Gemini unveils ‘Deep Research’ for complex research tasks

Google has enhanced its Gemini chatbot platform by introducing a feature called "Deep Research," designed to tackle complex research tasks and compile detailed reports. This new capability, available exclusively on the Gemini Advanced tier under the Google One AI Premium Plan ($20/month), employs "advanced reasoning" and "long context capabilities" to generate research briefs that can be exported to Google Docs for further refinement.

The feature acts like a research assistant, analysing relevant web content to address user queries. Users submit a question, and Deep Research devises a multi-step research plan for their review. Once approved, it conducts iterative searches, saving useful information, refining its focus, and generating a comprehensive report of findings along with source links. Initially, the feature is available in English on desktop and mobile web platforms, with a rollout to the Gemini mobile app expected in early 2025.

The introduction of Deep Research raises ethical issues. Critics argue that it might hinder critical thinking and learning among students, who may over-rely on AI tools for academic tasks, potentially leading to procrastination, memory loss, and lower grades. Moreover, publishers may face financial losses, as the tool’s ability to scrape and summarize content could reduce site traffic and ad revenue. Past examples, like Google's AI Overviews for Search, have reportedly caused traffic declines of 5% to 10%, with potential losses for publishers reaching billions of dollars.

While Google claims the feature could increase discoverability for websites, the actual impact remains uncertain, with concerns it may instead divert user attention from original sources.

In addition to Deep Research, Google has introduced "Gemini 2.0 Flash," an experimental version of its flagship AI model optimized for chat. Available for both free and paid users, the model promises faster and more efficient performance. However, some Gemini features may not yet be fully compatible with this experimental version. The complete 2.0 Flash model is slated for release in January.