Artificial intelligence has attracted broad attention lately, but most of the conversations surrounding this topic are irrelevant to the actual, impactful progress being made in so many areas. In the real world, artificial intelligence isn’t about recreating and supplanting generic human intelligence — it’s about automating specific tasks and solving specific problems that require techniques beyond simple pre-programmed, if-then-else logic.

Mobile check deposits and adaptive cruise control are just two examples of problems that AI has solved and that people now take for granted. In fact, as AI applications like these become commonplace and ingrained into everyday life, people no longer consider them to be “artificial intelligence,” reserving this term for the yet-to-come magical future. But the reality is that we already have at hand many powerful tools that save us time and money.

When taking this more pragmatic view of the near-term potential of AI and pondering the most compelling applications, many look to the home, where countless daily tasks are ripe for automation. While voice and smart home assistants are already lightening the burden of home management, they are just the tip of the iceberg.

One day, our homes will be able to proactively help us manage the many mundane responsibilities that come with home ownership. Imagine your home detecting that your HVAC system is nearing its end of life and arranging for a technician to drop by; knowing that you are away, it unlocks your front door when the technician arrives, locks up afterward, and then shares a summary report once you return from work, complete with relevant video clips.

But while we eagerly wait for this streamlined future to become reality, it is important to recognize that we will still be very much involved. In fact, for artificial intelligence to truly add value, especially in the home, it needs to be designed to keep people playing a significant role, one greater than they may be expecting. Here are three reasons why.

Understanding both state and intent

A truly smart home understands human intent, but first it must be able to understand the state of the home. While data from sensors and devices combined with artificial intelligence algorithms make it possible to develop an accurate model of what’s happening in your home (e.g., everyone is asleep, or the adults are at work but the kids just got home from school), occasional explicit clarifications in ambiguous situations can help make this learning process faster and more accurate.

Actually knowing what you want, however, is fundamentally harder for your home. As with even the best human personal assistant, your home will need to receive direction from you at first to ensure it can make the right suggestions. Eventually, as your home applies learnings from this ongoing feedback loop and develops a deep, actionable understanding of what you want to happen in different circumstances, it can transition from just making suggestions to automatically taking actions and providing confirmation.

Keeping people in the loop

While we often picture a smart home assistant toiling away discreetly behind the scenes, our research at Vivint Smart Home shows a different result: People actually want to stay in the loop. We initially thought that our users would have a strong preference for “silent automation” — i.e., having their home automatically manage tasks without bothering them — but this didn’t hold up during real-world testing.

Instead, it turns out that homeowners want to be notified of the actions their home is taking, even when they have high confidence that those actions are what they would want. For example, they were comfortable with the home automatically locking up after everyone left, but only if the system told them what it was doing and why. People need to maintain a sense of awareness of what’s happening for their emotional comfort, quite separate from any concern for error.

This puts smart homes in stark contrast to mobile apps, where there is strong user sensitivity to being bombarded with notifications. But unlike with a typical app, there is a higher threshold for the number of acceptable alerts from your home because each is personal, relevant, and valuable. It’s akin to the difference between being notified of incoming texts and being annoyed with pleas to return to playing that game some more.

Letting AI take the wheel

Getting to this ideal end state of a “self-driving” home that confidently lets you know the actions it’s taking and why will not be easy — for the humans, not just for the home. It requires people to cede control, but most people won’t understand at first what the AI can do and why or when it will do it, and as a result they won’t trust it. Getting there requires a gradual process of building that understanding and trust, and that means people will start out engaged and in charge, only slowly letting AI take the wheel over time.

Moreover, people’s lives are unpredictable, so no AI is ever going to be able to predict every special case or every random whim that breaks your ordinary routine. We need to be able to take back control in these moments, and the AI needs to make it seamless and easy. And even in the moments when control is not required, people still need the continual assurance of knowing that the option is available — it’s an emotional need in addition to a functional one.

We’ve been waiting a long time for our homes to learn what we want done and then do it for us. But as we inch closer to this ideal, it’s important to remember there will be a significant learning curve — not just within the machine learning algorithms seeking to understand home state and human intent, but also among people as we realize our vital role in shaping a true smart home experience.

Jeremy Warren is the CTO of Vivint Smart Home, a leading provider of smart home services in North America. He oversees the creation of new products and services at the Vivint Innovation Center.