What Is an AI Search Engine?

DataStax
14 min readAug 29, 2024

--

By William McLane

Leveraging AI to facilitate automation and simplify our daily lives has been at the forefront of conversation with the rapid adoption of generative AI in application development. One of the primary areas AI brings a significant amount of value to is changing how we approach search and retrieving data. When Google first introduced its search engine, it revolutionized how data could be mapped and presented so that people could easily find information based on keywords they were interested in. Searching for nearby restaurants was quick and easy as Google mapped data about location to restaurants in a 20 mile radius of that location and shared a list of options that might satisfy the query. Queries could be more or less broad which yielded different results but it was all based on keyword mapping and alignment.

With the advent of AI, searching for data has fundamentally changed. When keyword searching was introduced it transformed how we stored and mapped data, but the data was still stored and presented in a structured way that allowed humans to decipher it. With AI, data no longer needs to be stored in a structured way; it can be stored in what is known as a vector that includes thousands of attributes about the data all of which can be easily used by AI to determine if that piece of information is relevant.

That is the power of what AI-based search provides is that ability to move beyond deterministic keyword search to highly relevant non-deterministic vector search capabilities. In this guide we are going to explore what AI-based search is, how it is implemented and how applications can benefit from using AI-based search methods.

An overview of AI-based search

At the core of AI-based search is the ability to interact with the machines and algorithms leveraging data in a natural way. NLP, or natural language processing, provides the ability for machines to interact with human language without sacrificing all the complexity and diversity that language and speech provide. With NLP computers can detect patterns and similarity of words within language and use that to provide more relevant responses. Take for example something as simple as how we describe things in the English language. I can use the same word to mean two very different things, “My dinner was salty” means something very different compared to “Jane was acting very salty.” The word salty in these two contexts are very different, NLP provides the ability for computers to provide discernment on what is actually being stated and requested.

In addition to NLP, AI-based search leans on machine learning to provide cognitive abilities across large datasets. When machine learning is used, how the system reacts and responds is no longer a function of mathematics but becomes a function of recognizing patterns in data and using those patterns to make predictions and decisions. An interesting way to visualize this, actually uses mathematics which seems a little counter-intuitive, but it is how we were taught multiplication as children. If we are given the problem 3 * 4, we can quickly determine the answer is 12, but how did we determine that answer? Mathematically we know that 3 * 4 can also be represented as 3 + 3 + 3 + 3 or 4 + 4 + 4, but we didn’t come to the answer 12 by adding 3 + 3 + 3 + 3, we just knew it was 12. The reason for this is as children many of us were required to memorize multiplication tables so that when we saw the pattern of 3 * 4 or 4 * 3, we knew the answer was 12. Machine learning does something similar but on a much larger scale. Machine learning teaches computers to recognize patterns, memorize those patterns and provide responses much quicker.

Machine learning is like when we took timed multiplication tests over and over again in school, it allows for computers to learn and recognize patterns and ultimately use those patterns to provide more relevant answers quicker.

It is the combination of NLP and machine learning that is ultimately transforming how search is being implemented with AI. AI now can provide more detailed, contextual responses based on natural interaction. Take our example above of searching for a restaurant near me, with traditional search I will get a list of restaurants that are in the area, and I can search reviews and determine what might be of interest for me. With AI-based search I can ask “What is the hottest new restaurant in the area?” and AI can provide me with a detailed recommendation based on location, reviews, personal context, value, and much much more. It can take into account what hottest means in this context and what new means in this context to provide me with a relevant answer.

What are the benefits of using AI-based search?

The reality is that searching for information isn’t new, so what is the benefit of AI-based search compared to the traditional approach we have been using? In traditional search the correlation of data is done by the assignment of things like keywords and key/value pairs. This approach is highly scalable and very well proven, but it also has limitations. The results are tightly coupled to how keywords are assigned and weighted. The results are typically generic in nature, sure some personal contextualization can be done based on things like location of the query but individualized personalization can be difficult. The other challenge is that it is extremely difficult to find relevance in large volumes of data. The best example of this is how almost everyone uses Google’s search engine to find information. Admit it, have you scrolled below the first 5 results? Even if you do, I am willing to bet that you don’t scroll much beyond the top 10 or 20 results, which means there are hundreds if not thousands of results that you are never seeing but may be of interest to you.

This is where AI-based search is changing the game for how we deal with data. With AI-based search, search is not regulated to a small number of keyword or key/value associations; it can use vector embeddings on data that provide thousands of different inflection points that can be used to determine if that data is relevant to a particular query. Along with the specific information that can be used to define the data, metadata about context can also be used by AI-based search to provide a more personalized experience. It is this ability to correlate vast amounts of information with vast amounts of context that provides the end user with a much more personalized response ultimately giving them more relevant information and better user experience.

