Google I/O Conference 2025*

*features an interview with Dr. Andrzej Żurański, senior data scientist at Google

Link to Podcast Interview with Dr Andrzej Żurański

By Ananya Chopra, Kimberley Sun

The Lawrenceville School, NJ

Since ChatGPT’s first release in 2022, Artificial Intelligence (AI) private investment in the US alone has nearly doubled, from roughly 60 billion USD in 2022 to 109 billion USD in 2024.^[1] Amidst the explosion of the AI industry, tech giants have quickly seized this lucrative opportunity to compete for and build AI innovations. In a matter of a few years, AI chatbots have been integrated into nearly every aspect of society, from academic tools to corporate tools and customer service. Interestingly, this has ushered in a new era of the public perception of AI, where the term “AI” has gradually come to mean generative AI, specifically large language model (LLM) chatbots, such as ChatGPT, Deepseek, and Gemini.

Revenue Associated with AI

Source: Morningstar, Gartner. Data as of February 2025.

At the recent 2025 Google I/O Conference in Mountain View, California, thousands of developers gathered to attend the public announcement of Google’s latest technology and products. Google, like most other tech giants, has jumped on the AI wave to revolutionize not only corporate AI involvement but also the public perception of AI and robot thinking. With an aggressive approach towards AI, as stated by Dr. Andrzej Żurański, senior data scientist at Google and attendee of the conference, Google emphasized its mission with the products presented at the conference. Compared to previous years, Google barely presented any new hardware in this year’s conference at Mountain View, instead focusing almost entirely on AI technology and software. Dr. Żurański acknowledged that this may be a strategic decision, since Google “was never really very good at [hardware] and it wasn’t a priority for anyone.” The conference’s promotional nature became evident as Dr. Żurański noted that “quite often what’s being shown at that conference doesn’t necessarily exist or function, and the main purpose is to show interesting things that Google is working on.” Among other products, the 2025 conference introduced an AI search mode, one which, if commercialized, would completely overturn the conventional Google Search page rank algorithm; a release of Gemini 2.5 Pro, a new Gemini chatbot version leading in nearly all AI chatbot benchmarks. The conference also included a feature with real-time speech translation, encapsulating not only the message but also the tone and emotions of the speaker.^[1]

Google I/O 2025 centered around three primary themes: making AI more intelligent, agentic, and personalized, reflecting this vision through several key product launches. The Gemini app received major updates, including Veo3 and Imagen4 to DeepResearch and Canvas, representing Google’s push into comprehensive creative AI tools. Gemini 2.5 Pro and a new version of Gemini 2.5 Flash became generally available in early June, with enhanced language capabilities and real-time code processing. Major updates to the Gemini app include Gemini Live with camera and screen sharing, now free on Android and iOS. Most significantly. Gemini 2.5 Pro now outperforms competitors from xAI, Anthropic, and Meta, as well as high-performing Chinese models from DeepSeek and Alibaba, across various categories. AI Mode rolled out to all US users, with new features debuting there before being integrated with AI Overviews. Both AI tools utilize a custom version of Gemini 2.5, which includes personal context from Gmail for tailored results, marking a shift towards personalized search experiences. Furthermore, Gemini 2.5 Pro and Flash now include thought summaries in the Gemini API and Vertex AI, organizing the model’s thoughts into clear formats with headers and details on the model’s actions.

The evolution from I/O 2024 to 2025 shows Google’s strategic maturation in AI development. At I/O 2024, Google announced improved AI models, Gemini 1.5 Pro and Trillium TPU, along with image and video generation models, Imagen 3 and Veo. The 2025 iteration was built directly on these foundations with Imagen 4 and Veo 3, showing iterative rather than revolutionary progress. Google I/O 2024's keynote was completely dominated by Google Gemini and AI tools, but 2025 marked a shift from experimental showcases to production-ready implementations. While 2024 promised AI Overviews would reach a billion people by year's end, 2025 demonstrated that the promise was fulfilled with widespread deployment and enhanced capabilities. The hardware trajectory tells an equally important story. The Pixel Fold was a star of 2023's I/O, with 2024 continuing hardware emphasis, but 2025's complete absence of hardware announcements signals Google's recognition of where its competitive advantages truly lie.

Image sourced from MIT Technology Review

Announcements and products at Google I/O 2025 reflected Google’s focus on making AI more useful and accessible, weaving AI into everyday experiences with projects like Gemini 2.5 and Astra, making it a part of infrastructure. The conference showed several trends that align with Google’s AI strategy. AI is becoming increasingly intelligent and autonomous, transitioning from merely answering questions to actively taking action and completing tasks independently. We are witnessing this growth with systems like Gemini 2.5, which can think through problems and use different tools more effectively. At the same time, AI is becoming more personal by connecting to our email and other accounts for relevant data to provide tailored responses that actually relate to our specific situation and needs. Companies like Google are making this a reality by offering basic AI features to everyone while charging for the more advanced creative tools, which started rolling out to their premium subscribers right away. This shows how AI is becoming both more capable and more tailored to individual users. Additionally, the development of agentic AI shows a shift towards proactive systems that can execute complex tasks autonomously, as shown with Gemini 2.5. Furthermore, the trend of personalized AI experiences was evident with the integration of personal data from Gmail and other Google services. Most creative tools became available for paid users in the US, suggesting Google’s strategy to monetize advanced AI capabilities while making basic features broadly accessible.

