Google CEO Sundar Pichai says AI search will actually help the web
20 May 2024 (6 months ago)
Language vs intelligence (22s)
- Language encodes a lot of intelligence, but human consumption of information involves more than just language.
- Large language models (LLMs) have led to an explosion of interest, innovation, and investment in AI, but the intelligence of these models may not be increasing at the same rate as their facility with language.
- Computers are getting better at language but may be getting "dumber" in some cases.
- Google's Gemini AI model is natively multimodel, incorporating audio, video, text, images, and code.
- Multimodality will allow for a more comprehensive understanding of information and improve search results.
- Google's approach to AI is "bold but responsible."
- There is a growing concern about AI systems providing incorrect or misleading information.
- An example is given of a multimodal video search that provided incorrect instructions for fixing a broken film camera.
- Google consults with subject matter experts to review answers, but there can still be nuances and context that are missed.
- Search should provide more context and allow users to explore answers deeply.
- Multimodality can provide more opportunities for improving the accuracy and understanding of AI systems.
Future of Google Search (4m6s)
- Sundar Pichai discusses the future of Google search and the introduction of AI previews.
- Some website creators have expressed concerns about the impact of AI on their traffic.
- Pichai acknowledges the negative reactions from website creators, comparing it to the concerns about the web's demise during the transition from desktop to mobile.
- He remains optimistic, citing empirical evidence that human curiosity is boundless and that AI previews have been met with positive user responses.
- Pichai emphasizes that AI previews actually help users understand content better and engage with it more, leading to higher click-through rates for content and links within AI previews.
- He understands the concerns of website creators, as AI represents a significant platform shift, but he is engaging with them to address their perspectives.
"Google Zero" & website decline (7m31s)
- Google is the last large-scale referrer of traffic on the web for almost every website.
- Independent sites that aren't part of a large publishing conglomerate have seen their Google traffic drop to zero, leading to concerns about the survival of these sites.
- Google is aware of these concerns and is trying to understand the impact of its changes on individual websites.
- Google has historically provided more traffic to the ecosystem and driven growth.
- There are times when Google has made changes to send more traffic to smaller sites, but some of the sites that complain about decreased traffic are aggregators in the middle.
- AI overviews provide context and expose users to various branching-off points, leading to more engagement and growth over time.
- Google remains one of the largest referrers due to its deep care for the web.
- Google prioritizes approaches that send more traffic while meeting user expectations.
- Google is constantly trying to improve its understanding of high-quality content and values independent sources, smaller things, and authentic voices.
- Google relies on the value of the ecosystem and is committed to doing the right thing at an aggregate level.
- Google is considering breaking out clicks from featured snippets, AI previews, and regular Google 10 Blue Links in Search Console to provide more visibility.
- Google wants to avoid people designing content specifically for AI previews and seeks a balance between providing visibility and encouraging the creation of good content.
OpenAI's Sora & fair use training (14m15s)
- Google's search results have evolved beyond the traditional 10 Blue Links model.
- The AI era brings uncertainty about fair use arguments and data payments for training AI search results.
- Google has made licensing deals and believes in fair use for beneficial transformative use.
- There are cases where dedicated incremental value to models may lead to partnerships.
- Sundar Pichai expresses uncertainty about the details of OpenAI's Sora training on YouTube.
- Google's YouTube team is following up to understand the situation.
- Google has terms and conditions that users are expected to abide by.
- Pichai suggests that if OpenAI violated Google's terms and conditions, it would not have been appropriate.
Compensating creators for content (17m9s)
- Google understands the emotional reaction from the creative community regarding AI search potentially feeling like a 'taking' of their work without proper compensation.
- Google emphasizes the importance of supporting the creator community and is actively working on finding ways to bring value and incentives back to small creators and independent businesses.
- Google believes that players who do better in supporting content creators will emerge as winners in the long term.
- YouTube's relationship with the music industry is highlighted as an example of a successful licensing business model, where both sides benefit from the partnership.
- Google's approach to search generative experiences and AI previews on the web is contrasted with OpenAI's approach, with Google emphasizing its careful testing, iteration, and prioritization of approaches.
- Google stresses that it deeply cares about listening to feedback and balancing the needs of different stakeholders in the ecosystem.
The rise of AI spam (21m45s)
- People will exploit search engines by generating AI-written content and using it to steal traffic from competitors.
- The internet will become increasingly flooded with synthetic content created by AI.
- Users may prefer human-created content over AI-generated content.
- Google's search quality teams have been working to improve their ranking systems to better identify high-quality content.
- Google believes that users will value human-created experiences and is committed to ensuring that AI-generated content does not replace human-created content.
- Google is trying to strike a balance between promoting responsible use of AI and listening to user feedback on what they consider high-quality content.
- Google is confident that it will be able to approach the challenge of AI-generated content better than others.
Testing AI Overviews in Google Search (25m2s)
- Google CEO Sundar Pichai discussed the current state and future of AI-generated search results.
- Pichai demonstrated the need for AI-generated search results to respect user intent and provide a variety of options rather than pushing AI-generated overviews.
- He showcased an example of an AI-generated overview for the search term "JetBlue Mint Lounge SFO," highlighting its value and convenience.
- Pichai acknowledged that AI-generated search results may not always be perfect but emphasized the positive user response due to the summaries' value and convenience.
- He believes that AI-powered search will enhance the web by providing valuable and enjoyable experiences to users.
Multimodal search & competing with OpenAI (30m31s)
- Sundar Pichai acknowledges the competition from OpenAI's multimodal search demo, which resembles Google's own demos.
- He compares the current situation to the earlier competition between voice assistants like Google Assistant, Alexa, and Siri.
- Pichai expresses excitement about the rapid technological advancements and the potential to significantly improve user experiences.
- He believes that the underlying technology has advanced to a point where it can support a powerful assistant, leading to progress in this area.
- Project Astra is mentioned as an example of the magical and exciting possibilities of multimodal search.
- Google's mission is to organize the world's information and make it universally accessible and useful.
- AI is seen as a transformative technology that can help Google achieve its mission more effectively.
- Pichai emphasizes the importance of responsible AI development and the need to address concerns about bias, privacy, and security.
- Pichai believes that AI-powered search will actually help the web by making it easier for users to find relevant and accurate information.
- AI can provide more personalized and contextual search results, taking into account the user's location, preferences, and past search history.
- AI can also help users explore new topics and discover connections between different pieces of information.
- Pichai highlights the potential of AI to improve the overall search experience and make it more intuitive and user-friendly.
Solving AI hallucinations (32m47s)
- AI hallucination is still an unsolved problem.
- Hallucination is what makes language models creative but it can lead to incorrect factual information.
- Search can mitigate hallucination by grounding language models with context and knowledge.
- Google Lens is an example of how AI recognition and understanding has improved over time.
- DeepMind and Google Brain are working on solutions to the hallucination problem and making progress.
- Language models are one aspect of AI and Google is working broadly to advance AI.
What the web looks like in 5 years (35m57s)
- Sundar Pichai hopes the web will be:
- Richer in terms of modality.
- More interactive.
- More stateful.
- AI can help create a richer web by dynamically composing UIs.
- There will be entrepreneurs who figure out extraordinary ways to use AI to improve the web.
- Google creates a lot of incentives for development on the web through search, Chrome, etc.
- Pichai wants to make sure those incentives are aligned with the goal of creating a richer, more interactive, and more stateful web.
- Pichai acknowledges that not everything is in Google's control, but he wants to influence the web in a positive way.
- One of the biggest challenges is how to reward originality, creativity, and independent voice in the content ecosystem.
- Pichai believes this is an important principle for the web and for Google as a company.