# Finding the Signal in the Noise of Generative AI | Capterra > Valkyrie CEO and founder Charlie Burgoyne, discusses the state of generative AI, some of its problems, and introduces us to the next frontier: deductive AI. Source: https://www.capterra.com/resources/generative-ai-and-deductive-ai --- Types of SoftwareIT & Software Development # Finding the Signal in the Noise of Generative AI By Charlie Burgoyne Charlie Burgoyne Charlie Burgoyne is the founder and CEO of Valkyrie, an applied science and AI firm of top scientific talent. Austin-based Valkyrie works to solve industrial... [See bio & all articles](https://www.capterra.com/resources/author/charlie-burgoyne/) Published August 30, 2023 7 min read Table of Contents - [Deductive AI prioritizes context over content](#deductive-ai-prioritizes-context-over-content) - [Explaining the noise](#explaining-the-noise) - [Finding the signal](#finding-the-signal) - [Following the signal](#following-the-signal) - [Separating the signal from the noise](#separating-the-signal-from-the-noise) ## AI firm founder Charlie Burgoyne talks using deductive AI to make sense of cluttered content. It’s not hard to understand the buzz around generative [artificial intelligence](https://www.capterra.com/artificial-intelligence-software/) (AI). Industries as varied as software development and [customer service](https://www.capterra.com/customer-service-software/) have found novel applications of generative AI like OpenAI’s GPT-4 to streamline their workflows through automated content creation. Amid this widespread adoption, it’s crucial to acknowledge that current generative AI and associated large language models (LLMs)face serious limitations. While generative AI can provide instant access to seemingly infinite information and generate content that appears original, current implementations are unable to consistently return precise and reliable results. A generative AI LLM can generate as much content as you like, but whether that content is relevant, unbiased, or even accurate is hardly understood. AI can serve businesses in two ways: automating tedious activities and helping leaders make complex decisions. Generative AI tools are great for automating tasks like generating resumes, itineraries and summaries. However, when businesses need to make more intellectually complicated decisions—such as deciding how to channel operational resources for a major initiative—that’s where deductive AI comes in. ## Deductive AI prioritizes context over content This limitation means that generative AI models rarely return precise and explainable answers to user queries. All too often, all they return is noise.  The output of these generative tools is essentially the “average” of what thousands of experts have already stated. Generative AI is less about generating content as much as finding consensus at scale. However, utilizing AI to bridge the gaps between the context embedded within content requires a new domain in the field of AI: deductive artificial intelligence.  As the name suggests, a deductive AI tool is designed to be sensitive to the full context of a user’s query. Deductive AI can make sense of vast data sets to identify the most relevant and helpful information for each request. With this simple distinction, LLMs used to deploy deductive AI can separate the signal from the noise of generative AI, enabling new possibilities for AI-powered decision making. Takeaway If generative AI can turn a prompt into a research paper, deductive AI can show you that paper's main themes and thesis. ## Explaining the noise Context sensitivity is the key to extracting meaningful insights from artificial intelligence responses. After all, the lack of context sensitivity in current generative AI models explains why they tend to generate so much noise. When you give a prompt to a generative AI tool, the tool does not attempt to parse your intention or interpret the context behind the query. Instead, it runs a series of calculations to determine which words are most likely to follow the initial query, and then creates a response using those words in the likeliest order. No effort is made to confirm the accuracy or relevance of this output; it’s all just a calculation. This isn’t to say that generative AI models don’t contain helpful data. Quite the contrary: current LLMs are trained on enormous swaths of information derived from websites, books, articles, and everything in between. Yet, without the ability to detect the nuance of user requests, generative AI models are unable to extract the most relevant results from their vast data sets. Takeaway Ultimately, valuable insights and context are integral to the training data used by generative AI tools, but current models are not designed to exploit that context, as their primary function is text generation. ## Finding the signal Deductive AI provides the framework to make sense of the noise from endless libraries of content or even generative AI itself. Rather than generating content based on probability and likely word order, a context-sensitive AI platform is trained on specific, user-provided data sets and designed to comprehend the human context of each query.  A context-sensitive AI does not generate content, but rather can search and map out large bodies of text. This design allows context-sensitive AI models to return precise answers that are specifically attuned to user questions.  