Ode to Ethical AI: Be More Awesome

I’m new to the field of data science, but I’ve done my best to immerse myself in relevant material from podcasts to whitepapers. As I built my first prediction algorithm (I know… super impressive), I began thinking about the implications of such ability. What if the data I put in was bad? What if I unconsciously include parameters that are biased? It wasn’t as much of a problem since I knew the data I was feeding it, the weights of the variables, and could see exactly how the prediction was made.

But what if I couldn’t? What if it were a black box?

I noticed that internally, ethics is often received with scorn — it’s clunky and doesn’t allow one to move fast, nonetheless, break things. I understand the hesitancy. We are wired for results — no one wants to be held back by bureaucracy. But by pushing away any internal progress towards explainability, we risk welcoming greater external regulation.

From grassroots activists to NYU’s AI Now, there are calls from the public for socially ethical AI. Falling short of acceptance by the public risks an outright ban on the use of ML/AI in both the commercial and government sector.

And this is an issue (obviously). We need this technology now more than ever; to create a better world amidst the current crises. We know the potential for progress with data science and related fields. If we want to realize this future we all hold internally, we will have to take steps now to address the issues while the field is still pre-pubescent. Let’s look at a few places we can start.

Diversity

Yes, I know. DEI (diversity, equitability, and inclusion) is the trending topic at. But that’s because it’s relevant. And sometimes addressing it head-on stings because we see where we’re falling short.

The graph above is from a 2018 post on KDnuggets. And I’ll be honest, as a heterosexual white male from the Midwest, this graph hurts my heart a little. It should. It means there is more to be done.

Aiko Bethea, an executive leadership coach and equity consultant, cuts through the B.S. when she says, “Stop demanding the business case for investing in diversity, equity, and inclusion work. Just do it.” (I highly recommend reading her Medium post here.)

And it is important to “just do it” not only the financial success of companies but specifically for ethical issues that exist in ML/AI. Let’s face it, there is bias everywhere. In our data, in our code, and in our oversight. And when you don’t have a team representing various backgrounds and divergent ways of thinking, it will be difficult to see all of the issues. It reminds me of the David Foster Wallace parable:

There are these two young fish swimming along, and they happen to meet an older fish swimming the other way, who nods at them and says, “Morning, boys. How’s the water?” And the two young fish swim on for a bit, and then eventually one of them looks over at the other and goes, “What the hell is water?”

Broadening perspectives in the field of data science will improve the scope and angles of questions regarding our work. With increased diversity, we will not only be better equipped to address ethics, but also have higher quality output.

Collaboration

I first heard about Audrey Tang in a story about how Taiwan hacked COVID-19 with open-source monitoring programs. They’ve created a community chat platform that rewards cooperation and communication and disarms trolls with humor rather than feed them with fire. Holy crap. They literally open-sourced their government. What an incredibly powerful model that the data science community could use to bring the public representatives to the table in a non-destructive manner. Collaborating with the public is important to get buy-in from all stakeholders. It also improves trust in algorithms as people feel they’ve had a say in their development. Increased trust can only improve the efficacy of ML/AI tools.

An additional collaboration opportunity is with fellow AI companies. The open-source ML/AI movement has been quite successful so far with companies like OpenAI leading the way. Though we’ve done well to share code, we could do better at establishing consistent ethics guidelines across companies. So far, many of the ethic statements have been “paper tigers” as they lack actionability and enforcement. This goes back to collaborating with certification processes to ensure accountability. Additionally, if companies possess shared ML/AI ethics guidelines, they’re incentivized to holding each other in check. Cross-company ethics guidelines may not be sexy, but it is important for the future societal adoption of ML/AI.

Explainability

With the rise of neural networks, algorithm black boxes are the latest fad. Since these models are generally self-trained in the wild, it’s not always clear how they make decisions. And if you are unsure of your model’s thought-process, debugging will be a pain when and if it fails. Beyond hampering performance, lack of explainability raises a host of ethical issues.

When you are unaware of how decisions are being made, you cannot be certain that the algorithm isn’t using making biased choices. And, unfortunately, the bias does exist. Bias has been well documented recently as watchdogs ring alarm bells. The issue is compounded with bias in data (e.g. more photos of furry cats than hairless ones). There are things we can do like, funnily enough, use AI brethren to interpret our neural networks. And honestly, some of the explainability concepts fly right over my head. But I know there are smart people working on it — if you have any good articles, please comment below!

Intelligence That Lasts

The upside to ML/AI is huge and we could do better about sharing them. But to regal in the good, we need to cross the ethics river first. We are only hurting ourselves by not getting ahead of the discussion especially in relevant biases such as race and gender. If we don’t respond quickly to the current outcries, we will limit our ability to realize the great gains that are possible.

On another note, addressing ethics is critical for the morality of our future society. AI is a reflection of its data. The data is a reflection of the society. Much like history, it is controlled by the winners. If we aren’t careful, the winners may continue to win especially if we combine their power with artificial intelligence. But the benefit of history is the ability to not make the same mistakes made by our ancestors. This is wisdom. By learning from history, we can free ourselves from our past and create the future we desire.