (machine) translation.
• In the field of micro insurance, advanced predictive models will help in understanding how demand can be met financially, therefore increasing sustainability for these offerings.

Financial Inclusion

Developing countries still have very large unbanked and uninsured populations, putting individuals at risk when unforeseen expenses arise. Along similar lines as above, AI can help in many areas:

• Improve accessibility through automated customer service chat bots that can speak the same language.
• Tailored and automated financial education and processes geared towards lower income families.
• Better financial models, be it credit models in banks or mortality models for insurers, that take into account different forms of data that may increase accuracy in those countries where traditional data is scarce or often non existent. Where a customer today is turned down for a business loan or insurance policy due to lack of data history, using these new methods will open up markets which is advantageous both to customers and the financial institutions.
• Significant avoidable costs borne by financial institutions, such as fraud, can be more accurately detected and avoided using AI. By decreasing fraud and other unnecessary costs, these savings can be passed on to the customer.

This is just scratching the surface, and there are many more ideas where AI has the potential to impact developing countries. While in general, technology is already making considerable inroads, predictive models can turn it into something truly ground-breaking, far from the doom and gloom warnings from the likes of Elon Musk and Stephen Hawking. While some start-ups are already starting to explore this space, I hope that in the coming technology revolution many more will find ways to sustainably and responsibly contribute.

What do you think are the key issues that we can solve with AI?

Feel free to add your thoughts in the comments.