Search engines based on Generative AI, known as Generative Search Engines (“GSE”), provide actual answers to user queries, not just website links. These links suggest dozens, hundreds, if not thousands of websites where answers to your questions might be found or might not. GSE, once perfected, is going to change that and revolutionize the way people search online to find information. With GSE, you simply ask questions and explain what you want to know. Then in just seconds GSE searches for you, studies the web, and provides reliable real-time answers in a conversational fashion.

Finally, you will not have to dream up just the right keywords to find the information you want. Finally, you will not have to wade through advertisements and SEO inflated website rankings. You will just talk to an AI, chat with it, just like talking to a human, and it will tell you what you want to know. It will do so accurately and honestly without financial motivation to sell you products or steal your privacy.

Goodbye Google Keyword Search

Traditional search engines like Google rely on keyword-based queries that are juiced-up for advertising profit. Keyword search is a very old search method that I used in the eighties. It usually fails when searching large volumes of ESI. Ask any lawyer who has ever done e-discovery search in the days before Da Silva Moore. See Austin The Da Silva Moore Case Ten Years Later; (EDRM 2/23/22). All e-discovery experts know that keyword search leads to a flood of irrelevant information, whereas predictive coding is far more precise. GSE is like that. It goes beyond keywords and understands natural language, context, and concepts. These abilities, much like advanced search methods in e-discovery, enable GSE to deliver more accurate and relevant answers to search questions. Plus, a GSE will give you a well written (or spoken) response to your question. Upon request the Generative Search Engine should also be able to support its response with citations and yes, even links. Verification is always a good idea with generative AI where errors and hallucinations are still possible.

GSE is poised to revolutionize the way people access information online. Google acknowledges this potential, stating that GSE “can unlock entirely new types of questions you never thought Search could answer and transform the way information is organized to help you sort through and make sense of what’s out there.” In May 2024, Google began to integrate many Generative Search Engine features into its standard search. Previously, these features were available only to some users under the name Search Generative Experience, or SGE—confusing, right? Users had to opt-in through Google Labs. However, in May 2024, Google dropped the confusing name, made the features the default for everyone in the U.S., and rebranded it as AI Overview. Here’s how it looks on your browser when I asked this question:

Give it a try and see for yourself. I’ve been using it since Google Labs first released it for opt-in beta testing in May 2023. While it has improved, it still has a long way to go. I’ll go into more detail on later in this article.

Hello Advanced Online Search Using GSE

A perfected GSE will seamlessly combine the traditional search engine capabilities with generative AI. It will retrieve existing information from the web, analyze and select the highest probability correct answers and generate conversational responses. The primary new capacities of GSE are:

• Conversational Search: Engages in natural, ongoing dialogue with users.
• Answer Synthesis: Combines information from multiple sources into a single, coherent response.
• Contextual Memory: Remembers a user’s previous queries for better contextual understanding.
• Multi-modal Search: Handles and generates text, images, videos, and other formats.
• Real-time Information: Provides up-to-date information by synthesizing data from live sources.

Many new search possibilities arise from these new capacities to enhance the way the users retrieve information. The ones that stick out to me as likely transformative are:

1. Personalized Query Responses: GSEs can tailor search results to individual preferences, offering more relevant and personalized answers based on past interactions and specific user needs. For instance, it could emphasize law related issues.
2. Deep-Dive Research: Users can conduct complex research within a single search session, as GSEs can follow up on queries, refine responses, and provide in-depth analyses by synthesizing information from various sources.
3. Task Automation: Beyond just retrieving information, GSEs can perform tasks like generating code, drafting documents, or creating detailed summaries, making them useful for professional and technical applications.
4. Dynamic Content Creation: GSEs can generate new content on demand, such as personalized reports, creative writing, or custom visuals, directly from a user’s search query.
5. Interactive Learning: With the ability to engage in back-and-forth conversations, GSEs can be used for interactive learning, helping users understand complex topics through step-by-step explanations or simulations. The GSE can also suggested related topics, alternative queries or additional resources that users might not have considered

These new possibilities arise from the capacities of GSE, but new challenges will arise too, including the reliability and accuracy of GSE language. It will make errors, and it will to some degree be tainted by biases in the training data. It will also be prone to sycophantism and even hallucinations. I have written about those known dangers of AI many times before. See e.g. OMNI Version – ChatGPT4o – Retest of the Panel of AI Experts – Part Three (5/29/24) (expert views on AI hallucination); Worrying About Sycophantism (7/9/24) (How to protect yourself from AI’s tendency to flatter).

