How Google Assistant with AI can deliver a truly smart home

After watching the Google I/O keynote last week, I noticed that Google Assistant, which has been a staple of Google’s developer event for around eight years, was barely mentioned. Instead of Google Assistant being front and center, this year’s keynote was all about the company’s AI, or artificial intelligence, efforts. That suggests a very positive, seismic change when it comes to the smart home.

By the way, I’m not overselling how much of an AI focus the Google I/O event had this year. I lost track of how many times “AI” was said just in the first 20 minutes of the keynote. Just look at this user-generated compilation on YouTube of some, not even all, of the AI mentions at I/O this year:

At a macro level, I understand this messaging shift. OpenAI and other generative AI solutions have been the talk of the town for the last few months. And Google’s competitors debuted AI-powered software and solutions earlier this year. So it makes sense for Google to strike back and remind everyone that it too is a leader in this space.

But what does this mean for the future of the smart home? More specifically, what does it mean for Google’s vision of the smart home, which up to now has centered around Google Assistant? Sure, you can use phone apps to control your smart home, but smart displays and speakers in combination with Google Assistant is a compelling method as well.

However, Google Assistant is really just an interface to your Google Home devices; it never became the brains of the smart home. At I/O, instead of sharing new Google Assistant updates, we got a nice tablet with a barely useful speaker dock and a user interface refresh of the Google Home experience. That’s about it. Google didn’t mention its smart speakers, where the Google Assistant “lives,” at all. And aside from the Pixel Tablet in Hub Mode on its dock, it didn’t show the new Google Home interface on anything other than Android phones.

As a result, I can’t help but wonder if the Google Assistant we know today is effectively going away. Sure, I anticipate the name Google Assistant will still be around for years to come. But it won’t be the same. With a focus on enabling AI in all of its products, the Google Assistant smart home experience will be driven by Bard, PaLM 2, and other Google AI efforts. And Google isn’t the only one with this strategy: Amazon recently announced it plans to bring a large language model to its Alexa assistant.

When I first started thinking about this, I was dismayed. Most of the latest generative AI products use such massive data models that they have to be run in the cloud. So I figured a true AI-powered Google Assistant replacement, or enhancement, would reverse recent “on-device” smart home efforts. And that would be disappointing since moving away from local processing pushes our smart home data beyond the confines of our homes. Digging into the details of Google I/O, however, that doesn’t have to be the case.

Google’s PaLM 2 next-generation language model comes in four distinct sizes, ranging from smallest to largest. They are named Gecko, Otter, Bison, and Unicorn, with Gecko being “so lightweight that it can work on mobile devices” and “fast enough for great interactive applications on-device, even when offline,” according to the company.

However, PaLM 2 model training is focused on mathematics, coding, formal logic, and commonsense reasoning, which means that PaLM 2 is better suited to scenarios outside of the smart home.

But PaLM 2 model training is focused on mathematics, coding, formal logic, and commonsense reasoning, which means that PaLM 2 is better suited to scenarios outside of the smart home.

It’s really Gemini, Google’s next AI model, that could smarten up a local version of the Google Assistant. Google says Gemini will be “multimodal, highly efficient at tool and API integrations, and built to enable future innovations, like memory and planning.” And like PaLM 2, Gemini will have available models in different sizes, enabling it on various devices with a range of computing power.

Now we’re talking. I think it’s safe to assume that Gemini will have some offline capabilities, given similar multiple-sized models like that of its predecessor. And when it comes to a smart digital assistant for the home, you need API integrations and — perhaps the most important feature missing from today’s Google Assistant — memory. To have a truly “smart” smart home, the brains behind it can benefit from the learning patterns of those inside the home. It sounds like Gemini may be capable of that.

Although I like to think I lead an exciting life, it’s generally one of a few daily patterns in my home. I wake up, immediately go downstairs, make coffee, and sit in my recliner for an hour or two. Currently my family and I can manually set up automations and routines to power on the coffee pot. We can do the same with motion sensors, so if I wake up before sunrise, the downstairs lights will be illuminated on the way to that cup of java. But we create those rules. And sometimes they’re hard and fast rules, such as setting up something to occur at a specific time.

A generative AI model that can run offline could reduce or even eliminate the need to create manual rules in the smart home. The Google Assistant driven by Gemini could see patterns, for example. Then it could execute on those patterns automatically. Or it might prompt me by saying: “I notice you often wake up, come downstairs, make coffee, and read in the mornings. Shall I automate the coffee pot and lights for you?” This wouldn’t be that different than Google’s Bard AI, which can debug or create working application code from a prompt. In fact, it would probably be simpler than that.

Here’s another useful example. Today I asked the Google Assistant on my smart display, “What lights work with the Google Home?” The answer wasn’t helpful. Google Assistant told me that its lights have different functions and that I should check the manual. Clearly, it was talking about the lights of the device instead of telling me which bulbs are compatible with the Google Home app.

The answer to this question can be found on the internet. But it’s scattered across numerous places: device reviews, manufacturing product pages, and an outdated Google Home support page that doesn’t even mention Matter. Sure I can search the internet if I’m in the market for new connected bulbs and I use Google Home. But a well-trained, capable AI should be able to provide me with a better answer in less time. So aside from smart home controls and predictive automations, a mature AI-powered Google Assistant could be far more useful.

Don’t expect a Gemini-powered smart home experience anytime soon, however. Google says it’s still training the massive model and won’t release access to Gemini until it’s tested and deemed safe. Yes, responsible AI was also a big part of Google I/O this year.

When Gemini does arrive, though, I anticipate a return of Google Assistant to the I/O spotlight. Maybe it keeps its name and maybe it doesn’t. That doesn’t matter too much to me. What does matter is a smarter digital assistant in my smart home that’s vastly more helpful than today’s relatively stagnant assistant, which hasn’t delivered on the promise of a truly smart home.