There are a lot of technical and regulatory hurdles to be cleared before smartphones can natively run generative AI apps.
In the just-over six months since ChatGPT’s first public release, AI technology has continued to evolve at a dizzying pace, with seemingly game-changing announcements coming nearly every week. The last two months in particular have seen a number of technology companies making major announcements about implementation strategies for generative AI and AI technology in general.
However, as quickly as this technology is moving, there are two big reasons that it’s going to be a while until consumers see generative AI apps on their smartphones:
- The pace of technological evolution is likely to slow, given that government agencies and leading AI experts alike are calling for immediate regulation.
- Generative AI is still hugely processor-intensive.
Regulatory concerns have created a patchwork of AI availability, making new AI products unavailable in a growing list of countries.
OpenAI released an app version of its generative AI chatbot, ChatGPT, for iOS in late May, allowing native access to iPhone users with phones compatible with iOS 16.1 or later without needing to use a web browser — which offers many benefits for both convenience and accessibility. However, Android users still need to access ChatGPT via web browser — though OpenAI promises an Android version of the app is “coming soon”.
Of course, ChatGPT is far from the only generative AI chatbot on the market. In early May, both Google and Microsoft opened access to their own generative AI chatbots — Google Bard and Bing AI — for some users. Less than two weeks later, at their annual Developers Conference, Google announced plans to upgrade key Google Suite utilities with generative AI capabilities to help users with internet searching and writing emails — both features that are currently available to some Google users through Workspace Labs — Google’s “trusted tester program” that allows users to try new AI features.
Access to Bard as well as Workspace Labs is available to users in nearly 200 countries, including some notoriously repressive countries like Eritrea. What’s surprising is where it’s not available: Canada, with Germany, France, and several other European countries also considering restrictions.
The reason? The concern most cited by regulators currently is data privacy and data collection. By default, ChatGPT saves your chat history and uses it to train the AI model running the technology. However, ChatGPT has no age verification mechanism in place, giving it no ability to comply with tighter regulations for collection of data from children.
Surprisingly, even with the uncertainty created by regulatory scrutiny, top AI companies and researchers are also calling for limits. There are disagreements about the particulars of what regulation of AI should look like, but most major experts in the field — including Bill Gates, Steve Wozniak, and Elon Musk — agree that regulation is necessary.
Generative AI is still too processor-intensive to run natively on smartphones.
There’s a reason why text-based generative AI is also known as “large language model” AI. ChatGPT is trained on a data set of 300 billion words and has 570 gigabytes of text data. (Imagine trying to fit all that and your favorite music on your phone’s SD card!) The training data is also limited to data until September 2021, meaning that a truly up-to-date set would be even larger.
By comparison, Stable Diffusion — the protocol used in several popular online AI art generators like Midjourney — is a much more manageable 8GB. However, the processor power required to run Stable Diffusion is unmanageable for all but the most serious desktop gaming rigs. Stable Diffusion simply won’t run on anything with an integrated GPU (Graphics Processing Unit).
Even the process of installing stable diffusion is not for the faint of heart, requiring first downloading several utilities as well as using command line prompts to set everything up. (It took me about two hours to install at home, after some false starts and troubleshooting.)
All of which means that for a while, consumers will be restricted to using web-based applications for text-based AI like ChatGPT. Similarly, AI art will be handled through web-based tools like Midjourney for all but the most hardcore gamers and graphics enthusiasts.