What Government Agencies Need to Know About Artificial Intelligence
Artificial Intelligence can be a useful tool for government agencies to increase efficiency and advance their mission. But what exactly is AI? And how does it really apply to the government context?The potential of Artificial Intelligence (AI) is limitless if you believe everything you read: sleek digital tools are able to process big data sets to uncover new insights, ease work flows, and lead to quantum leaps in productivity and a happier workforce. You and your team record a jump in efficiency, and you drop the chains of old ways and walk into the light of a new digital era. Ok, maybe it’s not that simple, but every day we are learning about new applications of AI that are making things more efficient. Government agencies are increasingly calling for and using these tools. The President’s Management Agenda (PMA) pushes agencies to prioritize modernization including new tools like AI and Machine Learning. And the U.S. Senate recently passed bi-partisan legislation called the AI in Government Act to push agencies to explore the technology. With ever-increasing amounts of data, the resources required to process it are rising exponentially. These new tools hold tons of promise to solve complex problems. But navigating the proliferating array of AI tools is daunting. If you are a government manager or mission area director looking to advance your mission, or increase data processing efficiency, it can be difficult to know where to begin. To cut through the noise, here are some of the main terms you need to know, and some suggestions on where to begin if you want to explore AI within your agency.
The Basics: What is Artificial Intelligence?AI is broadly associated with the goal to use machines to assist with ordinary tasks that require the human skills of inference. US Senate legislation defines AI as, “any method implemented on a computer, including any method that is drawn from machine learning, data science, or statistics, to enable the computer to carry out a task or behavior that would require intelligence if performed by a human.” The field is producing lots of new terms, including offshoots like “Artificial Narrow Intelligence,” “Artificial General Intelligence,” “Deep Learning,” and many more. These are all attempts, some clumsy, others more useful, to more clearly define the field. It is an exciting time for innovation, as government practitioners and data scientists are, as we speak, designing the tools and writing the rules for this powerful technology. AI refers to the ability to learn (intelligence) compared with the ability to process data more quickly than humans can. AI and Machine Learning are often used interchangeably, but they are not the same. In fact, a lot of what you hear referred to as AI is actually Machine Learning.
Artificial Intelligence vs Machine LearningMachine Learning (ML) analyzes data sets and asks it for specific outcomes. It then uses that data to train the machine to “learn” and categorize data going forward to be able to make assessments on what it sees. AI, on the other hand, is the ability to learn, make the assessment and deliver an outcome without having to be programmed each time for the specific set of outcomes desired. In order to take steps without direction, a machine needs to be trained. An ML application is only as good as the training data used to train it; ask any data scientist and they might mention the “garbage in, garbage out” issue as ML tools still rely on the quality of the data sets with which you are working. So, be wary of anyone promising that ML will “fix” your data problem. This also raises tricky issues of bias that, used incorrectly, can lead to negative outcomes such as racial profiling. For example, what characteristics do you select to train the machine on the likelihood of criminal or terrorist behavior? The City of San Francisco recently restricted the use of facial recognition data by city agencies, citing the risk of bias. And a recent study conducted by the ACLU found that some facial recognition software produces false matches disproportionately from minority groups. Interesting, their facial recognition data did not lead to an increased ability to identify criminal suspects or behavior.
Natural Language ProcessingNatural Language Processing (NLP) has the advantage of being able to work with unstructured data in the form of words, and uncover trends or insights. Commercial companies use text collected in helpdesk chatbots to identify product deficiencies. Think about the notes your doctor takes during a visit as she types electronic health records. NLP tools can take huge amounts of this unstructured data and use natural language algorithms to surface previously-undiscoverable occurrences of selected variables. Amazon has pay-as-you-go services such as Comprehend and Lex that make it easy to use for text-based datasets.
Artificial Narrow IntelligenceArtificial Narrow Intelligence (ANI), sometimes referred to as “weak AI”, refers to a tool designed to do a single simple task and return a result, e.g., to check the weather or update the score of a live baseball game. These tools are not intelligent at all; rather they are searching established data sets quickly and coming back with an answer. Alexa, Siri and Google Assistant feel as though they are intelligent because they understand your question, when in fact they are ANI tools designed to take your question using NLP, search the internet for results, and return an answer based on that search.
Neural NetworksThis is where it gets interesting. Neural Networks, also known as “deep learning”, attempt to reflect the way the brain works by transferring experience and learning from discrete, apparently unrelated experiences. For example, think about how a toddler learns a toy truck requires two hands to lift. When toddling over to another toy of similar size and shape, there is no need to re-learn that this second truck is too heavy to lift with one hand. Learning is transferred from toy one and applied it toy two. Currently, all these tools still rely on data and are not learning how to learn in order to take the next step. They do not yet predict what will happen. However, projecting into the future tools using neural networks may be able to take that step. And, to get really sci-fi: where speech recognition is already used widely today with Alexa, Siri and Google Assistant, the artificial assistants of the future may be able to see what you see, process your environment as you do, and take action as a result using neural network-enabled skills of inference.
Getting Started with AI in Government
- Identify the problem you are trying to solve. How will solving a particular problem lead you to accomplish your goals more effectively? Are you trying to speed up existing processes or lighten the burden on your data team? Or are you trying to crack a problem you are not currently able to crack?
- Determine what kind of data you are using. The type of data your agency collects will drive the solution you choose. A few examples include:
- Statistical data: Education assessment statistics like those we work with under the US Department of Education’s Nation’s Report Card provides a huge amount of numeric data.
- Language or speech data: Citizen contact centers collect large amounts of text and speech data from people who need assistance from government agencies.
- Text data: Agency communications teams collect daily news articles and documents that need to be categorized and reported on. Another example is the huge amount of regulation, codes, and legal guidance produced by government agencies.
- Image data: Satellites collect vast amounts of image data. For example, the Department of the Interior recently requested AI tools to process terabytes of satellite image data of federal lands to analyze the impact of human activity over time. An excellent goal and use of AI!
- Work with experts. Find a partner who can help you think through these issues, prioritize what you really need, and assess options. Start small, pilot a solution, iterate, and document along the way.
- Make sure you are not using a solution in search of a problem. Endemic in applications of technology is the eagerness to use new tools like AI before you are clear about the problem you are trying to solve. You may, for example, “merely” have a data problem that does not require new tech. And many tools are available already that can get you started.
Looking for a partner for an upcoming AI project?
Forum One works with government agencies to identify and implement technical solutions that move agency missions forward. Get in touch today.