Introduction to Artificial Intelligence (AI)

Artificial intelligence is the umbrella term for a variety of fields focused on teaching computers to learn. There are different ways of doing this, including machine learning (ML) and deep learning. For the purposes of this guide, we will be focused on machine learning, as that is the most developed and applicable field for most organizations.

Machine learning is teaching a computer how to learn through observation. The goal is that over time the computer gets better and better at whatever task it has been given through repetition and correction. It’s what enables everything from self-driving cars to the images you see on Netflix to the recommendations on Amazon.

To create a good machine learning system you need a few things; good, structured data; a large enough quantity and variety of data to provide information to learn from; a repeated task; and patience.

There are three main categories of machine learning. The first is called supervised learning. Supervised learning is defined by its use of labeled datasets to train algorithms to classify data or predict outcomes accurately. A labeled dataset is when you take data, often existing, historical data, that has both a bunch of inputs and the target output you are looking to predict.

The second category is unsupervised learning. This is utilized when you don’t have a labeled dataset but rather let a computer cluster and organize that dataset on its own. In this category,  a computer identifies hidden patterns and connections in the data without a human intervening with labeled data.

The last category is a hybrid of the first two called semi-supervised learning. Semi-supervised learning offers a happy medium between supervised and unsupervised learning. While training your data, it uses a smaller labeled dataset to identify which features might be helpful in a larger, unlabeled dataset.

Just as organizations can use data for a wide variety of purposes, machine learning can be applied to just about anything. From helping nonprofits better target fundraising outreach to identifying diseased crops in Africa, machine learning can help drive real impact for organizations.

For example, Google is working with Oceana and SkyTruth to identify instances of illegal fishing in a tool called Global Fishing Watch. It gives information freely to organizations, governments, advocacy groups, and more to have a view of global, commercial fishing activity.

Crisis Text Line has long used machine learning to support its human counselors. They use machine learning to triage and prioritize incoming texts and match individuals to the counselors best suited to support them. Crisis Text Line also highlights the risks of using machine learning. Politico reported they had a for-profit spin-off using their data to inform call centers, which highlights the complexity of where and how sensitive data gathered for AI/ML can or should be used..

Quill, an organization that helps students learn grammar, uses machine learning to automate their learning platform. Largely using free open-source tools and using data from Wikipedia, they developed a sentence fragmentation algorithm to help students improve their grammar.

By now you have seen that machine learning can be used in so many different ways across your organization. So where to begin? First, check out our Data Strategy guide. It is important to anchor your use of ML in your larger data strategy. Identifying the goal of an ML system is critical to being effective.

Once you have identified what you are trying to accomplish, you need to ensure that you have appropriate data to answer that question with machine learning. This often means having enough data to help the machine learn. This isn’t purely about the quantity of data but about ensuring you have enough variance of data as well. Data that represents the real-world situation.

The good news is that ML systems are increasingly more straightforward to stand up. Tools like Tensorflow and SciKit Learn can make it easier to implement ML algorithms and develop custom models. There are also numerous flow charts to help identify what types of algorithms might be best for your task.

For many organizations beginning to use machine learning, it is helpful to augment their internal capacity with outside consultants and volunteers. Groups like DrivenData and DataKind can support your organization’s machine learning goals.