Today’s well-known virtual assistants—Siri, Alexa, and Google Assistant—don’t quite match up to the sophistication of the latest AI chatbots, such as ChatGPT or Google Bard. However, once the advancements from the recent surge in generative AI are integrated into these older assistant models, we can expect a significant leap in their capabilities and intrigue.

To glimpse what the future holds, WIRED experimented with an AI voice assistant named vimGPT. It demonstrated remarkable efficiency by navigating to the WIRED subscription page and locating the sign-up form when prompted to subscribe to WIRED. Though it couldn’t complete the purchase without credit card information, its performance was notable.

What Could Possibly Go Wrong?

This instance underscores the complexity of online purchasing compared to the relatively simple tasks current virtual assistants handle, like setting reminders or fetching sports scores. VimGPT, developed on the GPT-4V model and utilizing the Chromium browser, excels in tasks from searching humorous cat videos to finding budget flights without storing personal data.

VimGPT, a project by developer Ishan Shah, showcases the potential direction for tech giants like Apple and Google in enhancing their virtual assistants. Shah’s creation, leveraging the multimodal capabilities of GPT-4V, demonstrates a more intuitive interaction with the web than text-based approaches, promising a shift towards more conversational and less click-dependent computer use.

Supercharged Siri.

The evolution of virtual assistants is echoed by Ruslan Salakhutdinov, a former Apple AI research director, who envisions a future where assistants like Siri are supercharged with AI to perform tasks more autonomously. Salakhutdinov’s work, including developing test environments for AI agents, suggests a future where AI can accomplish tasks such as shopping for specific items or managing online preferences with greater autonomy.

However, this vision is not without its challenges. Experiments have shown that while AI agents can perform complex tasks with some success, they still lag behind human efficiency and are prone to errors. These testing environments allow for the safe exploration of AI capabilities and limitations, informing further development.

A New Era

The potential for AI agents to transform digital interactions is immense, promising to make navigating the web and managing online tasks simpler and more intuitive. As big tech companies continue to innovate in this space, the future of virtual assistants looks set to redefine our expectations of AI-powered help, and change the employment landscape dramatically.

An oddly beneficial outcome from a cybersecurity point of view is that these agents can be developed with security detection, defense and protection from the ground up. And if we like, they can also be trained in offensive security, heralding a new era in our long battle against bad actors.