The Ultimate Guide to Smartphone AI: On-Device AI vs Cloud AI vs Hybrid

Some might argue that translation and transcription in real time were available long before AI, but now that these systems are able to understand the context, the end result is far better. You can also use AI to summarize text for you and write or paint for you, which is the generative part.

Facial recognition is now much more accurate thanks to AI, for better or worse, and AI can easily find search patterns and offer contextual information tailored to a person's specific needs.

Navigation systems also utilize AI now, and they're better than ever. These include self-driving systems as well, which use computer vision to recognize objects. Speaking of computer vision, it's the mechanism behind Circle to Search and other image search systems. AI is able to make sense of images and interpret them.

So, AI is all around us and is not going away anytime soon. Let's see what we've got on our smartphones.

AI in smartphones

Every big (and not-so-big) smartphone brand is looking to hop on the AI train. Google, for example, rolled out AI-powered capabilities like Magic Editor for background fills and adjusting subjects and Best Take, which merges several pictures to produce the optimal group photo. Samsung has its own Object Eraser designed to eliminate objects from photos and some clever Galaxy AI features such as real-time translation in calls, photo editing suggestions, and generating summaries both in Notes and while browsing.

OnePlus introduced an AI Summarizer for concise phone call recaps, and Honor has its own system called Magic Portal that can extract text and images and send them to apps and services.

Another hot topic that smartphone companies talk about all the time is "On-device AI." They also seem to be more hush about the other type, called "Cloud-based AI," and there's a good reason for this. Let us explain.

On-device AI

Well, it's self-explanatory, really. On-device AI means that the system is running on the smartphone. It may use dedicated silicon or part of the main processor, but the important thing is that such a system, in theory, doesn't need an internet connection to work, as all the computation is done on the smartphone itself.

Google's Tensor chip, for example, has a dedicated edgeTPU (tensor processing unit) part for accelerating AI tasks, while Qualcomm has included a new AI Engine to support up to 10 billion parameters of generative AI models—solely on-device in the Snapdragon 8 Gen 3 chipset.

There are a lot of benefits to the On-device AI model, and that's one of the reasons many companies are vocal about it (even though most of the time their systems are hybrid ones, but we'll get to that in a minute).

Here's a list of the advantages and disadvantages of On-device AI:

Advantages:

• Latency: Immediate processing as data doesn't need to be sent to and from a server, resulting in faster response times.
• Privacy: Data remains on the device, reducing the risk of data breaches and ensuring better user privacy.
• Offline Functionality: It can operate without an internet connection, making it useful in remote or unstable network environments.
• Bandwidth: Reduces the need for constant data transmission, saving on data usage and reducing network congestion.

Disadvantages:

• Resource Constraints: Limited by the hardware capabilities of the device, such as processing power, memory, and storage.
• Update Management: Deploying updates to models and algorithms across many devices can be challenging.
• Scalability: It is more difficult to scale as each device needs to individually handle computations.

On-device AI is the holy grail of security and self-sufficient AI experience, as no information is (theoretically) sent to the cloud. The thing is, however powerful our smartphones might be, they just can't compete with a dedicated data center with hundreds of processors running AI tasks.

Cloud-based AI

Cloud-based AI systems run AI models and process data on remote servers hosted in data centers. Your smartphone must be connected to the internet in order to send and receive requests to those datacenters, and most of the time, not only requests but some user data is sent over the internet in order to fulfill the task at hand.

Cloud-based AI is extremely powerful, as it utilizes the huge processing power of data centers specifically developed to handle AI. However, there are drawbacks and security concerns with such systems.

Here's a list of the advantages and disadvantages of Cloud-based AI:

Advantages:

• Resource Availability: Access to virtually unlimited computational resources, enabling the use of more complex and larger models.
• Scalability: It is easier to scale processing power and storage as needed without worrying about individual device capabilities.
• Management: Easier to manage, update, and maintain models and algorithms from a central location.
• Data Aggregation: Facilitates the collection and processing of large datasets from multiple sources.

Disadvantages:

• Latency: Potential delays due to data transmission to and from the cloud.
• Privacy Concerns: Data needs to be transmitted to external servers, raising concerns about data security and privacy.
• Dependence on Internet: This requires a stable internet connection, which can be a limitation in certain environments.
• Cost: Can incur higher operational costs due to data transfer and cloud service fees.

You can see why companies don't talk that much about Cloud-based AI, especially in this day and age where people are extremely sensitive about their private data. Another caveat is that these systems will most definitely transition to subscription models after the first couple of years of hook and hype.

Hybrid AI

The reality is that most smartphone companies use a hybrid AI model, meaning some of the features are on your smartphone and some use a datacenter far away. Some AI features also depend on an internet connection, such as Circle to Search. Without the internet, you can't search for anything in the first place.

The sketchy part is that these hybrid models tend to connect to the cloud even when using on-device silicon, and when you switch off your internet connection, some supposedly on-device features stop working.

Galaxy AI, for example, uses on-device AI for photo gallery suggestions but connects to the cloud for generative photo editing. It uses both methods for speech-to-text transcription and live translation. (You can check out which Galaxy AI features run on-device and in the cloud here.)

Some companies are also quite vague about what features are on-device and which ones need internet access, and we might need some new regulations in place to protect users.

Conclusion

AI is here to stay, that much is clear. And for the most part, it will make our lives easier. Some people compare this AI revolution with the industrialization or the internet revolution, and they might be on the right track with such allegories.

The not-so-obvious problem lies in how we use such systems, not so much in privacy and security. Our reliance on AI might render some of our skills obsolete, and they might deteriorate with time. We won't need to learn foreign languages, learn how to use a map, or summarize text and edit pictures. What would come of it is still unclear. What do you think of smartphone AI and AI in general? Let us know in the comments below!