Feb 13, 2024 3:00 AM PT
Android Intelligence Analysis

The real problem with Google's new Gemini Android assistant

Hey, Google: We've gotta talk.

JR Raphael

I'll be honest: I was really hoping for an "aha!" moment over these past few days.

Ever since Google made it clear that it was working to bring its next-gen generative AI chatbot into Android as a replacement for Google Assistant, I've been skeptical. And ever since I installed the new standalone Gemini Android app and allowed it to take over my phone's assistant function, that initial skepticism has only grown into a grumbling sense of us being ushered in the wrong direction.

Now, let's be clear: None of this is to say that Google Assistant in its current form is perfect — far from it! We've been talking for years now about gaps in Assistant's advancements and ways the service could be made better. And lately in particular, it's felt like Google's been neglecting Assistant as it turns its focus to its shiny new buzzword-boosting toy.

Still, despite its shortcomings, Assistant knew what it was all about. It was a service with a purpose, and Google spent years building it up into a platform and making sure it was everywhere.

With the newer AI chatbot — known as Bard until a rebranding late last week — I've never quite felt like Google knew what it was creating and why (outside of the obvious business incentives). And having now spent several days with Gemini playing the role of virtual assistant on my personal Pixel 8 Pro phone, I'm scratching my head even more about how Google could possibly think this is a good idea.

[Get level-headed knowledge in your inbox with my Android Intelligence newsletter. Tips, insights, and other tasty treats await!]

So, yeah: I haven't had that "aha!" moment I was hoping for, to put it mildly. But I have had a hefty realization in thinking about what, specifically, the problem is with Gemini as a replacement for Google Assistant and why it seems like such a puzzling fit.

And it's really quite simple.

Android, Gemini, and the Google Assistant contrast

First, a fast pinch of pertinent perspective: When Google Assistant first launched in 2016, it was actually an evolution of a series of long-standing Android services.

The similarity was so striking, in fact, that I initially described it as the "elephant in the room with Google Assistant" — because Assistant truly was "an expanded and rebranded version of what [we'd] previously called Voice Search, Google Now, and Now On Tap."

But all of those services served the same underlying goal, and that was to help you perform simple tasks and get swift answers without having to use your hands or futz around with any on-screen menus. It was "a Google for your world," as Google poetically put it as the time of Assistant's christening.

Despite its imperfections and growing pains, Assistant has served that purpose relatively well for most of us over most of the time since then, at least up until the start of the AI chatbot infatuation. You need to change some random phone setting while on the move? Ask Assistant. Want to play a particular podcast or type of music? Adjust your allegedly smart office lights or thermostat? Set a reminder, send a message, create a calendar appointment, check in on your agenda? Easy peasy, all around — a quick spoken command to your silky-voiced Android companion, and the deed would be done.

Assistant was made to accomplish tasks, in other words — to help you achieve simple feats with your phone, control your connected apps and gadgets, and occasionally summon short, simple answers and information.

Gemini, in contrast, is at its core a generative AI chatbot. Like ChatGPT and other similar systems, its key capability is the ability to create — which, to be fair, is a generous term. But its ability to, well, mash up text and images and summarize large amounts of existing info is what sets it apart from earlier efforts and makes it useful (in theory, at least, for some people).

And hey, that's fine! That's a purpose, and it clearly has its place. But squaring that purpose with the purpose of our Android assistant is where I struggle to see the connection. And having spent several days living with Gemini in place of Google Assistant now, it's more apparent than ever to me:

The real problem with Gemini as the Android assistant is that Google's forgotten why a phone assistant actually matters — and what we, as actual users in the real world, need from such a service.

Plain and simple, using Gemini in place of Google Assistant feels like having a square peg awkwardly forced into a round hole. It feels more like an awkward adaptation of an AI chatbot than a phone assistant — something that's half-baked at present and not at all intended or appropriate for this context.

And the more time you spend using Gemini, the more apparent that disconnect becomes.

Gemini's awkward assistant additions

First things first: The relative strengths that Gemini brings into the picture just aren’t things that need to be in an on-demand assistant or that make sense in that environment.

All that generative stuff is fine, for the right purpose — and with all the usual asterisks around quality, accuracy, originality, and so on. But do you really need that sort of function in the context of an on-demand assistant on your phone? Aren’t you more likely to open an app to handle a task like that — or to seek out a function within an existing app like an email client, a word processor, or even an Android keyboard (all of which are places where similar sorts of systems have already been added)?

JR

Some of the suggested actions from Google's new Gemini Android assistant interface.

Those features are weirdly out of place and unnecessary in this context, particularly when the very same functions are also being added into every other imaginable app and service — including most Google services — anyway.

And that's to say nothing of the even more important part of the picture — the standard assistant intelligence we've come to expect from a service in this position.