Components powering AI search engines

AI powered search engines clearly are different from our traditional search engines, but how are they different and what are some of the core components at the foundation of AI search engines? The core of AI-based search engines comes down to two main technologies: machine learning, which has had significant advancement around how AI can understand and optimize information for fast processing and retrieval, and NLP, which has changed how we as humans interact with the data. Both machine learning and NLP have changed how information can be stored, processed and accessed. Within each of these concepts there are advancements around things like vector storage, vector search, nearest neighbor algorithms, and retrieval-augmented generation (RAG) that all provide AI search engines with optimizations and efficiencies for highly relevant understanding and responses.

Natural language processing (NLP)

Back during the dawning of the internet, before Google dominated the search engine market Yahoo!, Excite, and Ask Jeeves were just a few that offered people a way to find information on the web. Ask Jeeves was probably the one with the most promise because it allowed people to interact with the web by just asking questions like “What do I need to fix a broken drawer handle?”, and it would provide you with links to websites that could help provide you with the answer.

The problem was at the time this was all based on keywords so your question had to contain a keyword that would relate to the keyword of a website for the response. So while the engagement was more natural, question and answer, the result was still very mechanical.

Fast forward to today and the advent of NLP and solutions like ChatGPT. With ChatGPT you engage with information in a truly natural way, asking for it to explain how something works or how to fix something and the result isn’t a website with the information embedded in it, it is the actual explanation or steps to fix a given item. Interacting with ChatGPT using NLP allows for a conversation on a topic compared to a list of reference materials on the topic. With recent announcements by OpenAI and demos of how overlaying NLP with real-time audio generative AI, we now have the ability to not only interface with AI search via text but now we can engage and interact with them like we do anybody else in the real world, by simply asking our AI assistant a question.

Machine learning examples

Where NLP provides significant advancement in how we interact with AI-based search, machine learning is the major advancement that powers the brains behind AI search engines. The biggest advancement in machine learning for AI-based search really boils down to the general availability of large language models (LLMs) that provide training on extremely large datasets.

What LLMs provide is the ability for data to be indexed and vectorized in a way that was never possible before. Traditional search engines had to provide correlation of information through the use of keywords, which meant that information was indexed based on how it was labeled and tied to specific keywords. While this provides a fairly accurate mapping, it is fairly rigid in its implementation and execution. If I search for dog toys I will get a list of things that are tagged as dog toys, but what if I am really interested in getting my dog tennis balls? Unless tennis balls are tagged as dog toys, I might not even see tennis balls as an option because tennis balls are not really designed to be dog toys.

But with AI search, machine learning and LLMs, we have a different way to correlate information — semantic/vector search. With semantic search, AI-based applications have the ability to traverse massive amounts of data in real-time. So instead of looking for what things are tagged with the label dog toys, AI-based search can use nearest neighbor algorithms to pull all things that are within a close proximity to the term dog toys. More importantly, with RAG the inventory catalog can be updated in real time so that as people search for dog toys and purchase tennis balls, tennis balls will be moved closer and closer in proximity to the query of dog toys.

The above is a basic example and AI search using machine learning, LLMs and RAG have a lot of underlying detail in how they gather, store and index data with 1000s of different pieces of metadata for fast retrieval. The key, however, is the AI-based search query responses now have that ability to provide context around what is being asked and provide more relevant answers because of how data can be used to train and learn from patterns over time.

Supporting multi-modal and cross-modal searches with AI Search

The other area that AI-based search is changing how to interact with data is in the realm of multi-modal/cross-modal access of data. With the ability of LLMs and data to be vectorized in a way that allows for semantic correlation of that data we are presented with new functionality in search that have never been possible before, because we are no longer limited to just text-based search and retrieval. With AI-based search we now have that ability to encode data as vectors and use those vector embeddings not only to search for blocks of text but also use those encodings to search for samples of audio files, video files and image files.

In a traditional search approach, handling diverse data types and objects is extremely challenging. How do I search for an image that includes a specific logo for example? But with AI search, images can be turned into vectors and then I can create a query that includes the vector representation of the logo I am looking for. The AI search engine can use that query vector to find all images that the vector embedding included in their vector embedding and use that to return a series of results that are nearest to the original logo image.

This is where the true benefit of AI-based search comes in, as it allows for search across a much broader set of data that can be used not only to search text but also find patterns and similarities across any object that has been vectorized.

Examples of AI search engines

Like the early days of traditional search, there are currently a number of different options when it comes to AI-based search. Tools like Brave, Andi and Metaphor all have been built to leverage the concepts of NLP and machine learning to provide a simple way to get answers to queries in a natural way.

