New York Times Harnesses AI to Supercharge Investigative Reporting

The New York Times is pioneering a new era of investigative reporting by integrating artificial intelligence into its core journalistic practices. Through the efforts of its AI Issues team, the Newspaper is not just experimenting with AI but is actively institutionalizing its use to uncover deeper insights and enhance the efficiency of their investigations.

zach Seward, Editorial Director Of Artificial Intelligence Initiatives at The New York Times, highlighted the transformative potential of AI at WAN-IFRAS recent Congress in Krakow. He emphasized how AI tools are empowering journalists to sift through massive datasets in ways previously unimaginable, effectively giving them a “superpower.”

the Four Pillars Of NYT’s AI Investigative Toolkit

the New York Times has identified four repeatable patterns in its use of AI for investigative purposes, leading to the growth of a specialized toolkit:

1. Vibes-Based Search
2. Diving For Pearls
3. Augmenting Datasets
4. End-To-End Verification

Vibes-Based Search: Finding Meaning Beyond Keywords

Vibes-based search, also known as semantic or vector search, allows journalists to uncover connections that customary keyword searches might miss. By encoding text as numerical vectors, this method identifies semantically similar content, revealing variations in terminology and hidden patterns. This capability is invaluable when exploring complex topics where nuances in language can considerably alter the meaning.

Did You Know? Semantic search algorithms have improved by over 40% in accuracy in the last two years, thanks to advancements in transformer models.

The Math Behind Semantic Search

Semantic search uses mathematical principles to understand the relationships between words:

• Text is converted into numerical representations (embeddings).
• Similar concepts cluster together in a mathematical space.
• Distance calculations reveal semantic relationships.
• This allows for “equations with text,” such as: [king] – [man] + [woman] ≈ [queen].

Diving For Pearls: Extracting Insights From Vast Content Volumes

This data extraction tool leverages AI to pinpoint crucial insights from overwhelming amounts of content. By combining AI with the expertise of seasoned journalists, this approach structures findings into organized spreadsheets, categorized by topics of interest. The New York Times employed this method to analyze over 500 hours of leaked video from an election interference group, showcasing its potency in handling massive data troves.

Pro Tip: Journalists can enhance AI’s effectiveness by providing carefully crafted prompts that guide the AI towards specific areas of interest.

Augmenting datasets: Unlocking Details From Complex Documents

The New York times uses optical character recognition (OCR) to analyze complex document sets, including handwritten notes. The latest foundational models from major LLM developers have significantly improved OCR capabilities, enabling intricate analysis of messy datasets.This has allowed the NYT to monitor “manosphere” content and screen 10,000 individuals for a Puerto Rico tax investigation.

In 2023, OCR technology saw a 30% increase in accuracy when processing handwritten documents, making it an increasingly reliable tool for investigative journalists.

End-To-End Verification: Ensuring accuracy and Trustworthiness

The new York times emphasizes that AI-assisted reporting must always adhere to strict verification protocols. Journalists are required to trace all insights back to the original sources to ensure accuracy. The NYT’s verification tool links AI-generated findings back to primary sources via spreadsheet formats, mandating a review of the original material before publication.This commitment to verification is crucial for maintaining the integrity of investigative reporting.

“It’s an axiom in our newsroom that you should never trust an LLM… before any of that makes its way into a story, we’re always going all the way back to the original material,” says Seward.

The ‘Cheat Sheet’ Tool: Streamlining Data Analysis

The ‘Cheat Sheet’ tool, currently under development, assists journalists in making sense of large datasets. This spreadsheet-based interface transforms unstructured data into structured formats, extracts quotes matching specific criteria, performs translations and summarizations, and connects findings directly to source material for verification. It enables reporters to efficiently verify information and maintain journalistic integrity.

how do you think AI will further transform investigative reporting in the next five years? What ethical considerations should news organizations prioritize when implementing AI tools?