Over the past year, Elizabeth Reid has taken the helm of Google's ambitious initiative to revolutionize its core search product. Nearly a year ago, her team unveiled one of the company's most significant updates in recent history with the introduction of AI Overviews. This feature utilizes generative artificial intelligence models to summarize search results, making it easier for users to digest information quickly.

The initial rollout of AI Overviews was met with mixed reactions. Some users found themselves amused by the AI's absurd suggestions, which included bizarre advice such as recommending that people could eat rocks or glue cheese onto pizza. However, Reid assures that Google has since worked diligently to enhance the accuracy and usefulness of these AI summaries. The results have been transformative, as users are increasingly adapting their online information-seeking behaviors to take advantage of the new capabilities.

In a recent conversation with Melissa Heikkil, the AI correspondent for the Financial Times, Reid delved into the future of AI-enhanced search and discussed the broader implications for the internet's business model.

MH: You graduated from Dartmouth College, where the concept of AI was first articulated in 1956. How did your journey to AI begin? Did your time at Dartmouth shape your path?

ER: Attending Dartmouth was indeed pivotal for my entry into computer science. My high school experience was quite limited when it came to technology education, focusing mainly on typing and basic software skills like Microsoft Excel and Word. My first taste of programming came from a project involving my graphing calculator, which was a prerequisite for a class I wanted to take.

I started Dartmouth intending to pursue physics because I excelled in mathematics. During my freshman summer, I interned in material science, which I found theoretically intriguing but ultimately desired something more practical. This led me to consider engineering physics.

While taking thermodynamics and physics courses, I also enrolled in a computer science class, and I discovered a passion for it. For guidance, I approached Professor Thomas Cormen, a respected faculty member. He encouraged me to pivot to computer science based on my interests and aptitude.

As I sought employment in 2003, Dartmouths department provided solid training, though it did not carry the same prestige as schools like Stanford or MIT. Professor Cormen connected me with a former student at Google, which ultimately helped me secure an interview and, eventually, a position at Googles New York office. At that time, there were only about 10 engineers in the office and around 500 to 1,000 employees in the Mountain View headquarters.

Throughout my time at Google, I've witnessed the insatiable nature of human curiosity. People have endless questions that they seek answers to, reflecting a constant quest for knowledge.

I initially worked on local search, which later transitioned into geospatial mapping projects. My experience underscores that while generative AI is often conflated with AI in general, the latter encompasses a broader spectrum of applications. Over the years, I have utilized AI across various sectors, including local search and mapping technologies.

After transitioning to Google Search, I found myself in discussions with engineers about the evolving capabilities of our technology. It was an exhilarating time as we reached a turning point in what we could achieve.

MH: Now that its been about a year since AI Overviews launched, how has the experience been?

ER: The launch has been immensely successful. We are witnessing remarkable growth in search usage, with users increasingly issuing more queries. AI Overviews help users frame questions that were previously challenging to articulate, allowing them to access information that is often dispersed across the web.

Human curiosity knows no boundswhether its a toddler asking why repeatedly or adults hesitating to inquire due to perceived barriers. By lowering the threshold for asking questions, we've seen a surge in user engagement, with more individuals feeling empowered to seek answers to their inquiries.

Moreover, we've noted that users are not just asking more questions; they are also posing longer, more nuanced queries. AI Overviews facilitate this trend by allowing individuals to express their thoughts without having to conform to strict keyword formats.

This shift is particularly evident among younger users, who are often more willing to challenge expectations and embrace new technologies. We are witnessing a trend towards multi-modality in searches, where people are combining text and images in their queries to receive more comprehensive responses.

MH: With the emergence of tools like ChatGPT, behaviors in search are clearly evolving. Is Google considering adapting to more conversational search functions?

ER: Our approach differs from that of chatbots, which often create a more personified interaction. Rather than treating search as a conversation with a friend, we view it primarily as a means of obtaining information. We are experimenting with how to address questions that have multiple components, allowing users to follow up without starting from scratch.

Our focus remains on organizing information efficiently rather than simulating a human-like interaction.

MH: There has been substantial criticism regarding search efficacy, with many users feeling the need to include terms like 'Reddit' in their searches to yield satisfactory results. How is Google addressing concerns about inaccuracies and misleading AI responses?

ER: Including 'Reddit' in a search is not necessarily a negative trend. Many users want diverse perspectives, and the ability to express specific needs can be seen as a positive development. We recognize that younger users prefer firsthand accounts over traditional sources.

While the generative AI technology has its limitations and can sometimes produce erroneous suggestions, our team takes such feedback seriously. For example, the 'eating rocks' incident was a rare outlier, but it highlighted the need for continuous improvements in our models.

We've made significant strides in prioritizing factual accuracy, a crucial element of our search product. Unlike chatbots, which may prioritize conversational flow, our search models are designed with factual integrity at the forefront.

Addressing the technical flaws of language models, we strive to improve security and accuracy. Our search function fundamentally relies on high-quality information sourced from the web, which helps minimize the chances of errors.

AI Overviews are not intended to serve as standalone features; they are designed to guide users while encouraging them to explore further. We carefully consider when to deploy AI Overviews, ensuring that the responses generated meet high-quality standards.

MH: Are you observing changes in how users verify information sources? How often do they rely on AI Overviews?

ER: We do see users engaging more deeply with the content, often driven by a desire for confirmation or a spark of curiosity that leads to additional questions. This engagement indicates a higher quality of clicks, as users are less likely to abandon a webpage after realizing it doesnt meet their needs.

Interestingly, longer and more complex questions lead to a broader diversity of websites in search results. Users seeking thorough answers from generative AI are often redirected to pages that provide in-depth discussions rather than surface-level content.

MH: Some critics have pointed out subtle inaccuracies in language models that may go unnoticed by non-experts. How concerned are you about this?

ER: We place a high importance on maintaining quality, particularly for sensitive topics related to finance and health, which we refer to as your money or your life. In these areas, we take extra care to ensure our responses are appropriate and often suggest consulting a professional.

However, we also recognize that not everyone has immediate access to medical professionals or experts. Thus, we strive to strike a balance, providing essential guidance while encouraging users to seek further information when necessary.

MH: Addressing publishers' concerns, how is Google planning to prevent traffic and revenue losses due to AI-driven search?

ER: We believe that user interactions with genuine voices are fundamental to our search product. Many users seek quick answers, but they also desire insights from trusted sources. We continually explore how to present content effectively while integrating links that elevate reputable publications.

By reducing barriers to information, AI Overviews open new opportunities for websites and content creators, leading to higher-quality user engagement and clicks.

MH: Might we see a paid version of Google Search in the future?

ER: While nothing is off the table, we are committed to ensuring that search remains accessible and free for all users. There may be subscription-based features in the future, but the core search functionality will always be available at no cost.

MH: Looking ahead, what does the future of search hold? Are there additional modalities or agents on the horizon?

ER: Our primary goal is to make search as effortless as possible, which means embracing multi-modalities. People naturally express their needs in various ways, whether through text, voice, or visual inputs.

Over time, we expect search will become increasingly personalized, adapting to individual learning preferences, whether through video or text formats.

The challenge lies in ensuring that our technology meets users where they are, making exploration and learning as seamless as possible.

MH: Have your own search habits changed in this age of AI?

ER: Absolutely! For example, I find myself asking questions more frequently, especially when discussing topics outside my expertiselike cricket. Instead of feeling overwhelmed by unfamiliar terms, I simply pose questions to gain a better understanding. This shift highlights how AI can enhance our everyday interactions and learning experiences.

This conversation has been edited for brevity and clarity.