Machine Learning in Government Operations

So far this century our society has experienced the creation and implementation of an economy based on information and knowledge, transforming the world one day at a time, and creating exponential social and economic value. In just two decades, the transformation of information and knowledge into concrete solutions has become the fundamental equation to generate value, efficiency, prosperity, and wealth.

But before there’s information, there's data. Data is a set of numbers, symbols, characters, words, codes, or graphics that have no logical meaning, but when put in context can be converted into information that does have logical meaning. Data is measured in bits, information in units such as time and quantity. Human beings use information to make decisions, computers use mathematical formulas, programming scripts, and software applications to convert data into information, helping us save our most valuable resource: time.

One of the characteristics humans have is that we try to avoid repetitive tasks and always try to use shortcuts to save time. The mathematical methods we learn throughout life are precisely driven by that desire to avoid repetition. On the other hand, computers are designed to do small, repetitive, and incremental operations.

That’s the reason why it is not an easy task to create a computational algorithm: because we must leave behind how humans operate and think about how machines operate. 

This is shown in the study of computational algorithms, where we have the Big O Notation, which can help us determine how an algorithm’s run time or space requirements grow as the input size grows in a function, going from — Contant: O (1); Logarithmic: O (log(n)); Linear: O (n); Linearithmic: O ((n log(n)); Quadratic: O (n^2); Exponential: O (c^n); and Factorial O (n!). Interestingly, many times it is easier for humans to create a computational algorithm from an exponential function where the operations are grouped, while for the machine, a logarithmic function is easier to execute because it can be broken down into small but repetitive operations.

The application of these computer attributes has transformed the dynamics of how we operate as a society in many ways. The most obvious benefits include the optimization of repetitive processes, the reduction of human error, expedited decision-making, and the availability of services 24/7.

Artificial intelligence

The two biggest objectives of computer science have always been to simplify things for humans, and to reduce the number of resources needed to perform a task; time being the most valuable resource. Artificial Intelligence (AI), one of the key sub-disciplines of computer science, focuses on that: protecting human time, by enabling the machine to make certain decisions for itself. Machine Learning (ML), a subset of Artificial Intelligence, focuses on the use of statistical techniques that allow machines to learn (improve with experience), without having to be explicitly programmed to do so.

Machine Learning is being applied by default to many of the systems and products we use every day, from image and voice recognition, to product recommendations, autonomous vehicles, and spam filters. Machine Learning accomplishes this through three main methods: (1) Supervised Learning, (2) Unsupervised Learning, and (3) Reinforcement Learning. An example of Supervised Learning is the classification of spam, based on data that has been labeled previously, and is used as an example by the machine to determine new spam.

In Unsupervised Learning, the machine identifies patterns in unlabeled data, and tries to organize them based on similarities in their properties. An example of Unsupervised Learning is movie recommendation systems, where recommendations are made based on previously watched movies, by identifying similar patterns in their properties without any prior explicit instruction. Lastly, Reinforcement Learning is frequently used in video games, where the machine learns from its mistakes little by little, and is continuously improved until reaching a high level of prediction based on the statistics accumulated during the algorithm training process.

Machine Learning has optimized tasks that would have taken much more resources to be carried out through a non-ML algorithm, and this is shown in the productivity and generation of significant profits by private organizations that apply it to their products and services. According to consulting firm McKinsey & Company, ML-enhanced supply chain management greatly improves forecasting accuracy while simultaneously increasing granularity and optimizing stock replenishment. It can enable reductions between 20 and 50 percent in forecasting errors, and lost sales due to products not being available can be reduced by up to 65%.

Government Operations

In the last three years, there has been increased interest from governments in understanding how Artificial Intelligence can be applied to the public sector. The U.S. National Security Commission on Artificial Intelligence was established in August 2018, and its current Committee was established in June 2020. On the other hand, in May 2021 the White House announced the launch of the National Artificial Intelligence Initiative, with the aim of expanding federal innovation efforts in Artificial Intelligence, and encouraging the private sector to work together with the government on the design, development, and implementation of public solutions powered by Artificial Intelligence.

According to the latest report from the U.S. National Artificial Intelligence Commission in 2021, although there has been initial interest from the government in the application of Artificial Intelligence for processes improvement, they still need to take massive action. It also highlights that the government should partner with the private sector and professional research and development societies, to carry out an expeditious transformation before 2025.

To achieve a ML-powered transformation in government operations, it is essential for governments to understand the elements that allowed that transformation to happen in the private sector, being aware that the public sector can implement the private sector’s best practices, but also develop its own set of know-how. It is also key for them to identify more experimental uses of Machine Learning for more complex tasks that are only found in the public sector.

In recent years we have been observing and studying experimental applications of Machine Learning in the optimization of government procurement processes, and after having conversed with hundreds of government agencies at the local and federal levels in the United States, and Latin America, we have discovered that even though public staffers in charge of making purchases in different government agencies have databases of registered suppliers, contracts and tenders, it is complex and it takes them a lot of time to understand where to find what they need. 

