When Maps Make Decisions: Mobile AI,Navigation Apps,and the Mobilization of Mobile Data

Introduction

People once talked about “reading” maps. But the maps that we now use most are no longer straightforwardly readable representations of cartographic information; instead, they show us how, when, and which way we should travel. They update us on changed conditions, advertise local services, and talk us through navigation. They respond to spoken queries and are vocal as well as visual. They make predictions and decisions for us, through interfaces that are increasingly voice-driven and conversational. Contemporary maps are creatures of “mobile AI,” the linkage of machine learning with communication and mobility, which we now think of as a significant “next phase of digital communication” (Goggin, 2025). In this article, we examine the current AI-facilitated developments that are reshaping mobile maps and navigational apps. Building on a rich body of scholarship on smartphone map use (including Alvarez Leon, 2019; Duggan, 2018; Özkul, 2015; Wilmott, 2020), in this commentary paper we consider how AI tools are changing the maps that we have (how maps are made), their function (what maps are doing), and how we interact with mobile maps and navigation services (how maps are used).

Our argument is that contemporary mobile navigation depends on an intensive mobilization of mobile data, with far-reaching consequences for the mobile digital economy. Navigation apps rely on the continuous capture of locational and contextual data from user devices, their aggregation with extensive historical and real-time datasets, and their processing within large-scale networked and computational infrastructures. Much of this work occurs upstream from the device, in cloud platforms and data centers that model traffic patterns, predict travel times, and generate recommendations in real time. The mobilization of mobile data transforms navigation from a discrete, user-initiated act into a persistent, automated service that accompanies everyday movement.

This transformation involves a distinctive mode of automation. Navigation apps automate tasks that were once performed by drivers, passengers, and transport authorities—such as route planning, wayfinding, and the interpretation of traffic conditions. But automation here also operates at an organizational level. Decision-making about routing, timing, risk, cost, and attention migrates from users and public institutions into platform models and private infrastructures that classify, predict, and optimize mobility. These systems necessarily prioritize certain objectives—often speed or efficiency—while the criteria for decision-making are opaque. It follows that navigation apps are likely to generate new externalities, redirecting traffic, reshaping urban rhythms, or privileging certain destinations and services over others.

To understand how this form of automation has emerged, this commentary paper situates mobile navigation within longer trajectories of geospatial artificial intelligence (GeoAI) and platform development. While global positioning systems were crucial in enabling mobile devices to determine their location, machine learning has long been essential to the further capability of navigation itself: recognizing spoken inputs, predicting traffic conditions, and dynamically recalculating routes. From early experiments in AI-assisted cartography to contemporary platform investments in large geospatial models and generative AI “grounding,” GeoAI has reshaped both the production of maps and their integration into digital infrastructures.

We then turn to recent developments in mobile AI that bring these transformations into everyday use, focusing on the growing prominence of voice-driven and conversational interfaces. Features such as conversational reporting, open-ended natural language search, anticipatory commute alerts, and persistent place histories extend navigation beyond the act of driving or walking. Maps now speak, listen, and intervene, offering recommendations and notifications even when users are not actively navigating. These developments foreground a renewed importance of orality in mobile media, alongside the visual, and further entrench navigation as an automated, predictive, and personalized service.

By analyzing mobile maps at the intersection of GeoAI, platformization, and mobile AI, we show how the automation of navigation reorganizes everyday mobility, redistributes decision-making, and generates new political, economic, and social questions. Mobile maps offer a particularly revealing case for understanding how AI is being embedded into mobile media, how value is extracted from movement, and how infrastructural decisions are increasingly delegated to platform systems.

GeoAI: Geospatial Artificial Intelligence and Digital Platforms

Emergent developments in mobile media and AI stem from long-standing research within the geosciences disciplines that explores what has become known as geospatial artificial intelligence, or GeoAI (Li, 2025). Early, speculative work exploring the possibilities and potential interactions between AI and geoscience dates to the mid-1980s and mid-1990s (e.g., Couclelis, 1986; Openshaw and Openshaw, 1997; Smith, 1984). Geographic and cartographic interest in GeoAI accelerated in parallel with rapid developments in AI, large language models, and machine learning systems from the 2010s onwards (Gao et al., 2023). In a 2024 review of contemporary GeoAI developments, it is noted that AI applications of greatest interest within cartographic research are those “related to the map’s ‘mechanics’ or basic construction, such as map generalization, map reading, and map production” (Kang et al., 2024: 617).

