We live in a world being transformed by artificial intelligence, known as AI. Every sector and industry is affected, including nonprofits.
Many definitions of artificial intelligence exist, but they all describe AI as a program that acts and thinks in ways that resemble human thought. AI is broadly split into two categories. One is artificial general intelligence, known as “Strong AI,” which, if it existed, would exhibit human-level intelligence. This is, for now, the realm of futurists and science fiction.
The other category is artificial narrow intelligence, known as “Weak AI,” which is what is available today. Weak AI “assistants” perform a limited set of tasks (perhaps only one), such as speech recognition and autonomous driving. One category of Weak AI is machine learning, which is the kind of AI nonprofits are most likely to encounter. It allows AI to learn from data without human intervention. Four components are called for: data, models/algorithms, tools/infrastructure, and computing power.
If you’re not paying for the product, you are the product. You must decide whether to pay for a service or allow others to use your nonprofit’s data.
Among nonprofits, early uses of AI include identifying viable donor prospects, recommending messages for prospective donors, and determining the optimal amount to ask a potential donor. In the near future, nonprofits will start using AI to fix duplicated and “dirty” data, and use predictive modeling to answer specific fundraising questions: for example, “Which of my repeat donors is most likely to lapse this year?” Increasingly, nonprofits will turn to AI to intelligently automate labor-intensive processes to save time and money while accelerating results.
As nonprofits explore this technology, here are 10 questions to ask to better understand and evaluate the services and products that AI companies provide:
- What data sets does your AI rely on? Data amount to the fuel that powers machine learning, the most common form of artificial intelligence that nonprofits will encounter. Data are used in two ways for machine learning: to train the AI how to make predictions, and to give the AI something to make predictions from. A nonprofit should ask which data sets are used for each.
- How does your AI deal with missing, duplicated, and otherwise messy data? Most nonprofit data are not usable for training AI, plagued as they are with duplicate data, incomplete records, and inaccurate data fields. A nonprofit should understand how the AI provider deals with such “dirty data.”
- What “features” in the data does your AI use for training, and where does it obtain them? A feature is a characteristic from a record, such as a name, address, income, or prior donation. AI teaches itself using the features in the data. If the nonprofit’s data doesn’t have many features, the AI won’t have much to learn from. A nonprofit should understand which features in the records the AI is using to train itself and where those data are coming from. For example, an AI tool may recommend the likeliest donors from a list of contacts. The nonprofit should ask which characteristics from the data set the AI tool uses to make its recommendations, as well as where that information came from.
- Which type of machine-learning tasks is the AI performing? The most common machine-learning tasks are supervised learning (classifying things into categories, predicting a value), unsupervised learning (clustering things into groups), and reinforcement learning (learning from mistakes). An AI service provider should be able to explain, in nontechnical language, which of those tasks its AI performs and how that relates to the service or product the nonprofit is receiving.
- Is the AI tailored to the nonprofit (that is, trained on the nonprofit’s data)? An AI could be tailored by training on the nonprofit’s data, or it could be trained on data from other sources, which results in a more generic approach.
All AI tools make predictions from the data provided. AI that is not trained on your data will not produce recommendations or predictions tailored to your organization, because it will have been trained on data from other organizations.
It’s similar to the difference between reading a report about giving trends at all nonprofits versus reading an annual report about giving to your nonprofit. “Tailored” AI, however, can be more difficult, expensive, and time-consuming to provide. Unfortunately, there’s no way to tell whether AI is tailored or generic by looking at what it provides — you’ll have to ask the company how its AI is trained.
- Does the AI continue to learn? Artificial intelligence can have a feedback process by which it continually learns on the basis of the results of its predictions. Ask whether the AI will improve over time using this feedback.
- Does the AI keep copies of the nonprofit data sets it uses? Your data are valuable. They can also be used for analytics, advertising, and similar monetization by third parties. You should understand what the AI provider does with your data, both while you’re a client and after you’re no longer a client. Otherwise you may be unwittingly giving it away. Remember: Your data belong to you.
- Does the AI provider sell information to third parties? Be wary of terms of service that allow unlimited use of the data you provide. Ask direct questions. In some cases, selling your data may be how a company makes money, which is likely if you are receiving the product or service free. There’s a saying: “If you’re not paying for the product, you are the product.” If that’s the case, you’ll have to decide whether it’s preferable to pay for a service with money or by allowing others to use your nonprofit’s data.
- How does the AI provider ensure that the AI results are fair and free of bias? Algorithmic bias — misleading or unfair results because of bias in the data used to train the AI — is a real and growing concern. Ask vendors how they account for and control for potential bias.
- How does the vendor enable AI to work with people? AI is intellectual automation and works best as a tool to improve, assist, and enable human judgment, initiative, and awareness. An AI/human team outperforms either AI or humans operating alone. Be sure to understand how the company enables its AI to work as part of a human/AI team. One way to evaluate this is by asking for case studies and success stories. For example, does the company promote studies and stories about how AI replaces people? Or how AI works with people?
As artificial intelligence matures, nonprofits will face a bewildering set of AI products and services and have many questions: Is the AI really AI? How is one product or service different from another? Which AI makes sense for my nonprofit? Answering the questions above will help decision makers learn about the range of AI services and products available. Comparing the answers provided by different vendors will help you make a more informed choice when implementing AI.
France Hoang is co-founder and chief strategy officer of boodleAI, which provides AI assistants to nonprofits to fix data, find answers, and enable peer-to-peer fundraising.
