I once invested $250,000 in a startup that promised to revolutionize news verification. They had a slick pitch deck, a team of PhDs, and a tool that they claimed could spot deepfakes with 99% accuracy. Twelve months later, the company was dead. The tool was a disaster. It flagged a video of a cat playing the piano as a potential deepfake and completely missed a manipulated video of a politician giving a fake speech. That expensive failure taught me a hard lesson: in the world of computational journalism, most tools are junk.
As an investor in over 200 companies, including some of the biggest names in AI like Anthropic and OpenAI, I get pitched new AI tools every single day. They all promise the world. They all claim to be the next big thing. But when you actually get under the hood and see how they work, you realize that most of them are built on shaky foundations. They’re either too simplistic to be useful, too complicated for anyone but a data scientist to operate, or just plain don’t work.
This is especially true in computational journalism. The idea of using AI to analyze data, uncover stories, and fight misinformation is powerful. But the reality is that the tools we have are not up to the task. They are often built by people who don’t understand the realities of a newsroom, the pressures of a deadline, or the nuances of journalistic ethics.
The Big Problem with Most AI Journalism Tools
The fundamental issue is that most of these tools are designed to find easy answers in messy data. They are built for a world of clean, structured datasets, not the chaotic reality of news gathering. Real-world data is a nightmare. It’s full of errors, biases, and missing information. A tool that can’t handle that is a tool that’s going to lead you to the wrong conclusions.
I saw this firsthand with a data journalism project I was advising. We were trying to analyze a massive dataset of campaign finance records to look for evidence of illegal contributions. The tool we were using, a popular open-source data cleaning library, kept crashing. When it didn’t crash, it would “correct” the data in ways that were just wrong, like changing the name of a contributor from “John Smith” to “John Smithson” because it thought “Smith” was a typo. We ended up having to clean the data by hand, which took weeks and almost killed the project.
Another big problem is the black box nature of many of these tools. They take in data, spit out an answer, and give you no real visibility into how they got there. This is a huge problem for journalists, who need to be able to show their work and explain their sources. If you can’t explain how your AI tool came to a certain conclusion, you can’t use that conclusion in a story. It’s that simple.
I remember a tool that was supposed to be able to predict the outcome of elections based on social media sentiment. It was incredibly popular for a while, and a lot of news organizations were using it. Then, in one election, it got the result spectacularly wrong. It turned out that the tool was heavily biased towards a certain type of social media user and was completely ignoring a huge chunk of the electorate. The news organizations that had relied on it had to issue embarrassing retractions.
What I Use Instead: A Battle-Tested Toolkit
So, what’s the alternative? Over the years, I’ve developed my own set of tools and workflows for computational journalism. It’s not a single, magical solution. It’s a collection of battle-tested, often unglamorous, tools that actually work. Here are a few of them:
For data extraction and cleaning, I use a combination of custom Python scripts and a tool called
DataWrangler.DataWrangleris an interactive tool for data cleaning and transformation. It’s not as flashy as some of the newer AI-powered tools, but it’s incredibly powerful and gives me a huge amount of control over the data cleaning process. I can see exactly what the tool is doing at every step, and I can easily undo any changes that I don’t like.For network analysis, I use a tool called
GraphTool. It’s a Python library for the manipulation and statistical analysis of graphs. It’s not the easiest tool to use, but it’s incredibly fast and powerful. I’ve used it to map out everything from the spread of misinformation on social media to the hidden connections between shell corporations.For text analysis, I’ve built my own custom tool on top of the Hugging Face Transformers library. This allows me to fine-tune pre-trained language models for specific tasks, like sentiment analysis or named entity recognition. This is a much more powerful approach than using an off-the-shelf API, as it allows me to create models that are tailored to the specific domain I’m working in.
For geospatial analysis, I use a combination of
QGISandPostGIS.QGISis a free and open-source geographic information system.PostGISis a spatial database extender for PostgreSQL. Together, they provide a powerful platform for analyzing and visualizing geospatial data. I’ve used them to do everything from mapping out the locations of polling stations to analyzing the impact of sea-level rise on coastal communities.
The Future of Computational Journalism
I’m not saying that all AI journalism tools are useless. There are some genuinely exciting developments on the horizon. I’m particularly interested in the potential of large language models to help journalists with tasks like summarizing long documents, generating story ideas, and even writing first drafts.
But we need to be realistic. AI is not a silver bullet. It’s a tool, and like any tool, it can be used well or it can be used badly. The future of computational journalism is not about finding the one perfect tool that will solve all our problems. It’s about building a culture of critical thinking and experimentation. It’s about teaching journalists how to use these tools effectively and ethically. And it’s about holding the creators of these tools accountable for the claims they make.
I’m still an optimist. I still believe that AI has the potential to revolutionize journalism. But I’m also a realist. I know that we have a long way to go. The road ahead is going to be full of hype, disappointment, and hard work. But if we’re willing to put in the effort, I believe we can build a future where AI and journalism work together to create a more informed and just world.
I’ve seen enough to know that the real breakthroughs don’t come from the tools with the biggest marketing budgets. They come from the scrappy, open-source projects built by people who are passionate about solving real-world problems. They come from the journalists who are willing to get their hands dirty and learn how to code. They come from the news organizations that are willing to invest in training and experimentation.
That’s where I’m putting my money. And that’s where I think the future of computational journalism lies.
Frequently Asked Questions
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.