The Google Assistant to Gemini downgrade

I won't beat around the bush: The core functions we rely on from an Android assistant take a significant step backwards when you move from Google Assistant to Gemini.

As we established a moment ago, a universal smartphone assistant is most suitable for answering quick questions, performing basic on-device tasks, interacting with connected gadgets — the things Google Assistant has traditionally done well.

Gemini generally does all of those things slower and less reliably — and in some cases doesn’t even do 'em at all. Want to set a reminder? Interact with your calendar? Remember where you parked? Play some music? Gemini can't do any of that stuff yet. When you encounter a task it can handle, like asking the system to find certain emails, it's clunky and slow with its processing — to the point where most people would probably swipe away and go do it themselves before it finishes.

JR

Trying to use Gemini as an Android assistant is an exercise in frustration.

It requires awkward extra taps to submit input, too, even when you're talking. And it rarely gives you spoken responses, which kind of defeats the very purpose of interacting with a hands-free, on-the-go Android helper.

Beyond that, Google's separate new Circle to Search system (and even just the plain Google Lens technology that powers it) is significantly better at visual search. Gemini feels like a substantial downgrade in almost every traditional assistant regard.

And all of that's to say nothing of the actual issue of accuracy and being able to rely on the information Gemini gives you as an on-demand assistant. As we're all aware by now, these large-language-model chatbots have a fun little habit of "hallucinating" — or, to skip the euphemism, flat-out making up info and presenting it confidently as fact.

Here, for instance, is a screenshot one of our Intelligence Insiders sent me showing a question he asked Gemini about the Super Bowl on Saturday — the day before was the game was played. At right, you'll see my attempt to replicate his results on Sunday morning, still hours before kickoff.

JR

Gemini's varying answers to a simple question in the hours leading up to this year's Super Bowl.

All other issues aside, what good is an Android assistant if you can't trust it to provide even the most basic information reliably and you always have to question if anything it tells you is trustworthy and true?

A puzzling Google pivot

So here's what it boils down to: When it comes to an on-demand mobile device assistant, we don't need the ability to have mediocre text or creepy images created for us from anywhere across Android. We need a fast, consistent, reliable system for interacting with our phone and other connected devices, getting things accomplished with our core productivity services, and getting short bursts of basic info spoken aloud to us in response to simple questions.

Google Assistant did that. It did it with a recognizable, known brand Google has spent endless energy working to build up over the past several years and a recognizable, known voice Android users have come to trust and appreciate. Throwing all of that away now to create an entirely new system that introduces out-of-place, unnecessary additions and doesn't do the Assistant basics as effectively is a puzzling — if perfectly Googley, in the most facepalmy sense imaginable — move.

Now, polishing up Assistant, fixing its woes, and adding in Gemini as an optional add-on you could summon for its generative capabilities, if and when such a need were to arise? That could make an awful lot of sense. But positioning Gemini as a flat-out replacement for Assistant when it's so much worse at practically everything is an awfully strange move to make — one that seems to be forcefully trying to make the wrong tool work for a very specific purpose.

To be fair, it is early. And maybe Gemini will get better. (I sure hope so!) But either way, this is the future of Assistant Google is prominently showing off and making available to users this minute. And even if some of its mechanics do improve, the broader philosophical issues we've just been chewing over seem unlikely to shift substantially.

Speaking of which, at the start of this year, I posed a broad, philosophical question about all the hype-driven AI obsession that was clearly headed our way:

How much of the current rush to cram some form of "AI" into everything imaginable is actually about what's useful and advantageous for us, as human users of these creations? And how much is more about chasing the latest buzzword du jour and finding a reason to use the term "AI," no matter what it accomplishes or how it fits into our lives?

As I get to know Gemini in its role as an ultimate replacement for Google Assistant on Android, I can't help but come back to this query. It's painfully apparent why Google is pursuing this change from a business perspective — when you think about competitive pressure and investor demands around public perception and all of that. Sure. Fine. Whatever.

From an actual user perspective, though, it's tough to see how the shift to Gemini as an Android assistant makes sense or is in any way an upgrade for most people — y'know, us regular ol' mammals actually using this stuff in our day-to-day lives. Most of us don't need a "creative" chatbot at our fingertips all day long, in every area of our Android experience. We don't need on-demand image and text generation at our constant beck and call. And we certainly don't need long-winded, on-screen answers of questionable accuracy for our short spoken questions.

What we need is a simple, reliable task-handler and an accurate and concise info-relayer. Assistant established the framework for that. And living with Gemini for a while now, at least in its current incarnation, it's hard not to feel like Google's gotten caught up in the latest tech industry hype and forgotten what a phone assistant is actually for — and what we actual humans who use these products want and need from such a service.

Sign up for my free Android Intelligence newsletter to get practical tips, personal recommendations, and plain-English perspective on the news that matters.