How Nonprofits Are Actually Using AI Today

Artificial Intelligence (AI) is a hot topic in Silicon Valley these days. Here, we’d like to share some good news about projects in research and development about how AI can be used for social and, environmental good. Last year, the Asilomar AI conference laid out a list of AI Principles that can be used to guide development of AI for good, which Salesforce.org also supports. These also align to several Salesforce values, which include trust, growth, innovation, giving back, equality for all, and transparency.

Let’s talk about a few examples of how machine learning and AI are currently being used for good, and how they could be implemented in the near future.

AI for Good for Nonprofits

Dr. Vivienne Ming, neuroscientist and founder of Socos, shared her AI for Good insights at Dreamforce 2017 (watch video).

Seven areas she covered were:
1. Alzheimer’s Disease
Currently in the prototype stage, Dr. Ming discussed how AI is being used to detect the neural signature of memory failure and can give a person a “heads-up” about something they may be about to forget. Think “Hey, here’s the name of your child.” This type of “augmented intelligence” – not artificial intelligence – could help you to “be you” for as long as possible, and only turn on when you need it.

2. Autism
Using Google Glass, Dr. Ming created a system for autistic kids to read facial expressions. This also helped them learn theory of mind and empathy to understand why people do what they do. What’s outstanding about this example is that the too preserves what’s amazing about being different – and scaffolds skills for what autistic kids aren’t as good at.

3. Diabetes
Dr. Ming had a child with diabetes, so she built a model to predict blood glucose levels hours into the future. Eli Lilly is currently working on building an artificial pancreas. Pretty cool.

4. Finding Refugees
There are one million orphan refugees in refugee camps. That would be a lot of faces to look through – a million photographs, if you had to go through them in a paper book. Fortunately, with a deep neural network for faces created with Refugees United, you can use an iPad in a few minutes to help find a kid in a camp anywhere in the world.

5. Cochlear Implants
If you’re deaf, you may have a cochlear implant to help you hear. But cochlear implants are far from perfect, and having one is like “hearing the world through a garbage disposal” – there is a lot of noise. Dr. Ming worked on creating an artificial ear that learned how to hear on its own, using an algorithm to improve hearing, and adults could hear 50% more of what was said within 30 minutes!

6. Mental Health
We use our phones more often than we see our doctors. Accordingly, an app for phones used data that passively estimated emotion states could predict manic episodes in bipolar sufferers. Dr. Ming researched bipolar disorder and found that 3-4 weeks before the subjective onset of a manic episode, there was evidence of change in people’s activities, mainly in movement patterns. She helped create an system using machine learning based on data from 12 sensory packs on a smartphone to help predict changes in emotional states – enabling individuals and care providers to take intelligent action.

7. Education Outcomes
Predicting life outcomes of young children doesn’t always help – if you’re in a low income neighborhood, being reminded of that fact won’t help. So Dr. Ming created a game every night to play with kids based on of research on activities that correlate to positive life outcomes. This became a tool with SMS based tips that is easy to use with questions like “Do you have 20 minutes to spend with your kids?” and recommendations on how to spend that time to make a parent, grandparent, or foster parent’s time with kids more valuable and change life outcomes.

Although poorly deployed AI could increase bias (e.g. sending more high paying job listings to men than women on job websites), there are ways that AI can help with accelerating change for the better is on the social justice front. Raheem.ai, a chatbot that works on Facebook Messenger, allows the public to rate police interactions. Tested in Berkeley, California, Raheem asks users to answer basic questions about their experience with policing, and then funnels that information back to local precincts. Farley told a Dreamforce audience that during a three-month pilot testing Raheem earlier this year, twice as many reports were made as were collected during the entire previous year. The results promise greater transparency, more agile policing and a closer bond between local citizens and the law enforcement community.

AI doesn’t magically make the world better. It’s about an effective tool that can take a real human solution and share it.

What Nonprofits Can Do Today with AI for Good

1. Be goal-oriented
If you’re at a nonprofit, consider how artificial intelligence can boost your organization’s goals at a high level. “Look at your industry and your vertical when looking at how AI can help your org’s mission,” says Austin Buchan, CEO of College Forward, an Austin nonprofit, in his Dreamforce 2017 panel discussion with Cheryl Porro from Salesforce.org and others.

2. Think about people
To make AI work for good, check out this intro, or see this list of 9 positive steps you can take to keep AI honest. Make sure you think about the core problems – people, not algorithms – that you want to optimize for. In this way, you can make an intelligent impact.

3. Explore fundraising
Another approach for nonprofits is to start with fundraising. In this piece by Adam Martel at Gravyty, he summarizes how fundraisers are starting to use AI to reach donors more effectively.

And – if all this talk of AI for Good sounds intimidating, start with the basics – getting better nonprofit reporting, analytics, and all your data in one place on an impact platform.