These issues can all be managed by skilled users, but this means the hybrid human-machine approach will still be required. From Centaurs To Cyborgs: Our evolving relationship with generative AI (4/24/24); Predictive Coding 4.0 – Nine Key Points of Legal Document Review and an Updated Statement of Our Workflow – Part Three (9/26/16). Human common sense will still be needed, especially in the early days. As will be shown later in this article Google’s AI Overview has already given some pretty ridiculous answers.

ChatGPT4o-Omni Explains Six Key Features of GSE

The Six Key Features of a Generative AI Search Engines are, to quote ChatGPT4 omni:

1. Natural Language Understanding: These search engines understand complex, natural language queries, allowing users to ask questions in a conversational tone rather than relying on specific keywords.

2. Content Generation: Unlike traditional search engines that only return links or snippets from existing web pages, generative AI search engines can create original content based on the input. This is useful for summarizing information, answering questions directly, or generating content like emails or articles.

3. Contextual Awareness: The AI can maintain context over a conversation, allowing for follow-up questions and more detailed exploration of a topic.

4. Personalization: These engines can tailor responses based on the user’s preferences, history, and specific needs, leading to more relevant and customized results.

5. Integration with Other Tools: Generative AI search engines often integrate with other AI tools and services, allowing for tasks like code generation, document creation, or data analysis within the search environment.

6. Ethical and Safe Responses: Advanced AI search engines are designed to generate content responsibly, avoiding harmful or biased outputs, and respecting user privacy and data security.

Goodbye Google Monopoly

This expected dramatic improvement in search likely means that GSE will eventually replace traditional search engines. For all practical purposes, as of 2024 traditional search engines means just one company, Google.

Ironically, this market shift may be helped along by an antitrust case started by the DOJ in 2020, US v. Google. After a nine-week bench trial a 286-page final judgment was issued on August 5, 2024 by U.S. District Judge Amit Mehta. United States v. Google LLC, (E.D. Va. 8/5/24). Judge Mehta held that Google search is an illegal monopoly. Google loses massive antitrust case over its search dominance (NPR, 8/5/24) (Download the Opinion by Judge Mehta).

This is a beautiful monster of an opinion, a real work of legal art, including wonderful language, 38 redactions, color charts, and impeccable reasoning. It even includes a very interesting discussion of Rule 37(e) sanctions for preservation omissions. United States v. Google LLC, supra at 272-276. Judge Mehta wisely decided not to muddy the appeal and denied the government’s request for sanctions against Google for failure to preserve employee chat messages. But he also warned Google that if they do this again they “may not be so lucky in the next one.” Id. at 276.

Here are the first three paragraphs of this landmark opinion:

The general search engine has revolutionized how we live. Information that once took hours or days to acquire can now be found in an instant on the internet with the help of a general search engine. General search engines use powerful algorithms to create what seems like magic. Enter a search query, and the general search engine will retrieve, rank, and display the websites that provide the exact information the user seeks at that very moment. And it all happens in the blink of an eye.

General search engines make money by selling digital advertisements. Type the words “running shoes” into a general search engine, and sellers of running shoes will compete with one another in a split-second auction to place an advertisement on the results page, which if clicked takes the user directly to the seller’s website. This is a highly effective way of reaching consumers. It is also an incredibly lucrative business. In 2021, advertisers spent more than $150 billion to reach users of general search engines.

For more than 15 years, one general search engine has stood above the rest: Google. The brand is synonymous with search. Once a scrappy start-up founded by two Stanford University students in a rented garage, Google is now one of the world’s most valuable companies. Its parent company, Alphabet Inc., today has a market capitalization (the value of its outstanding shares of stock) of more than $2 trillion. Much of that value is due to Google’s extremely profitable advertising business.

The appeal and enforcement will likely take years. In the meantime the holding may become somewhat moot by post-appeal facts on the ground. Namely the rapid advancement of GSE, meaning generative search engine of course, not Google search engines. Ironically this newly emerging AI tool was beyond everyone’s imagination when the suit was filed in 2020. The attorneys and judge used the acronym GSE to stand for “Google Search Engine.” No doubt the change of market conditions since the trial will be one of Google’s many arguments going forward.

Still, no matter what the appellate court may do with this decision, the days of Google’s dominant search are numbered—either by innovation or by the courts. Probably both forces will be involved.

