Decision automation is just a continuation of data-driven decisions

I've been meaning to jot down some thoughts on AI and while doing research I came across this banner ad that I think sums up the zeitgeist:

There is a feeling that vastly powerful tools are coming and that they will change everything (cue the requisite FOMO). However, at least for the foreseeable future, we are the masters of these machines and we will dictate how they "change every business"... so how?

I think the prospect of self-driving cars has skewed our perception of how AI will transform business. The data-processing challenges of driving a car are immense, but the rules dictating the decisions made with that data are pretty simple. For most of our jobs, the situation is reversed. We have a relatively small amount of data to help us make very complex decisions. We are, through a variety of technologies, getting more and better data and as that happens we will naturally get more data driven in our decision making. This will lead to incremental changes in the way we make decisions that should be far less frightening than "DON'T BE LEFT BEHIND." would lead us to believe*.

The framework below lays out what I see as the broad types of decision making from completely manual to completely automated (and less data to more).

Running quickly through the first four, more traditional approaches:

• Gut feel/experience: Do what feels right or you've done before
• Brute force analysis: Put the data in a spreadsheet and play with it until something insightful falls out
• Visual analysis: Begin to abstract the data by plotting it an hope that patterns emerge that weren't clear in a table
• Statistical analysis: Use more robust techniques for abstracting and finding patterns

Now we're to the bolded section where "AI" comes into play.

• ML with human decision: In most cases this is where even the most advanced companies are with AI adoption. Advanced Machine Learning techniques parse vast quantities of data to recognize patterns and present them to the decision maker. This step uses better tools and more data than statistical analysis, but it should be clear it is an incremental change.
• ML with human anomaly handling: In this step we let the algorithm take action in most cases but kick things out to a human when the certainty it's recognized the pattern is below some threshold. Now we're getting to machines actually making decisions for us. Skynet is coming, right!?! Not so fast. In the step above, you're already trusting the algorithm to discern the pattern properly and not discard critical information. You've delegated a great deal of responsibility already, this next step is akin to letting a subordinate work without oversight (and trusting them to escalate things as needed).
• Artificial Decision Making: This is where things get interesting, right? No more humans in the chain! True, but as long as there is a human somewhere in the organization, there will be detection and oversight of massive anomalies. So, you're delegating this decision as far down the chain as possible, but there is a meta-decision you keep (even if it's just to continue letting the AI make the call).

The organizational and social impacts of widespread AI are huge (more on that another time) and there is certainly the risk of being left behind if our skills fall behind. But, in another sense, this is just an extension of trying to get decisions made where they are made most efficiently. Using a decision rights tool like Bain's RAPID (or the inferior but more common RACI), we could lay out the current and future state of the critical decisions for our organization and see where AI is (or should be) moving from an "Input" to a "Recommend" to a "Decide" position. Those decisions that benefit from speed, scale, low cost, lots of disparate data, or other factors are natural choices. Those that are interpersonal, impact culture, require a high level of explainability, or perhaps are just deal-breakers for the way we envision our firm's mission should be kept in human hands.

AI is a crazy powerful and at times inscrutable tool, but ultimately it is just a tool for helping make data-driven decisions. The organizations that will win with AI are the ones who are already thoughtful about how and where they make their decisions.

*At least for a while. I think there is a tipping point I'll discuss sometime in the future.