If a user query can be interpreted multiple ways, a context-sensitive AI can determine which interpretation is most relevant and use that guidance to return an accurate and relevant response. Context sensitivity in deductive AI models is made possible through a combination of two distinct AI domains: knowledge graphs and natural language understanding (NLU). Working together, these components allow LLMs to graph the relationship between pieces of information and clearly communicate a response to the user. Knowledge graphs provide the information architecture necessary for a deductive AI to understand contextual signals. Knowledge graphs enable a deductive AI to semantically map the themes and context of large data sets, which allows the AI to rank data by relevance and only return the most contextually appropriate information for each user request. Given its potential, deductive AI can dramatically alter the capabilities within industries that have large data sets of prose / natural language, such as investment firms, legal firms, government administrations, and tax firms. ## Following the signal Knowledge graphs provide the technical architecture and visual tools necessary to explore the interconnectivity of objects in a data set. When paired with deductive AI LLMs, incredibly powerful search and explainability can be derived from large, otherwise unwieldy data sets.  This functionality can support a vast range of new AI-driven use cases. Here are a few examples of how deductive AI can solve problems that can’t be addressed by existing generative AI models: - **Intelligent search**: With context sensitivity, an LLM can index and catalog data across repositories to enable easy, direct access to complicated information. Intelligent search goes beyond keywords; a deductive AI LLM provides results based on relevancy identified through its knowledge graph structure, rather than by keyword-sensitive elastic search alone. - **Contract auditing**: Using knowledge graphs, context-sensitive AI can audit contracts to compare how previously agreed upon terms compare to recent contracts, streamlining otherwise tedious and time-consuming contracting and procurement processes. - **Knowledge sharing**: While knowledge/[file sharing platforms](https://www.capterra.com/file-sharing-software/) have limited searchability, deductive AI can form the basis of a more intuitive, searchable platform. You can feed a deductive AI hundreds of thousands of documents and the tool will effortlessly surface the main themes and most important information for each document. It’s critical for businesses to understand their knowledge systems—how their organization is sharing information and how those knowledge-sharing processes are instrumented and orchestrated. The efficacy of AI in the workplace highly correlates with how well businesses structure their internal data systems, specifically around knowledge management. ## Separating the signal from the noise It’s easy to get caught up in the buzz of generative AI and settle for the noise: the inaccurate results, the unwieldy data sets, the limited searchability. However, implementing deductive AI from LLM development can complement the already impressive achievements of generative AI to deliver more precise and actionable AI insights out of any data set.  Added context sensitivity allows deductive AI to add intuitive searchability to complex information and deliver direct access to the insights you need.  Ultimately, when you’re considering how to implement AI in your organization, identify whether you are solving tedious challenges that can be automated through generative AI or thinking through complex challenges that can be informed by deductive AI.  While both are derived from LLMs and provide real value, their application couldn’t be more different. Dig deeper on navigating generative AI, ChatGPT, and more with these Capterra resources:  - [Do Marketers Need ChatGPT Now? A Leading Expert Guides Us Through 2023's Biggest Martech Trends](https://www.capterra.com/resources/marketing-expert-on-latest-martech-trends/) - [Why Marketers Need a Brand Monitoring Strategy in the Age of AI](https://www.capterra.com/resources/ai-brand-monitoring/) - [A Renowned Tech Futurist Explains How AI Will Revolutionize B2B Marketing Through Deep Learning](https://www.capterra.com/resources/how-to-use-artificial-intelligence-in-marketing/) ## Capterra's 2026 Software Buying Trends Report ### Download our 2026 Software Buying Trends Report to see how successful software adopters avoid disappointment and how your business can, too. * * * ### Was this article helpful? * * * ## About the Author [### Charlie Burgoyne](https://www.capterra.com/resources/author/charlie-burgoyne/) Charlie Burgoyne is the founder and CEO of Valkyrie, an applied science and AI firm of top scientific talent. Austin-based Valkyrie works to solve industrial challenges with custom AI solutions, invests in asset classes using AI-informed theses, provides philanthropic scientific services, and optimizes car performance in world class autosport. Prior to founding Valkyrie, Burgoyne was the principal director of data science at Frog Design, director of data science at Rosetta Stone, and a research... ### RELATED READING - [How AI Onboarding Tools are Reshaping HR Operations](https://www.capterra.com/resources/ai-onboarding-tools-for-hr/) - [Project Resource Management: What PMs Need to Know](https://www.capterra.com/resources/project-resource-management/) - [5 Types of Accounting Software and How To Choose the Right One](https://www.capterra.com/resources/types-of-accounting/) - [What is the Difference Between Accounting and Financial Management Software?](https://www.capterra.com/resources/accounting-vs-financial-management/) - [AI vs Automation in Project Management: Where Each Actually Delivers Value](https://www.capterra.com/resources/ai-and-automation-in-project-management/) - [What’s Behind the Pricing Jump in Project Management Software Plans](https://www.capterra.com/resources/project-management-software-pricing/) - [What is a Job Requisition? Here’s Everything You Need to Know](https://www.capterra.com/resources/what-is-a-job-requisition/) - [Free CRM Software vs. Paid: Features, Costs, and Comparison](https://www.capterra.com/resources/free-vs-paid-software-for-crm-guide/) - [Accounting Trends in 2026: How AI Is Changing Work, Skills, and Strategy](https://www.capterra.com/resources/accounting-trends-ai-software-changing-work/)