Hello OpenAI’s New GSE product: Search GPT

Just before Judge Mehta’s landmark antitrust ruling, OpenAI announced its imminent release of a new GSE product: SearchGPT. Right now it is in a prototype version only. I am one of many still on the private waiting list to try it out. But according to the reviewers at TechTarget:

SearchGPT will offer users many features to extend the capabilities of ChatGPT. Key features include the following:

• Conversational AI. As SearchGPT uses the same LLM as ChatGPT, the service provides a natural language interface for asking questions.
• Follow-up questions. The service will let users ask follow-up questions, making the experience more interactive and similar to a human conversation.
• Up-to-date responses. There is no longer an arbitrary knowledge cutoff date. SearchGPT can ingest real-time information from the web to provide users with up-to-date information.
• Summarization. In contrast with a traditional search engine that just provides a series of links to web addresses where more information can be found, SearchGPT provides concise summaries of the information.
• Attribution. OpenAI is including clear attributions and source links enabling users to verify the information and explore further details on the original websites.
• Visual results. A feature of SearchGPT is the delivery of visual results in the form of pictures and video to answer user queries.

Sean Kerner, SearchGPT explained: Details about OpenAI’s search engine (TechTarget, 8/14/24).

The author of this well done TechTarget article, Sean Michael Kerner, includes a chart comparing the GSEs available today or coming soon.

Many issues arise from this paradigm shift, most based on the reliability and accuracy of Large Language Model based Gen A. As mentioned, it can be prone to errors, biases, sycophantism and even hallucinations. I have written about those known dangers of generative AI many times before. See e.g. OMNI Version – ChatGPT4o – Retest of the Panel of AI Experts – Part Three (5/29/24). It remains to be seen how well the GSE software manufacturers can reduce these problems with careful design, guardrails and prompt wrappers.

Hello Microsoft CoPilot, Perplexity AI and the New GSE Version of Google, AI Overview

Joining the new battle for GSE software is Google’s own AI Overview, Microsoft’s Copilot, and a few newbie tech companies like Perplexity AI. Google’s AI Overview is still a work in progress and so is Copilot, although, so far at least, it is better than AI Overview. I have not used OpenAI’s GPT search, so have no opinion, but based on my positive experiences using ChatGPT, and the favorable review by TechTarget, I expect good things. It may cause me to switch from traditional Google search as my primary search tool, but I will still keep it for verification purposes.

Google’s efforts with AI Overview got off to rocky start, a full-on public relations nightmare. That is primarily because it recommended that users eat glue and rocks. Nico Grant, Google’s A.I. Search Errors Cause a Furor Online (N.Y.T. 5/24/24) reported that AI Overview:

… instructed some users to mix nontoxic glue into their pizza sauce to prevent the cheese from sliding off, a fake recipe it seemed to borrow from an 11-year-old Reddit post meant to be a joke. The A.I. told other users to ingest at least one rock a day for vitamins and minerals — advice that originated in a satirical post from The Onion.

People also shared examples of Google’s telling users in bold font to clean their washing machines using “chlorine bleach and white vinegar,” a mixture that when combined can create harmful chlorine gas. In a smaller font, it told users to clean with one, then the other.

. . . struggle with presidential history, saying that 17 presidents were white and that Barack Obama was the first Muslim president. It also said Andrew Jackson graduated from college in 2005.

I asked Copilot, Microsoft’s GSE powered by ChatGPT, the following admittedly leading question: “How is Bing better than Google search.” Here is Copilot’s response:

Microsoft Copilot and Google’s Gemini (formerly known as GSE) are both powerful AI assistants, but they have some key differences that might make Copilot a better choice for certain users:

1. Integration with Microsoft 365: Copilot is deeply integrated with Microsoft 365 apps like Word, Excel, PowerPoint, and Outlook. This means it can help with tasks such as drafting emails, creating presentations, and analyzing data directly within these applications¹. If you already use Microsoft 365, Copilot can significantly enhance your productivity.
2. AI Quality and Responsiveness: Copilot is powered by OpenAI’s GPT-4, which is known for its high-quality output and responsiveness to feedback¹. This makes it particularly effective for generating accurate and contextually relevant content.
3. Comprehensive Features: Copilot offers a wide range of features that drive efficiency, such as summarizing documents, generating reports, and providing insights based on your data². These features are designed to integrate seamlessly across the Microsoft ecosystem.
4. User Experience: While both AI assistants have user-friendly interfaces, some users find Copilot’s integration within Windows 11 and Microsoft 365 to be more intuitive and less cluttered³.
5. Security and Compliance: Microsoft places a strong emphasis on security and compliance, which can be crucial for businesses handling sensitive information⁴. Copilot benefits from these robust security measures, making it a reliable choice for enterprise use.
6. Ongoing Development: Microsoft continuously updates Copilot with new features and improvements, ensuring it stays at the cutting edge of AI technology¹.