For large technology platforms, GeoAI opens the door for the pursuit of “new AI-supported approaches for making maps” (Griffin and Robinson, 2025 - emphasis added) and new or revised means of interacting with maps. Large-scale digital platform businesses, such as Meta, Apple, Alphabet, and others, have been quick to seize upon these opportunities. In 2016, Meta revealed they had been working with OpenStreetMap or OSM (an early adopter of AI mapping tools and techniques), and using advanced deep neural net models, to perform AI-assisted road tracing from satellite data (OSM, 2021). By 2019, this work became Map With AI, a new service for accelerating the cataloguing of roads and other physical features (Gao et al., 2019). For its part, Apple is repurposing photos gathered from the “Look Around” feature within Apple Maps to train the AI models that power Apple products and services, “including models related to image recognition, creation, and enhancement” (Liszewski, 2025). This repurposing of mapping data for training consolidates Apple’s position across multiple markets simultaneously—strengthening its computer vision capabilities while deepening the dependency of developers, businesses, and municipalities on platform infrastructure they do not own and cannot audit.

Alphabet, meanwhile, continues to integrate AI into its geo-related operations. In 2025, it launched Grounding with Google Maps in Vertex AI (Google for Developers, 2025), a feature that helps developers build generative AI applications that draw from up-to-date Google Maps data so that applications can display the most accurate geolocation information possible (Maguire, 2024). Alphabet also employs AI to stitch aerial photographs together and remove clouds (Kelly, 2024) so as to ensure that “it’s springtime everywhere” on Google Earth (Chapman, 2020). Further, in 2025 Alphabet announced AlphaEarth Foundations, “an artificial intelligence model that functions like a virtual satellite” (Google DeepMind, 2025b) and which now powers its Google Earth Engine. As one Google DeepMind engineer puts it, “much like Google Search has indexed the web, with AlphaEarth Foundations, we’ve indexed the surface of the planet” (Google DeepMind, 2025a).

Niantic Labs is seeking to move in the same direction. Niantic recently sold off its games business—including Pokémon GO—to Scopely (Niantic, 2025) and has relaunched as Niantic Spatial, a geospatial AI company. Just as there are large language models (LLMs), Niantic Spatial aims to create a Large Geospatial Model (LGM) that is populated from its store of 10 million mapped locations and game player contributions of “anonymised data through scans of public landmarks” (Criddle et al., 2025). The idea is that LGMs might build a “world model” that “can better understand” and interact with “human environments” (Criddle et al., 2025). In this way, Niantic Spatial’s ambition is to employ AI-facilitated “geocomputation” (Openshaw and Openshaw, 1997: 317) to provide “geospatial intelligence” (Brachmann and Prisacariu, 2024) as a service.

To recapitulate, GeoAI combines the spatial analysis capabilities of Geographic Information Systems (GIS) with AI to extract meaningful insights from large-scale spatial data, such as satellite imagery, aerial photos, and various forms of sensor data. Large digital platforms have been quick to use GeoAI to automate many of their workflows and to further enrich their databases with geolocation information. These platform developments are consequential for contemporary critical understanding of mobile media and maps as they reveal the full extent of GeoAI integration into all aspects of the metaphorized “platformization tree” (van Dijck, 2021), from the “deeper layers” of data capture through to the upper-most “branches” that comprise “sectoral apps,” such as mobile maps and navigation apps.

Mobile AI

Mobile AI has been defined as “AI designed for, integrated in and associated with mobile technologies and mobilities” (Goggin, 2025: 3). AI integration into smartphones, in general, has advanced rapidly in recent years, with “mobile device vendors, carriers and technology providers, infrastructure and data companies, and other key actors all hav[ing] progressively expanded the role and salience of AI in their offerings” since the early 2020s (Goggin, 2025: 3). It is valuable to understand the integration of AI into mobile devices as an imperative manifest at all levels of system design, from fundamental hardware to the user interface. Among the major device vendors, for example, Apple has Apple Intelligence, its overarching iOS AI framework, powered by its Neural Engine. This is the label Apple has given to its neural processing unit (NPU), hardware that is dedicated to the acceleration of AI and machine learning operations (Stine and Wilken, 2025). Samsung, meanwhile, has Galaxy AI, which incorporates a range of AI-facilitated assistive features (for image and text editing, transcription, translation, and health monitoring).

Mobile AI has emerged over a comparatively long period, at least in terms of the life of the smartphone. While they have not been obvious to users, the tools and techniques we now think of as “AI” have been integrated into core mobile services since the early years of the smartphone, and many of the features of apps we have long associated with smartphones would not exist without some kind of AI. Image processing using machine learning was developed in the early 2010s, serving, for example, as a critical element in the production of Google’s Streetview. Voice to text input is another key example, a service built into the system level of smartphones within a few years of the launch of the iPhone and Android ecosystems. The reliability and utility of voice capabilities also developed rapidly during the 2010s with the deployment of deep learning neural networks. When it comes to mobile-mediated navigation, AI-facilitated assistance manifests most directly as an embrace of voice-driven systems, which, as we detail below, can be seen in the use of conversational AI and open-ended natural language search.

