How my team trained AI to Read the Bill
The new Congress in Your Pocket app accomplishes what many lawmakers and the public have demanded for years. Here's how we did it.
Programming notes: New episode of The Ground Game posting tomorrow with
as well as fresh content from Mr. , proprietor of “The Booboisie” here on 24sight News.Catch Tom LoBianco in-studio with Julie Mason on SiriusXM Channel 124, Wednesday night 8-9p ET, during President Joe Biden’s farewell address.
**Support independent journalism with a paid subscription — $5/month, $50/year.**
There have been two times in my professional career where technology instantly changed the world and I ran to meet it. The first was the release of the iPhone in 2008. It ushered in the Age of Mobile, which has hyper-personalized our lives as well as our politics. I didn’t just want one on Day One, I wanted in on the shift.
I remember watching Steve Jobs deftly introduce the new thing and immediately knew what I wanted to do with it. That became Congress In Your Pocket, a suite of mobile apps which help connect voters to Capitol Hill with not only members but their staff who do majority of the hard work for constituents.
The second shift was the release of ChatGPT and the dawn of generative artificial intelligence.
It has ushered in the Age of AI. The underlying technology has been around for years and widely used in the political world. The big advance was that generative AI could create political content, not just analyze it or voter behavior.
I immediately knew what I wanted to do. Congress In Your Pocket’s mission was to help connect voters to their government’s first elected institution. Its new directive was going to be to help voters understand what they’re doing. This meant adding legislation summaries to the app for everything Congress debates.
My team launched Congress In Your Pocket AI (CIYP/AI) last week, and it’s magic. We were first-to-market again, this time with AI helping voters understand improve their understanding of what’s inside the bills being debated in Congress.
Evolving past a simple hashtag of #ReadTheBill took more work than you might expect. I won’t reveal all of how we accomplished that here, but I will give a peek.
First, I had to convince my partner and chief technology officer that it was worth it. Would what we spent in time and treasure result in a net profit or was this just one of those things worth doing without a solid business case for it? We felt that the technical cache of being first-to-market on mobile was worth it and time will tell if we’re right.
Next, we had to figure out how to build it.
The first year and a half was difficult because the open-source tools were not great and not widely available. But around the turn of year in January 2024, we started seeing some models perform almost as well as ChatGPT. Later in the year, they were effectively operating at the same capacity and efficiency. So, we began with open-source and built our custom large-language-model on top of it.
The core issue to solve was that congressional legislation is long and so your basic chatbot experiences won’t do. We had to figure out how to break legislation into finer pieces for processing then recombine them into a summary that made sense for the average voter. On that, I’ll keep the details private but let’s just say there was a lot of hard work involved.
The result is that you can now get a summary of any piece of legislation distilled down into a couple of paragraphs, generally one or two screens on an iPhone. We also offer the full text of the bill so you could read the whole thing if the summary piques your interest.
We envision this for two separate groups: voters and policy professionals. Voters can get the most benefit because they generally don’t follow even major bills closely. After hearing about a major health, immigration, or tax bill they can look it up and get a quick summary. If they want to learn more they can see who is sponsoring it, where it is in the process, as well as the entire text if they have that kind of time.
Policy professionals, or those like Tom who covers them, can benefit from monitoring more bills at once, diving deeper when the topic is on point, and scanning the others that might be important someday just for awareness. It amplifies the teams of nonprofits and trade associations that have small government relations teams. One of our clients, IBM has a very small team.
In both cases, CIYP/AI doesn’t solve information overload, instead it helps move us toward a broader solution.
That’s what I’m seeing more broadly in the political marketplace. This past cycle, we saw presidential campaigns experiment and discard ads that used fake voices and fake video. It turns out fakes are easily spotted and almost universally cringy. In our Age of Authenticity, voters prize the real, unvarnished, and definitely-not AI created candidate.
But under the hood and downballot, outside of the view of voters, campaigns were inching their way forward, using generative AI tools to draft content, test messaging, and model potential voter turnout projections. There are even firms that are working on ways to use generative AI to draft campaign plans, but I am not concerned that robots will be running campaigns anytime soon.
I’ll have a lot more to say about AI in the second edition of my book, Modern Political Campaigns, due out in late April. You can subscribe to the Substack here.
We need to pump the brakes on looking at AI as the solve for everything or the end of all things. It’s not. In technical parlance, it’s a framework, albeit a powerful one, that is becoming more useful in all facets of our life, including our politics. This is progress we should celebrate.
Michael Cohen, is the author of the book Modern Political Campaigns, president of Cohen Research Group and a 30-year veteran of the polling industry. He writes The Level regularly for 24sight News, analyzing polling and campaign trends with a keen eye and level-headed approach.
 | A guest post by
|
Hmm, I am a newbie - he says, slyly flashing his deep knowledge of what the kids say these days (or perhaps decades ago) - but does AI info inspire a deep dive or dispense the need for it?
Not here to quibble - he says quibblingly
- but surely the introduction of the World Wide Web was a game changer. I had worked when the Internet still had no graphics—the WWW tech helped bring graphics to the fledgling net.