An AI compliance questionnaire – Part 1
As generative AI becomes more powerful and practical for enterprise applications, large companies are establishing protocols to vet the use of generative AI in their business operations. Legal, compliance, and technology leaders want to comprehend the potential risks and rewards of this increasingly popular, but still poorly understood, technology.
When it comes to market and competitive intelligence research, the argument for using generative AI is compelling. Here’s what corporate “AI compliance” teams are asking, and how Northern Light is responding.
1) What problem is generative AI being asked to solve?
Large organizations typically have tens of millions of documents in their market and competitive intelligence research repositories – often a mix of primary research, licensed secondary research, business news, industry and government databases, technical and scientific journals, corporate financial reports, and more – that employees regularly search to find information and distill insights relevant to their research projects.
The business problem being addressed with generative AI in this environment is supporting business decisions in product management, product development, marketing management, and commercial operations with better information.
Northern Light provides a search engine optimized for business research that produces search results of highly relevant documents. For some questions, the user might have to download and read many documents and synthesize a summary to understand the overview. This is a time-consuming process and many times a user may not have the time or patience to glean the insights from many documents relevant to the user’s research project. Also, users vary in their ability to summarize disorganized bodies of information. This is aggravated by the fact that the user may not have a means to determine which documents are most useful just from the documents’ positions on a search result. As a result of these factors, the resulting synthesis may be missing relevant competitive and market research insights.
2) Why is generative AI the optimal solution to this problem?
Generative AI for question answering and summarization can address these problems. Generative AI can find the answers to questions consulting many documents and present the answers directly to the users. This is wildly more efficient than the traditional process and means better answers will be produced on average, better supporting business decisions with superior research insights.
Secondly, generative AI can summarize a topic for the user from across the most relevant documents related to the user’s research topic. This has the advantage of informing the user on what the overall topology of the topics are that the user should be considering. The summary provides the context for understanding and weighing the insights in the content.
3) What is the specific objective for generative AI in this application?
The specific objectives of applying generative AI in a market and competitive intelligence research knowledge management platform are to efficiently answer users’ competitive intelligence and market research questions with higher quality answers and to efficiently summarize research results with higher quality summaries. The output is text that answers the questions and summarizes the answers.
4) What is the generative AI model architecture in this application?
Northern Light uses natural language processing (NLP) to create a concentrated version of each document that only contains the “Summary Worthy Sentences.” These are the sentences that are likely to express insights.
The NLP text from the most relevant documents to the user’s question is submitted to the LLM — Northern Light is using the large language model (LLM) GPT 3.5 Turbo for the solution — via the relevant API in the prompt and the prompt contains instructions to answer the user’s questions from the submitted text. These “Answers” are presented in the UI to the user. A second transaction then occurs in which the Answers are gathered up and submitted to the LLM in the prompt and the LLM is prompted to summarize the answers. This is presented to the user as an “Executive Summary.”
NLP greatly reduces the negative impact of the severe capacity limitations the APIs of OpenAI and Microsoft have for transmitting text and significantly improves the performance of the solution in terms of both question answering and summarization because without the NLP text, the capacity limitations on the APIs exclude content that might be highly useful.