The Key Ingredient To AI Success: Our Values

Artificial Intelligence (AI) can enhance humanity; and it can also destroy it. It all comes down to one thing: values.

Values Before Ethics

People are starting to talk about the term “Ethical AI,” correctly identifying that we need to have some controls over how we use and deploy AI. But those discussions often miss a key point, ethics are subjective. They exist to reinforce a set of values, and those values can vary widely across people.

Of course, everyone tends to believe that there is only one ethical standard, as they confine their thinking to their own values. To demonstrate, I’ll jump into a very controversial topic: abortion.

A pro-lifer may see a woman’s decision to have an abortion as ethically wrong while a pro-choicer (I’m one of those) may see it as ethically right. Both of those are ethically correct; as they are consistent with their values. Their differences aren’t in ethics, but in values. One group values a strict interpretation of religious transcripts, while the other values an individual’s right to make decisions about their own body.??

As you can see, aligning values can be very difficult. But we can never agree on ethics until we expose and debate our values. Now let’s jump out of the abortion discussion (probably not soon enough), and get back into my discussion on AI.

Introducing Values-Driven AI

If we want AI to operate consistently with our values, then we need to adopt explicit practices to make it happen. That’s why I’m introducing a new approach called “Values-Driven AI” (VDAI), defined as:

Applying AI in an ethical manner based on a clearly defined set of values

There are four steps to VDAI:

1. Values Clarifying: Be explicit about what‘s most important for your organization.
2. Values Nurturing: Teach the analytical models how to operate in accordance with those values.
3. Values Sharing. Share your values with all of the people affected by the AI.
4. Values Maintaining: Continuously validate and update your values.

Step #1: Values Clarifying

For most of our lives, we operate using an implicit set of values that we amass over time. Most of the people we work with have grown up with similar backgrounds, so they naturally share a lot of the same values. But AI does not “grow up” in the same way, as it does not have access to the subtle clues about right and wrong that human beings experience during their early lives. Lacking this grounding, AI will look for clues about right and wrong from how people behave. But, as we all know, human beings often make decisions and act in ways that are inconsistent with their values.?

In order to drive AI in what we would consider an ethical direction, we need to be explicit about our values. Here are some steps for getting clarity around your values.

• Develop 10 to 15 statements that capture your organization’s values
• Prioritize your values statements
• Build alignment across your organization around the listing of values

Step #2: Values Nurturing

We have to help our AI understand and behave consistently with our values. This is not an one-time effort, as it is more like the ongoing process of bringing up a child.

AI is like an infant; eager to learn what the world teaches it. And just like with a child, we can’t fully control what AI learns so we need to instill it with a strong sense of values to guide its learning and actions.?

As a simple example, let’s say that we offload our executive recruiting to AI, asking it to find the best candidates for a new Senior VP role. If it is only looking for the candidates who are most likely to succeed, then it might look for a white, middle-aged man since that looks like the most dominant profile of the company’s senior executives. That’s one of the flaws of generic AI models, they tend to reinforce ingrained biases.?

Think about a world where all of our worse tendencies are automated and replicated, and any bad actions can be amplified at a rapid pace. Pretty nasty! We can already see that happening on social media sites where AI is being used to provide people with content that will keep them engaged, often without any consideration for the quality or accuracy of that information.

In order to nurture our values within our AI, we need to:

• Explain your values to AI by sharing examples of what behaviors are consistent and inconsistent with those values
• Tune the models to correctly apply our values by giving AI feedback on how its actions are consistent or inconsistent with our values

Step #3: Values Sharing

If I had to pick one recommendation that I had to use for any situation, then it would be to “be transparent.” So many of the world’s problems either stem from a lack of transparency or would be dramatically improved with greater transparency. AI is no different, and may even be one of the most important areas for transparency. Why? Because AI is a black hole. Most people have no idea about how it operates.

While it is not feasible to explain to most people how AI works, it is important that they understand the parameters that are driving its actions; its values. This provides people with the information they need to decide if and how to interact with an AI-driven system, and will make them feel more comfortable with any of their interactions.?

People who either use an application driven by your AI or are affected by decisions being made by your AI should have easy access to information about the values that are being used by the AI to shape its actions.?

To ensure transparency, make it easy for all stakeholders to discover:

• Values that are driving the AI
• Data sources that AI is using to learn from
• Range of actions that AI are automating
• Controls you have in place to maintain the AI’s ethics
• Process for reporting potential values mismatches

Step #4: Values Maintaining

While an individual’s values are relatively stable over time, an AI’s guiding values need to be constantly re-evaluated. Why? First of all, since this a brand new practice there will undoubtedly be ways to improve how you describe the organization’s values. Also, organizations shift over time, so it’s important to ensure that the values being used actually reflect the current state of the organization.

Here are some ways to ensure that your values remain aligned with the organization’s actual values, it’s critical to have an annual review of your AI efforts where you:

• Review your organizational values
• Update your organizational values and re-endorse support for them
• Identify ways to better teach your AI models to demonstrate your values

It’s Time To Establish AI Governance

I haven’t introduced VDAI as an exact blueprint, but instead as a model to spur critical discussions. AI is an important enough emerging capability that organizations need to proactively plan for it before it becomes widely used in uncontrolled ways.

Every large organization needs to establish some type of AI governance model with some resources and an executive overseeing the effort. This team needs to explore and establish guidelines around many different areas, such as:

• Values-Driven AI
• High-value opportunities for AI deployments
• Privacy
• New skills and roles required for deploying AI
• Future of employee roles, after AI is deployed
• And of course, experience management!

The bottom line: AI will grow up to be whatever we teach it to be, so we need to be good guardians.