How a College Student Outsmarted Major News Networks with AI

At one point or another we have all found ourselves doing this where we are at home watching a live political debate, press conference, or even a viral video clip containing someone speaking inappropriately or unusually out of character and try to figure out if they really are lying. You pick up your phone, enter information into Google, browse for 3 different articles that might contain information and try to find out if what has happened is accurate and after hours of searching discover that what the individual said was inaccurate. Now 3 different topics have gone past since the time that what was said in that political debate or press conference.

This cycle is very frustrating because journalists and major media networks are supposed to call out lies, however it’s impossible for anyone to provide real-time fact checking on live television without making many mistakes.

But imagine if you didn’t need to wait for the media to provide you an analysis of what occurred at that political debate after the fact? Imagine if you had the truth at your fingertips the precise moment a politician made a statement?

Instead of being created by a massive tech company in California's Silicon Valley or a large news agency with a multi-million dollar research and development budget, Intrude was developed by a single college student on his/her college campus (typically a dorm room). Intrude is an AI-based, real-time lie detector for use with political candidates. Intrude uses Anthropic's Claude as its artificial intelligence engine.

Here is how this ground-breaking new tool works, why it has everyone talking, and what it says about where we are today in terms of artificial intelligence.

How Intrude Changes the Game

If you ever spend time in a discussion on technology or politics on Reddit, you must have come across viral clips. A user installs Intrude, gets a YouTube livestream of a political speech, and keeps the video in view.

A sleek, unobtrusive overlay appears on one side of the browser window as the politician speaks. The minute the speaker either makes a factual claim, Intrude intervenes.

At first, using it in real-time audio transcription is easy. The real exciting events happen in a split second afterward. Intrude takes that claim that was transcribed in real time and instantaneously got it to Claude. The AI searches the live web, gathering actual verifiable proof, and cross-matches the statement of the politician against highly authoritative sources−think peer-reviewed journals, data portals controlled by governments, and trusted historical records.

It's doing a lot of heavy work before the politician's next sentence has ended.

Lo and behold, a little, neat bubble appears next to the video player. It highlights the claim and delivers a verdict. However, why Intrude is going viral is not just speed but also how it looks.

The Nuance of "Gotcha" Moments

If Intrude merely placed a huge red "FALSE" or green "TRUE" sticker on the screen, it would be an entertaining gimmick but would ultimately not be useful. Everyone who has been involved with political campaigns knows that most of the statements people make in a political campaign are not black and white: politicians are very good at using the gray area. For example, they use numbers to support their arguments and they quote people out of context, or they manipulate the truth to create the appearance that what they are saying about a subject is incorrect.

The college student who created Intrude understands this concept perfectly.

Rather than having a simple true/false system, Intrude classifies its outcomes. A statement that is made will likely have several categories associated with it, such as "Substantially True," meaning that the main part of the statement is true but that there is at least one piece of information (for example, a detail) that is not true.

"Misleading" is the most beneficial component of the extension.

Example: If a politician says, "Unemployment dropped 50% under my watch," Intrude will identify this statement as "misleading," because while the raw percentage may have actually fallen, the true reason for the decline was due to a significant drop in the size of the workforce in comparison to job growth.

"Unverifiable" is another category that is included in the functionality provided by Intrude. If a politician makes a claim regarding a private meeting or a very personal story, there will not be a public record of this activity, and so the AI would not make up a response; it will simply inform the user that there is no information to support or refute the claim. As a result, there is a very small number of early-stage Artificial Intelligence applications that take this type of approach, which demonstrates a strong understanding of responsible AI development practices.

Demonstrating The Evidence.

The "black box" issue is the most frequent, and accepted, criticism of the use of AI right now. How can we trust an algorithm to tell us what is true? If AI makes up a fact, it is simply substituting a human-made falsehood with a digital one

Intrude addresses this issue by not requiring you to blindly trust them.

Each and every judgment you see on the screen after you have made the inquiry will come with proof. For every "Misleading" and "Substantially True" result, Intrude provides you with links that can be clicked to view the exact source information used by Claude to make that determination. They will even provide you with the calculations used to derive the answers.

For example, if an AI fact-checks a statement relating to the environment, it will provide you with a link directly to the EPA report that was used. If it fact-checks a statement regarding historical tax rates, it will direct you to the IRS tables of historical tax information. This way you can fact-check the AI directly/independently since the information comes from a credible source, ensuring that AI is determining the truth rather than simply providing its opinion. AI behaves more like an extremely efficient/highly capable alternate research assistant who is sitting next to you helping you find the information rather than a definite arbiter of what is or was true.

Why the Internet is Going Wild

The reason the Reddit threads around Intrude have tens of thousands of upvotes have to do with one specific thing: it is a huge wake-up call for established organizations.

Established media organizations have told us for years that they will bring improvements to their fact-checking, and they have done lots of talking about it - they have hosted panels of experts; they have dedicated segments for them, and research teams are working 24/7 on fact-checking. However, when they are trying to live fact-check as statements are being made in real-time on a debate, the best that they usually do is a post-debate show that streams an hour after the false claim has already been made.

What this one college student (the developer of Intrude), using a laptop and some straight-up code to integrate the API for Claude and a great idea, was able to accomplish with regards to creating a real-time, accessible, transparent fact-checking tool is a complete reflection of where Artificial Intelligence (AI) currently is.

What one developer has accomplished represents the fact that building world-changing applications has never been easier because barriers to entry have fallen tremendously. You no longer need a large development team to solve even the most complicated, real-world problems. All you need is a good idea and the ability to pull together the mass amount of amazing AI tools that currently exist.