I Could've Used Qualcomm's ChatGPT-Like Phone AI on My Trip to Hawaii
Commentary: Qualcomm showed off generative AI on phones during its Snapdragon Summit, and I already thought of ways it could be helpful.
David LumbMobile Reporter
David Lumb is a mobile reporter covering how on-the-go gadgets like phones, tablets and smartwatches change our lives. Over the last decade, he's reviewed phones for TechRadar as well as covered tech, gaming, and culture for Engadget, Popular Mechanics, NBC Asian America, Increment, Fast Company and others. As a true Californian, he lives for coffee, beaches and burritos.
Expertisesmartphones, smartwatches, tablets, telecom industry, mobile semiconductors, mobile gaming
At Qualcomm's Snapdragon Summit in Hawaii in late October, the company revealed its next set of chips including the Snapdragon 8 Gen 3, which will power next year's top Android phones, likely including the Samsung Galaxy S24 series. In addition to expected improvements in performance and battery efficiency, the new silicon brings on-device generative AI to phones. At the summit, I saw Qualcomm give demos highlighting situations in which the feature could help users out. But it was during a day trip Hawaii adventure that I thought up specific ways that smartphone generative AI could be truly helpful.
Following the summit, I remained on Maui to, ahem, do more research. Other summit-goers suggested a staple adventure of the island: the famed Road to Hana. From the Kahului Airport at the north of the island, it's a bit more than a two-hour drive to Hana on the northeast side of Maui, but only if you don't stop at the dozens of rest points, attractions, swimming pools, waterfalls, state and national parks, eateries and scenic views of the Hawaiian coast.
The Road to Hana is as famous for its beauty as it is infamous for its challenges. The road rises hundreds of feet above the waterline and drops back down several times over the course of the 620 hairpin turns that block views of oncoming traffic. The road is barely wide enough to fit two cars abreast, except when it narrows to a single lane over 59 old concrete bridges that only fit one car at a time, forcing drivers to negotiate crossings with cars on the other side.
And for most of the drive, there is zero cellphone service.
As I encountered obstacles and made discoveries about the road, I couldn't help but ponder what on-device AI could've helped with. In some ways, it might have told me exactly what I needed to know before I did it. But there are still things our current vacation survival guides and simple word-of-mouth do better to help us than any AI.
Making plans for a last-minute day trip while abroad
Let's start from the beginning. After waking up Saturday morning from a post-summit sleep, I made the call at 8:30 a.m. to commit the whole day to the Road to Hana. I did as much research as I could, reading official guidebook sites and TripAdvisor forums and watching YouTube videos for a rundown of the day. I had a baseline of knowledge to make a timeline, plus advice from previous travelers about must-see experiences.
I needed to make a limited agenda since I set out in my rental car at 10 a.m. Hana Highway veterans suggest leaving at daybreak to get the most out of the experience. I guessed which attractions would suit me and hoped for the best as I drove up from Maui's southwest coast.
Here's how I think on-device AI could've helped: Several of Qualcomm's demonstrations at the summit showed people asking a Snapdragon-powered generative AI to make a travel plan. Rather than spend an hour researching, I could've had the AI give me a plan in seconds, and refined it with follow-up questions. Or better yet, I could've jumped in the car earlier and had the AI make plans for me on the go.
It's not just that the on-device AI could make planning faster. Qualcomm says that the virtue of on-device AI is that it can consider more of your personal data since it's being processed on the device and not going up to the cloud. The AI can draw conclusions about what people like from their behavior patterns, locations visited and photos. My device-generated itinerary could've recommended more beaches among the other attractions, which is what I ended up doing anyway only after guessing what things would've been less appealing for me.
Getting directions when signal was out
As online forums and videos suggested I do, I bought an app audio guide (the Gypsy Guide specifically for Road to Hana) that voiced suggestions and area lore, which was brilliantly triggered when I passed specific GPS coordinates, effectively timing the advice ahead of upcoming attractions so I could choose if I wanted to see them.
This audio guide was invaluable in helping me decide whether I should pull over or keep driving, especially as a solo explorer. But here's the rub: Even if I had other people in the car with me to research while I drove, that wouldn't be helpful once their phone signal cut out, which was the case for much of the Hana Highway.
On-device AI could supplement the audio guide in other ways while on the drive. If I veered away from the guide's path, I could ask the on-device AI for navigation help to find my way back to the main road. If there's a map downloaded to the phone, the on-device AI could give me step by step directions contextually dependent on where I'd driven off to. As it was, I got too worried about leaving the main road to venture out.