Mobile (Geo)AI and Maps

How, then, do advances in GeoAI converge with recent developments in mobile AI? Our contention is that the integration of mobile AI into smartphones is now increasingly reliant upon geospatial artificial intelligence. GeoAI works in conjunction with mobile AI to provide context-aware, location-based intelligence that allows mobile devices to interpret user inputs relating to the physical world and act upon these in real time.

Mobile AI (with GeoAI capacity) is now being further integrated into smartphone mobility and navigation services in ways that meaningfully impact how consumers use and interact with these services. To take one example, Alphabet subsidiary Waze has rolled out a Gemini-powered “Conversational Reporting” tool that seeks to make it safer for Waze to capture crowd-sourced data from its users relating to road incidents and traffic hazards through hands-free interaction (Schoon, 2025). As Waze explain it, “All you need to do is tap the reporting button and speak naturally, as if you’re chatting with a friend: ‘Looks like there are cars jammed up ahead!’” (Berkovich, 2024). Private maps firm Mapbox has taken this even further. Led by a US$280m investment by Softbank, Mapbox has created the playfully titled MapGPT service that equips connected cars and apps with AI maps. MapGPT is billed as “the first AI assistant that can have in-depth conversations about directions, landmarks, roads, and other highly dynamic aspects of the world,” which provides Mapbox and its automotive users with constantly updated location data (Mapbox, 2025).

Another significant step change in mobile navigation involves the integration of open-ended natural language search. This is being rolled out in both Google Maps and Apple’s Maps. For Google, map-related “Gemini curation” (Lekach, 2024) involves a shift toward processing “open-ended search queries,” such as searching “things to do with friends at night in Boston” or “fun fall activities in Seattle” (Cai, 2024), to use two US examples. The Gemini LLM, it is envisaged, will also enable natural voice conversations between the user and the AI about the particular attributes of possible destinations such as restaurants. It will draw upon both the navigational mapping data and Streetview to provide clearer, human-relatable, contextualized recommendations, such as “turn left at the park” instead of “turn left in 200 meters.” Apple, for its part, also aims to use natural language search in its Apple Intelligence, encouraging users to “search the way you talk,” by asking geolocation questions such as “show me cafes with Wi-Fi” (Nield, 2025). It is also rolling out personalized Preferred Routes and Visited Places. For Preferred Routes, the iPhone now uses “on-device intelligence” (Apple, 2025) to learn and anticipate frequently visited places, such as home and work. An Apple Maps widget then presents a preview of a user’s commute, together with notifications, to alert the user of “significant delays and offer alternate routes, even before their journey begins” (Apple, 2025). This means, of course, that a user does not need to be using a maps app or navigating a route to be notified about changes in traffic flows or other material developments. With Visited Places, it is suggested that “users can allow iPhone to intelligently detect the places they visit and spend time in – like restaurants or shops – and they’ll automatically be saved to Maps” (Apple, 2025).

These developments signal a significant shift in the form and nature of mobile navigation and place-finding. Through conversational interfaces utilizing natural language processing, the mode of engagement shifts from instrumental directional instructions (“turn left in 200 meters”) to conversation-driven, contextually aware, navigation, and anticipatory problem-solving and route/venue recommendations. For instance, a mobile AI might interpret a navigational inquiry in its temporal, locational, and social context, drawing the inference that it could be a suggestion to reserve a table at a restaurant, or respond to a more complex logistical problem. ¹ Furthermore, in all these instances, we can see a strong element of automation in the capabilities made possible by AIs, where they are substituting machine-driven communications—notifications, messages, queries—for tasks once undertaken by humans, such as the driver of a vehicle or their route-finding companion. Researchers have pointed out that some important shifts are underway in this transformation, giving rise to new kinds of social and practical problems, as well as addressing users’ needs. For example, AI-driven recommendation systems such as those used by maps applications may often fail to take into account the interests of third parties, if they are designed to optimize travel times. Will the AI take into account the value of a quiet neighborhood street, or the frequent presence of children on an alternative route? Will it consider the environmental footprint of different routes? (Google’s Maps app suggests that it does.) Mobile AI creates not only new capabilities for its users and new sources of value for platforms, but also new kinds of politics.

Conclusion

Goggin (2025: 2–3) suggests that AI may change the trajectory of the smartphone by pushing it further away from being a “phone.” This is a persuasive claim when we consider generative AI features that foreground multimodal composition, together with the prominence of platforms geared toward social distribution. Yet the integration of AI into mobile maps and navigation complicates the narrative, raising questions about what could be “phone-like” in the AI-enabled smartphone. Orality may be one of those attributes. While mobile navigation remains in large part a visual experience for most users, its growing reliance on voice interaction, conversational interfaces, and anticipatory notifications points to a renewed centrality of communicative infrastructures in everyday mobility. Mobile AI intensifies personal communication, embedding speech, listening, and continuous signalling into the coordination of movement. The AI-powered smartphone, as Fortunati (2023: 21) suggested, continues to function as a “connecting tool,” which will also continue to have far-reaching political, economic, and social consequences.