If artificial intelligence (AI) is one of the biggest and most disruptive advances in technology in our lifetimes, understanding the ethical issues associated with it could be one of the greatest challenges. To be sure, navigating the ethics of artificial intelligence and figuring out how to develop ethical machine learning systems is something that every business and risk team needs to be thinking about. That’s because although the upside potential is huge, if you lack a sound, ethical approach to artificial intelligence, the implications could be devastating.
At best, you could be putting the future of your own business at serious risk. At worst, your decisions could negatively impact on society as a whole. Let’s take a closer look at why.
Artificial intelligence has the potential to change how businesses build and strengthen relationships with their customers. Using data and machine learning, businesses can turn every interaction into an opportunity to learn what people value and want. If that weren’t enough, at a macro level, artificial intelligence can optimize margins, directing spend and human resources to those customers where outreach and engagement would lead to the highest return.
On the other side of the coin, AI requires enterprises to use customer data in new ways. That means businesses that take advantage of AI have an automatic responsibility to their customers to use their data appropriately. Underscoring the urgency around this are some simple realities: People feel shocked when they learn that their sensitive information was leaked. Suspicious when they sense businesses want to manipulate their behavior. Powerless when an automated system denies them a product without any explanation for why. The bottom line is that trust isn’t a constant. It’s something that’s earned over years, and that can be lost in an instant.
This makes taking a risked-based approach to the ethical use of artificial intelligence essential. Unfortunately, we are not at the point where you can simply download a tool to do the job. Data ethics is still new and it requires critical thinking. Businesses have to build up the right muscles to asses machine learning uses cases, think about the risk that they can and cannot tolerate, and make difficult judgment calls. While it’s never going to be perfect, you can increase your chances of success if you approach these decisions with the right framework.
"Trust isn’t a constant. It’s something that’s earned over years, and that can be lost in an instant. "
That framework should help break things down into the various small decisions that teams need to make when building an AI system. It should also offer an agile approach to ethics and risk management that aligns with agile software development practices. Perhaps most importantly of all, it should be based on some clear principles to make sure everyone in the business shares the same intuitions around what matters and where cross-functional teams can focus their efforts.
Principles to guide you toward responsible, ethical AI
To operationalize ethics in artificial intelligence and machine learning systems, you have to ask the right questions to the right people at the right time. You also have to think critically and make tradeoffs. We suggest adopting the following principles to facilitate this:
- If you want the future to look different from the past so that you don’t replicate biases, you need to design that into your machine learning system.
- Be clear about what proxy metrics do and don’t optimize. You may learn they exacerbate bias or lead to downstream consequences that conflict with your values.
- When you deal with abstractions and groupings, you run the risk of treating humans unethically.
- Be aware of correlations that mask sensitive data behind benign proxies.
- Context is key for explainability and transparency, and business and risk teams should assess context and communicate required constraints to technology teams.
- Privacy is about appropriate data flows that conform to social norms and expectations.
- Accountability is a marathon, not a sprint. Develop a plan to catch and fix errors that arise after machine learning models are in production.
- Step outside the walls of your organization and ask communities and customers what matters to them.
Ethical AI and our future
AI offers immense potential to businesses and society. Our ability to process data at scale and use machine learning to learn from that data has shifted the balance between enterprises and consumers. Thanks to machine learning, relationships between consumers and businesses are becoming bidirectional: the actions consumers take provide a window into they are, what they want, and what they value. As in interpersonal relationships, businesses can listen to this feedback and use it to provide more relevant products, services, and experiences. The impact AI will have on society starts with the mindset we adopt to imagine its potential and the tasks we choose to apply it to.
Our framework will empower you to apply machine learning and innovate. We hope it will spark ideas and spur conversations between teams in your company or with new communities outside your company. AI is here to stay. We can use it for good. We simply have to ask the right questions and activate our ethical prerogative to express our values in the systems we build.
Industry voices about the ethics of artificial intelligence
This article originally appeared in Integrate.AI