Below are three examples of searching for an answer to this question. What is interesting is if you look at both Andi and a traditional Google search you have to weed through the results to find the actual answer. The Google search seems to indicate that the Allure of the Seas is the largest ship but digging deep you find out that the Icon of the Seas is actually the current largest ship. Andi does a better job and points you to Royal Caribbean’s website, but it also provides a number of supporting articles that clearly state the Icon of the Seas as the largest ship in their fleet. Brave on the other hand provides the most accurate and most relevant answer to my question. By pulling context from wikipedia.org, Brave provides me with an answer that is factually accurate and provides additional details that I might be interested in as well as potential follow up questions. If you notice it did it even though I misspelled the word royal caribbean in the query, too.

Google search for “What is the name of royal caribbean’s largest ship”
Andi search for “What is the name of royal caribbean’s largest ship”
Brave search for “What is the name of royal caribbean’s largest ship”

The relationship between AI search engines and vector search

At the core of all of these AI-based search engines is a fundamental optimization they have made in accessing data via vector search. It is this shift in approach from tagging information with keywords to using highly dimensional vector representation of 1000s of points of metadata that are used to define how data relates to other data that changes how AI-based search engines query and retrieve information.

With vector search and AI-based search engines, queries for information behave more like having a personal assistant that can do all the research on a topic for you and provide a summarized report of what you are requesting. With vector search, AI can easily gather all the semantically related information on a specific set of metadata, process it, summarize it and reference it all within a matter of seconds. The challenge and key to providing this however is relevance and responsiveness when issuing queries and indexing data for retrieval. You can find a very detailed report of performance semantics of vector search from GigaOm here that highlights some of the important factors of vector search performance evaluation.

Vector search and multi-modal search semantics

The other area that vector search plays a huge role in AI-based search is with multi-modal search semantics. For example with vector search, you can now search video footage for all segments where a news anchor is facing the camera to eliminate all footage that is not usable or you can search an audio file for all segments that only have a single person speaking to eliminate sections where people are talking over each other. This ability to search data types and formats for information in new ways opens up a whole world of possibilities for AI-based search engines.

AI’s role in contextual understanding and user intent recognition

Probably the biggest benefit of AI-based search however has to do with smart context. Traditional search engines provided some level of contextual understanding but this was done with gathering data that was available to point at keyword associations that potentially made sense around the original query. Like in our example above where we searched for dog toys, in a traditional search if I then searched for tennis balls I can use the context of dog toys to provide results for tennis balls in the context of dog toys.

With AI search however we have the ability to embed not just the data itself but context around the data and provide a much broader scope of results based on individual characteristics, regional context, past search history and level of scope and detail. AI search brings in the assistant mentality where each individual can customize their engagement. With architectures like retrieval-augmented generation (RAG), AI search engines can constantly be updated with the latest information, but also with the latest understanding of the individual querying for information. With AI search I can prompt the AI to provide more or less technical detail. I can prompt the AI to respond with only results with a specific set of criteria that are more inline with what I am looking for.

Applications of vector search and AI search engines

This is where the true power of vector search and AI search engines comes into play. The ability to enable consumers of information with the ability to search enterprise specific data with smart context. The challenge is that AI search right now has been primarily enabled with general purpose LLMs that are using data that is publicly available. However, to truly provide the next generation of AI-based search engines and assistants the data used has to include private enterprise data. Think about being able to search your medical records using an AI assistant that can help summarize test results and provide preventative care suggestions. Think about being able to search your financial records to find patterns in how you can optimize spending and remove frivolous purchases or subscriptions. All of this is now possible with AI based search, vector search and RAG.

Technical considerations for AI search

There several technical considerations that need to be considered when it comes to implementing AI-based search on your data:

  • Is the data ready for AI-based operations?
  • Where does the data need to be stored and processed: on-premise, hybrid cloud, public cloud, etc…?
  • What is the sensitivity of the data and what steps are made around security and governance of the data?
  • How and where does the data need to be integrated with enterprise applications?
  • Do systems need to be upgraded or modernized for maximum potential and hardware utilization?

Get started using vector search with DataStax

AI-based search engines have the potential to truly revolutionize how we interact with data. The benefits from interacting with data in a more natural way, using machine learning to augment the results and provide more relevant information with smart context, can truly change how we interact with information in our daily lives.

To implement AI search for enterprise data can be challenging and there are lots of things to consider. That’s what DataStax does: simplify your journey with your existing data infrastructure and provide a simpler way to bring your data into the generative AI realm. Whether you’re looking for an easy way to store and retrieve your traditional and vector-based data using DBaaS functionality with Astra DB, or if you’re looking to modernize your on-premise infrastructure to be enabled for a fully virtualized environment built and designed for hyperconverged infrastructure with DataStax Hyper-Converged Database, DataStax can fast-path your journey to enable your data for AI-based integration.

--

--

DataStax

DataStax provides the real-time vector data tools that generative AI apps need, with seamless integration with developers' stacks of choice.