“Thoughtfully Biased” AI – Our Best Option
by Sarah Di Troia
(Reading time: Less than 4 minutes)
Your Netflix account and my Netflix account don’t look the same.
We may both have access to all the same programming, but based on our demographics and past behavior, artificial intelligence (AI) is dictating how our respective home screens look and the programs that are recommended.
The same principle holds true for the ads we see on our phones, the deals we receive from credit card companies, the job interviews we are offered (or not offered), and on and on. AI is differentiating between us and acting accordingly.
Of course, there are benefits to this kind of customization, quickly finding a movie on Netflix among them. But there are also dangers:
#1. AI is Biased
AI can magnify racism. And sexism. And ageism. And every other “isn” across the board.
It works based on historical data. To the extent our history includes marginalization, this is baked into the tools themselves.
These biases may not be part of the algorithm, but the data itself — your zip code, credit score, education level, etc. — all reflect the best and worst of our society. It’s so pervasive that we have to be deeply thoughtful about how the data is gathered and used.
#2. Nonprofits Always Get Hand-Me-Downs from the Tech Sector
Developing new technology is expensive. It starts by being tailor-made for companies with the biggest budgets; it then becomes less and less expensive and expertise dependent. Remember when websites required a team of coders and now your marketing intern can update your Squarespace-hosted site without coding skills?
So, as AI improves and its cost inevitably drops, the companies that develop these tools will make them available to us; they will become integral to the way we work. That is happening in the realm of AI with “no code” tools already available.
Again, some good news — this will expand the speed and scale of what an individual can do, largely because AI will create efficiencies with repeatable tasks, make recommendations, and help with decision-making. But here as well, all of it will be inherently biased.
And, since the social sector gets these tools further along in the process, by the time we do, our opportunity to exert influence over the underlying design will have passed — just as it did in the case of Salesforce and other automations that first appeared in the for-profit world and that were really hard for us to use when first offered to our sector.
In the case of AI, this is especially dangerous. In the absence of our insights about bias, we run the risk of products and implementation support that are not optimized for equity.
Thoughtfully Biased
Eliminating bias entirely is not possible, especially for a technology that is fundamentally based on what has come before. Our opportunity, then, is to remove what bias we can and acknowledge what remains — to make AI thoughtfully biased.
Some suggestions…
Get Involved Early
Now, while the experiments are happening and before the paint is dry, we need our sector to interrogate the use of AI in ways that drive equity and inclusion.
We need our own pilot programs and learned insights if we hope to have examples we can point to of how AI can be thoughtfully biased in ways that optimize equity and impact. That’s the only way to raise equity concerns that always seem to arrive after the fact, when the damage is already done.
Consider healthcare. We have only recently begun to understand (thanks to a great deal of advocacy and community-based pressure) that it’s the social determinants — food, neighborhood, housing access, etc. — that establish the bulk of an individual’s wellbeing. But our built healthcare system is hard to change and remains inadequate since most of the investment and emphasis has historically focused on care and treatment, rather than cause and prevention.
Find the Efficiencies
AI can support humans doing work “at the top of their license” — work that takes full advantage of their level of training and education. For example, instead of staff, chatbots in multiple languages could answer routine questions from any population, delivering them asynchronously at whatever time and place is best. This would enable human to do more of the work only they can do for clients.
Further, AI’s superior efficiency can allow for working more profitably, lessening the constant need for fundraising as a means of solving societal problems with charitable offerings. Not only will this lessen the burnout that is so prevalent in our industry, but it may also permit different organizational structures (e.g., a for-profit with a nonprofit subsidiary), enhancing sustainability and unlocking further investment.
Drive Greater Impact
AI can run tests against naturally occurring pairs inside your data to identify and understand patterns. Our Netflix accounts don’t look the same because Netflix has used its data to identify over 2000 different taste profiles — the basis they use for creating customization and value for customers.
AI’s ability to run continually organic matched comparison groups will allow for both recognizing efficiencies (maybe certain individuals only need three parts of a program model rather than the standard ten) as well as unlocking greater impact through the detection of sub-patterns in populations that can be further customized and prioritized.
Overall, this leads to both cost savings and enhanced impact for each individual.
AI Is Already Having an Impact
We can’t afford to be at the tail end of this game-changing technology. Pretending it’s not going to affect us or simply waiting for it to be handed to us once all the big decisions are made, are both losing strategies.
When the dust settles, let’s make sure that whatever biases are incorporated are consciously and thoughtfully implemented!