Filling in photo edges? Only if my photo is crooked
Here's another fun challenge about exploring alone: You're taking all your own selfies, especially if kind strangers aren't around to help. Even with friends, there are situations like in low light where I'd want to use the more powerful rear cameras to take better photos of us. I've developed a curious workaround of setting my camera to ultra-wide and flipping the phone around to blindly shoot selfies. If I want to make more of an effort, I use the Apple Watch's camera app to try to preview the image, requiring extensive angling of phone and watch. It works about as well as you'd think.
Qualcomm showed off another Snapdragon 8 Gen 3 on-device trick at the summit: using generative AI to expand the borders of a photo and predictively fill in what wasn't originally captured. In practice, this could mean a photo taken with the main camera is made to look like it had been taken with an ultra-wide lens instead.
But I found another potential use: fixing those janky photos you take accidentally. If I mistakenly took a photo at a Dutch angle, most phones' existing editing tools will let you tilt it back to being level with the horizon — but that might end up shaving off too much of the sides. Generative AI could fill in the photo's sides while keeping most of the original subject in the center of the shot.
There's also the neat trick Qualcomm showed off on stage for removing unwanted subjects out of videos. Like Google's Magic Eraser or Magic Editor that deletes people or things out of photos, a Snapdragon 8 Gen 3 phone could do the same fix for video, which would be nice given all the other people I inadvertently captured while shooting my litany of "I'd come back to this beach" videos.
Telling me if I had time to stay out or turn back
I triumphantly passed through Hana and technically finished the road, but my audio guide said there were still a few must-see areas beyond. As the sun started dipping behind Haleakalā, I got nervous about the hairpin turns and bridges that I didn't want to navigate in the dark. I stopped at a crossroads with the famous Wailua Falls ahead of me — and out of an abundance of caution, I turned back up the road to reverse my journey.
I could've pushed on, and I regretted not feeling confident enough to see the vaunted pools below the waterfalls. Still out of signal range (I occasionally got a single bar of LTE), I would love to have asked a specific question: Do I have time for a quick visit and to get past the worst of the turns and one-lane bridges before nightfall?
Those are the kind of "I wish I had an expert sitting next to me" questions I'd have loved to get answered on the spot. The audio guide cleverly checked in with seasoned advice depending on how far people got, so I had some sense of whether I should return, but nothing smart enough to consider timing the sunset for the current date or forecasting road conditions like whether traffic would slow me down enough before it got dark. Or, with personalization, taking into consideration how long my "quick stops" usually ended up being.
With up-to-the-minute travel info, on-device AI could take into consideration my grumbling stomach and laid out a few options for dinner, given my timetable. In a place like Maui where shops don't always keep regular or expected hours (alas for missing Sam Sato's that closed at 2 p.m.), it would be nice to have an inkling of dinner options on the drive back.
Keeping me company on the long road
Much hay has been made about generative AI's capability to instantly write short stories or create images based on prompts, and unskilled people have rejoiced in the possibility of using ChatGPT or Midjourney to get rich quick with regurgitated material. But there's potential in storytelling that isn't used to feed capitalism's yawning maw for content, and instead to help bored travelers pass the time.
The audio guide came in extremely useful here, especially on the return journey as it thoroughly summarized the history of mankind's millennia coming to and living on the Hawaiian islands. But what it couldn't do was answer any follow-up questions I had. For instance, when Sanford Dole led business leaders in a coup dethroning the last monarch of Hawaii, Queen Lili'oukalani, leading to the annexation of the islands by the US and disenfranchisement of the Hawaiian people in favor of business moguls, I could've asked for more clarification on the impact of America's colonization of the islands.
There are other neat ways generative AI could help other folks in the car, like backseat gremlins getting restless who may calm down if a generative AI tells them stories. If some promising AI models are to be believed, we could even see AI-generated video soon, which would certainly help placate folks until a return to mobile signal range to stream their trusted favorites.
And then there's the stalwart driver, stoically handling every harrowing turn, who might welcome some conversation, however auto-generated. If I hadn't had the foresight to download a couple podcasts to my phone, I would've had silence between the audio guide's travel tips and history lessons. Heck, even though I fully downloaded those podcasts, Apple's perpetually horrendous Podcasts app sometimes demanded internet connectivity to play them.
In summary, I survived my solo adventure thanks to quick planning, and making tough calls on the road to forgo what could have been incredible sights and envy-inspiring Instagram photos. But offloading a lot of that cognition or even having a voice to help me sort my thoughts could make travel a lot easier. Take this all with a grain of salt, of course, as we've yet to personally test Qualcomm's generative AI in a real product. (The Snapdragon 8 Gen 3 launched in its first phone, the Xiaomi 14, during the summit, but we'll have to wait until next year to see most of the handsets that will pack the chip.)
Still, the possibilities remain. Who knows how having a bonus brain in your pocket will remove some of the friction of travel and make the views that much easier to enjoy.
Editors' note: CNET is using an AI engine to help create some stories. For more, see this post.