Despite the impressive announcements, Google I/O 2025 highlighted several areas of concern and criticism. The integration of personal Gmail context into AI search raises questions about data privacy and the extent of Google's information access. This integration of personal context may further Google's data dependency while exposing sensitive personal information to AI, which is susceptible to errors. Additionally, the approach with Pro and Ultra subscribers creates accessibility barriers, limiting Advanced AI capabilities to users who can afford premium pricing, contrasting with competitors like DeepSeek and other foreign companies that offer more democratized access to AI tools. This aggressive AI integration has led to a decline in traditional search quality since users have shown frustration over AI-generated responses replacing direct access to source material.

The conference showcased Google’s “aggressive approach to AI”, as Dr. Żurański described, with AI now affecting “almost all of Google’s products.” This transformation is particularly evident in Google Search, where AI Overviews have become increasingly common, fundamentally changing how users interact with information. Despite its hallucinations and initial misgivings, it is bound to dramatically change Dr. Żurański and his team’s work.

Dr. Żurański’s insights from Google’s advertising team reveal the strategic challenges driving Google’s AI push. Working on a team that develops strategies for Google’s ad products, which account for approximately 80% of the company’s revenue, he explained that Google realized that the “ad business is shrinking relative to competitors, like Facebook or TikTok.” The shift to AI-powered search presents a fundamental challenge. “In the past people used to actually visit websites, and that was one of the main ways where ads were presented,” Dr. Żurański explained. It is becoming increasingly popular to use AI overviews, meaning that AI is executing the searching and gathering process, causing users to visit websites less often. The consequences of this  transformation force Google to balance monetization with user experience, seeking ways to “monetize content without being overly aggressive” compared to platforms that are “full of ads.”

Among the technologies presented, Dr. Żurański highlighted a real-time translational tool that exemplifies Google’s focus on practical AI applications. The tool was designed for video calls with foreign language speakers, not only translating words but also analyzing whether the speaker is “agitated or speaking fast or slow” and capturing their “emotional state” to influence the translated voice and convey the same emotion. Such a skill represents a significant advancement over traditional translation services, suggesting Google’s AI development is moving beyond basic functionality toward more sophisticated human-computer interaction.

Moreover, Dr. Żurański acknowledged a growing concern of Google having a “very aggressive approach to AI” , with people reporting that “Google search quality is going down compared to the original Google search.” Stemming from users expecting Google to “answer longer and more complex questions,” they are driving changes in how queries are understood and processed. The underlying content has also shifted, as Dr. Żurański noted, “we are past the era of personal blog posts where there was some pretty interesting content”, suggesting that the internet’s information landscape has also fundamentally changed since Google’s early days. However, AI overviews do provide an interesting solution to this, since although they initially hallucinated, they have become increasingly common with technological improvements. The experimental AI search mode now executes many search queries, reads the information, and provides a summary of the information it has read.

While addressing concerns about AI-generated misinformation, Dr. Żurański offered a pragmatic perspective, asserting that the internet was “already full of misinformation,” and although AI does hallucinate, people will eventually “become used to those errors.” He argues that while AI may initially seem unreliable, “in a lot of cases these tools are just too good to not use.” His comment reflects the broader industry belief that AI’s benefits will outweigh the risks, but also shows the challenges of segregating what is true from what is false in an AI-driven system.

Furthermore, Dr. Żurański predicts a shift away from algorithmically provided content toward “content that’s generated instantly with the use of AI,” with trends driven by the younger generation. This aligns with Google’s broader strategy of making AI generation tools more accessible and immediate. With only about 1000 data scientists among Google’s approximately 80,000 employees, Dr. Żurański represents a specialized community within the company. These professionals, often with backgrounds in statistics and natural sciences, focus on determining what and how to measure, tackling the often complex challenge of distinguishing between correlation and causation in vast datasets.

For aspiring technologists, Dr. Żurański emphasized curiosity as the most valuable skill. “Curiosity to understand how complex systems work can be applied to anything,” is especially valuable in his opinion, because “a lot of systems are complex” and “there isn’t a single person who can understand” every aspect completely.

Google I/O 2025 revealed a company in transition, leveraging its AI capabilities to maintain dominance while adapting to fundamental changes in user behavior and competitive pressures. The newfound focus on practical AI applications reflects exactly that. However, there exist several challenges, including concerns about information quality, intensifying competition, and tension between AI-powered search and traditional web monetization. As Dr. Żurański’s insights reveal, Google’s path forward requires balancing AI development, user trust, business needs, practical utility, innovation, and accountability. The 2025 conference remains crucial for its recognition that the future of search, advertising, and information access is being rewritten by the very AI technologies Google helped pioneer.

Citations

McHugh-Johnson, Molly (2025). 100 things we announced at I/O, Google Keyword. https://blog.google/technology/ai/google-io-2025-all-our-announcements/

Romannoff, Dan (2025). Is AI Investment Poised for Growth? Top Picks and Promising Applications for 2025, MorningStar. https://www.morningstar.com/stocks/is-ai-investment-poised-growth-top-picks-promising-applications-2025

Maslej, Nestor (2025). AI Index 2025: State of AI in 10 Charts, Stanford University Human-Centered Artificial Intelligence (HAI). https://hai.stanford.edu/news/ai-index-2025-state-of-ai-in-10-charts

Heikkilä, M. (2024, May 14). What to expect at Google I/O. Retrieved September 19, 2025, from MIT Technology Review website: https://www.technologyreview.com/2024/05/14/1092375/what-to-expect-at-google-i-o/

‌