We’ve discovered there’s a big fragmentation in the systems that inform government acquisitions, specifically, their warehouse systems, their contracts databases, their registered suppliers databases, and their existing suppliers’ ratings and reviews. This fragmentation creates a limitation for them to know if they’re actually buying at the best price in the market at the time they make a purchase. For the government to be able to make a small or large purchase that is fully compliant, it has to carry out a thorough evaluation and compare different pieces of information that are not connected or have not been updated in months or years, so the decision made when they make a purchase, is suboptimal.

Humans cannot process large amounts of data manually, but Machine Learning excels at tasks that are clearly defined and involve massive amounts of data.

To attack this problem we have been studying and working on the initial development of an algorithm that applies Unsupervised Learning to assist public procurement teams.

Specifically, we are preparing a statistical model created from public data of recurring government purchases, organizing the cases of purchases with similar properties, synchronizing databases where the relevant information for the buyers is stored, and enabling an interface capable of helping them find what they need in a simple way. 

Through the K-Means method, a clustering method of vector quantization that divides observations into  clusters, we seek to automatically group semantically similar expressions and thus accelerate the derivation and verification of the intention of a common user. In this way, an authorized user can simply say what they need, and the algorithm identifies whether that is available in the agency's warehouse, in an existing contract, in its list of registered suppliers, or in the open market. With this architecture, the algorithm is capable of making an expeditious analysis layer by layer, of finding what the user needs, regardless of which layer it is in, and of allowing the user to make the purchase immediately, with the same level of compliance.

In the context of predictive analysis for public procurement, the algorithm could take a sample of past purchases and intuitively divide it into separate groups, taking into account properties such as price range, product category, deadline, purchase frequency, among others, and based on the similarities, suggest the ideal purchase option instantly.

When will we see Machine Learning in the public sector at scale?

The use of Machine Learning in government operations has the potential to reduce the workload of public employees in repetitive tasks, enabling the improvement of their processes and providing more precision in the results; however, it is not something that should be applied deliberately to all processes.

My experience working in the study, preparation and initial development of Machine Learning algorithms for the public sector has left me with three key learnings so far:

1. Machine Learning is not the solution for all kinds of tasks in the public sector. At the moment it is ideal for small, repetitive and incremental tasks.
2. Start small. Before allocating resources and amplifying a project that involves Machine Learning, a preparation and experimental testing phase should be carried out based on the initial hypothesis, and refine it little by little.
3. Data is key. For a Machine Learning algorithm to be efficient, you need a lot of data, and it should be high-quality, so the pre-work of collecting and cleaning the data is essential before creating a predictive model.

Starting small is key, but Machine Learning’s impact as a tool for governments is big. An example is in the prediction of traffic congestion. Traffic congestion is a major issue in big cities, and it is difficult to predict. Many governments worldwide are using ML methods to predict short-term traffic conditions, with an estimated prediction accuracy of 90%, helping governments’ personnel take proactive measures to ease traffic congestion across the busiest areas of a city’s transport infrastructure.

The governments' interest in using Machine Learning for the public sector already exists. Now the next step is that government agencies learn from methodologies that have already been nailed by the private sector, and create strategic partnerships that allow a real transformation that our societies can experience in the short, medium and long terms.

About the author

Gerardo Mateo is the Head of Government Software Research and Development (R&D) at GLASS — a Silicon Valley-based software ecosystem company powering high-performing governments.

He's part of the founding team that took the company from $0 to more than $3MM in revenue as a new startup. He's currently leading R&D efforts to unlock Artificial Intelligence (AI) algorithms to optimize governments’ operational processes to make decisions with minimal human intervention. He's co-led the company's operations, and the development of the initial version of the product suite.

Gerardo is an Engineer with a B. Eng. in Computer Science and Technology, graduated with a full scholarship from the Organization of American States (OAS) in Beijing, China at CUFE — considered one of China's key national universities in the fields of Economics, Science, and Engineering. He's led the development of high-impact global initiatives and projects in the fields of technology and innovation in Germany, the U.S. and the Dominican Republic, specializing in the development of prototypes and digital tools to simplify manual processes within complex organizations.

He's also an OAS Scholarship awardee for Language and Science Studies in Mandarin, and a Global Entrepreneurship Summer School scholar — a world-class think tank by the Ludwig Maximilian University of Munich, in Germany. His leadership is also reflected in his previous work. He was part of Youth Service America’s Board of Directors, becoming one of its youngest appointed members for his outstanding work leading strategic partnerships as a Global Youth Council member.

He was handpicked to be part of the 4th International Forum for Young Entrepreneurs of the Caribbean, and was one of the winners of Udactiy’s Challenge — an international competition for young individuals in the science, tech, math, and programming fields, awarded by renowned German-American entrepreneur, educator, and computer scientist — Sebastian Thrun, in Silicon Valley.

Gerardo's work has been highlighted multiple times, including being featured on National Hispanic Heritage Month by the U.S. Embassy, for his exceptional trajectory and for his proven track record leading the creation of high-impact technology solutions and initiatives worldwide.