In contrast, Google’s Gemini is still in the process of fully integrating into Google Workspace, and while it excels in creative writing and editing tasks, it may require more frequent fact-checking⁵.

Have you had a chance to try out either of these AI assistants? What features are most important to you?

Pretty good, although I am a little dubious about point five on Microsoft security. See e.g. Ravie Lakshmanan, Microsoft Fixes ASCII Smuggling Flaw That Enabled Data Theft from Microsoft 365 Copilot (The Hacker News, 8/27/24).

Plus, the whole answer sounded a bit too much like marketing, although diplomatic. Responses should be totally objective, but I may have prompted this tone by my leading question. The effort so far of combining Microsoft’s previous so-so search tool, Bing, with AI from OpenAI, seems to be doing well. I never used Bing before and will now occasionally use Copilot. By the way, when I ask Copilot “Are you better than Bing?” it replies:

That’s an interesting question! Bing and I actually work together. Bing helps me find up-to-date information from the web, while I focus on having engaging and meaningful conversations with you. So, it’s not really about being better, but about how we complement each other to provide you with the best experience possible. 😊

I also looked into Perplexity AI, which I had never heard of before the TechTarget review. I created an account at Perplexity.AI in ten seconds, ironically using my Google account. It then took me to a screen that told me:

What makes Perplexity different: Answers that are accurate and always cited. We continuously search the internet and identify the best sources, from academic research to Reddit threads, to provide the perfect answer to any question.

Then I was taken to a screen that offered a Pro Version for $20 per month, or continue for free. I chose free, which I later learned is only powered by old ChatGPT3.5, plus its own natural language processing (NLP) software. If I had paid for the Pro Version the Perplexity software would supposedly be powered by GPT-4, Claude 2, and others.

I did not investigate what that meant. Instead, I asked free Perplexity to explain how it was different from Copilot search. It began by explaining Copilot’s Purpose, Functionality and Integration in three short bullet points. A fourth bullet point was Copilot’s Limitations, which stated: “Copilot does not retain chat history and does not support third-party plugins when commercial data protection is enabled.” Not much of a limitation to me; sounds more like a security protocol. Perplexity provided a link to support this Limitation to a Microsoft webpage, Overview of Microsoft Copilot, which, strangely enough, was partially in Thai. Click it and see. Some of the other links were to webpages partially in the Thai language. Very perplexing.

Here is rest of the Perplexity response where it explains what it is.

Perplexity AI

• Purpose: Perplexity AI is designed as a search assistant to provide concise and accurate answers to user queries by drawing from search results. It focuses on delivering precise information rather than assisting with tasks like content creation or coding.
• Functionality: Perplexity AI is primarily used for searching and providing information based on user queries. It is not integrated into specific applications like Microsoft Copilot but operates independently to deliver search-based answers.
• Integration: Unlike Copilot, Perplexity AI is not tied to any specific platform or application suite. It functions as a standalone search assistant.

In summary, while both Copilot and Perplexity AI utilize AI technology, Copilot is more focused on assisting with tasks within Microsoft applications, whereas Perplexity AI is centered around providing search-based information.

Seems to me like perplexity is trying to make its own limitations seem like strengths. Color me unimpressed. But who knows, in ten years a small unknown company like this may be the next Google of generative AI search. Let’s hope not. And I say that not because I have anything against Perplexity. My objection is to monopolies.

Conclusion

The days of one company stifling innovation by market dominance should be over. There should instead be a half a dozen or more GSE search platforms to choose from. All of them good but offering slightly different features and strengths. Monopolies are not good for innovation or free market economies. Creative inventiveness and technological progress instead thrive on competition. I also hope that the entire pretend-free internet marketing economy will fade away. It is inherently dishonest. I would rather pay with currency for a good product instead of pay with my soul, my privacy. The days of advertising scams and rigged markets should come to an end. I for one am hopeful that the DOJ and prosecutors will continue their good work, as shown, for example, by United States v. Google LLC. Also see Prosecutors and AI: Navigating Justice in the Age of Algorithms (8/30/24).

Let us embrace a marketplace where lies and deception are punished, not rewarded, where the rule of law and honest integrity prevail. That day will come, the only question is when. Will the country backslide again before it moves forward? Will most of the courts maintain or reattain independent integrity? Will our political leaders, our big tech CEOs? Will we the people?

It will be interesting to see how this all plays out in the next five to ten years, maybe less, especially if democracy remains strong world-wide. It will also help if OpenAI hits another home run with GPT Search, or some other company does so. I will let you know about GPT Search after I get my hands on it. Hopefully it will not suggest we add glue to keep the cheese on our pizza from slipping!