Google I/O 2018 Recap

Google I/O 2018 was a quieter event than Apple's WWDC the same summer, and it shouldn't have been. The Duplex demo, Android P's behavioral changes, and the Maps AR rollout were the more important platform shifts of the year.

Tech Talk News Editorial4 min read
#google io#android#duplex#ai#developer conference
ShareXLinkedInRedditEmail
Google I/O 2018 Recap

Key takeaways

  • Google I/O 2018 happened in early May, and the Duplex demo was the biggest live tech demo of the year, showing Google Assistant booking appointments by phone with filler words like "mm-hmm" and "uh".
  • Google later clarified that Duplex would identify itself as an AI when calling, after public debate over whether the demo was real or scripted.
  • Android P used machine learning for system behavior via Adaptive Battery and Adaptive Brightness, and surfaced app actions and slices as one-tap shortcuts outside the app.
  • Google Maps AR used the Visual Positioning System to match the camera view against Street View imagery for positioning within a meter or two, but it did not ship to general users until 2019.
  • Gmail Smart Compose predicts the rest of your sentence in gray text that you accept with Tab, exemplifying Google's bet on ambient AI woven into existing tools rather than standalone products.

Google I/O 2018 happened in early May and was overshadowed in the news cycle by Apple's WWDC announcement a month later. Looking at what each event actually shipped, that ordering is backwards. WWDC 2018 was a refinement event. I/O 2018 was an inflection event, and the Duplex demo alone was the biggest live tech demo of the year. Worth recapping the parts that mattered now that the dust has settled.

The Duplex Demo

The headline moment of the keynote: Sundar Pichai played a recording of Google Assistant calling a hair salon and a restaurant on behalf of a user, holding a real conversation with a human on the other end, including filler words ("mm-hmm," "uh"), navigating an unclear hours question, and successfully booking the appointment. The audio quality was so good that within a day there was a real public debate about whether the demo was real or scripted.

Google walked back some of the messaging in the following weeks, clarifying that Duplex would identify itself as an AI when calling. The technical achievement was real regardless: a TTS model good enough to fool a human in a phone call, paired with a dialog system that could handle real-world ambiguity. Three years from now, this moment is going to look like the one where the assistant category jumped from "set timers and play music" to "actually do tasks for you."

Android P (9) Behavioral Changes

Android P was the version that started using machine learning for system-level behavior. Adaptive Battery learns which apps you use when, and aggressively backgrounds the ones you don't, in a way that meaningfully extends battery life on phones that adopt it. Adaptive Brightness learns your preferred brightness curve in different lighting and adjusts more naturally than the old rules-based auto-brightness.

The other big change: app actions and slices. The OS started surfacing deep-link actions from apps in the Google searchbar and the launcher, so a user could type "Lyft" and see a one-tap shortcut to request a ride to their default destination, without opening the Lyft app. That's a quiet shift in how Android wants users to think about apps: less as discrete launchpads and more as service providers that the OS surfaces contextually.

Google Maps AR

The walking-navigation AR demo was visually impressive: hold up your phone in an unfamiliar city, and Maps overlays directional arrows, street names, and points of interest on the live camera feed. Solves the problem every Maps user has had where you start walking the wrong direction because GPS hasn't quite figured out which way you're facing yet.

The technology underneath is Visual Positioning System, which matches what the camera sees against Street View imagery to determine position to within a meter or two, much more accurately than GPS alone. As of mid-2018 this was still in preview; it didn't ship to general users until 2019. The demo mattered anyway because it pointed at where Maps was going.

Smart Compose

Gmail's Smart Compose feature: as you type an email, the model predicts the rest of your sentence and renders it in gray text, which you accept by hitting Tab. After two weeks of using it, you stop noticing it's there and start subconsciously typing in a more predictable way to get the predictions to fire more often. That's a small UX change with a meaningful productivity impact.

Underneath, this was a signal of where Google's ML investment was pointing: at quiet, ambient features that improve everyday tools rather than at flashy standalone AI products. The same approach shows up in Photos, Docs, and the Google Assistant. Google's AI bet at I/O 2018 was on infrastructure-style integration, not on standalone products. It's still mostly the right bet five years later.

The Takeaway

I/O 2018's biggest contribution was demonstrating that ambient AI could meaningfully improve products people already use, without a new app, a new device, or a new paradigm shift. Duplex was the crowd-pleaser. The Android P battery model was the one that quietly affected the most people. Both pointed in the same direction: AI as plumbing, not as product.

Frequently asked questions

What was the Google Duplex demo at I/O 2018?
Google Duplex was a demo where Sundar Pichai played recordings of Google Assistant calling a hair salon and a restaurant and holding real conversations with humans, including filler words and booking appointments. The audio was convincing enough that a public debate broke out over whether the demo was real or scripted.
Did Google Duplex identify itself as a robot?
Google walked back its messaging in the weeks after the demo and clarified that Duplex would identify itself as an AI when calling. The underlying technical achievement stood regardless: a text-to-speech model good enough to fool a human paired with a dialog system that handled real-world ambiguity.
What did Android P change with machine learning?
Android P used machine learning for system-level behavior through Adaptive Battery, which backgrounds apps you rarely use to extend battery life, and Adaptive Brightness, which learns your preferred brightness curve. It also surfaced app actions and slices, letting you type something like "Lyft" and get a one-tap shortcut without opening the app.
How does Google Maps AR walking navigation work?
Google Maps AR overlays directional arrows, street names, and points of interest on the live camera feed when you hold up your phone. It relies on the Visual Positioning System, which matches what the camera sees against Street View imagery to determine position within a meter or two, more accurately than GPS alone.
What is Gmail Smart Compose?
Smart Compose is a Gmail feature that predicts the rest of your sentence as you type and renders it in gray text you accept by hitting Tab. It reflects Google's I/O 2018 bet on quiet, ambient AI woven into everyday tools rather than flashy standalone AI products.

Written by

Tech Talk News Editorial

Computer engineering background. Writes about software, AI, markets, and real estate, and the places where the three meet.

More about the author
ShareXLinkedInRedditEmail