Empathy’s new tool uses AI to generate obituaries, and it’s not half bad

Writing an obituary isn’t an easy task. That’s an understatement — it’s incredibly painful, usually expensive too. But someone has to do it.

Or perhaps not. Consider leaving it to AI.

That’s the pitch Empathy, a platform that provides support for families who’ve recently suffered a loss, is making with the launch of its new tool that uses AI to create obituary drafts. Called Finding Words, the tool generates obits from basic info provided by family members.

“With the overwhelming number of tasks and emotional strain grieving families face, Finding Words allows them to worry less about the task of drafting the text for an obituary and focus more on honoring the memory and legacy of their loved one,” the company wrote me in a pitch email.

But not everyone would agree. Offloading the work of writing an obituary to AI doesn’t sound particularly sensitive, at least to my ears. Wouldn’t family want to be more involved in writing a remembrance of a loved one’s life? Doesn’t letting AI handle the work cheapen it somehow, or feel less thoughtful?

I asked Empathy CEO Ron Gura.

“Many people who experience the loss of a family member struggle to write personal and thoughtful tributes for their loved ones, for a variety of reasons,” he told me in an email interview. “They may be too emotionally overwhelmed to know where to start or preoccupied by the enormous volume of administrative tasks that typically follow a loss. It’s a terrible feeling to be sitting at your computer staring at a blank screen and feeling like you are letting your family and your loved one down. Any support that can guide people through this process is beneficial, and it’s essential that access to such support is democratized and made available to as many people as possible; generative AI serves as an equalizer in this regard.”

Those are fair points. So — in the interest of giving Finding Words a shot — I plugged in some dummy info and had the tool write an obit for me. (Cause of death: Grease fire. Plausible enough in New York City, I thought.)

The tool walks you through a questionnaire, serving prompts like the deceased’s name, date of birth, date of death, location of death and last city of residence. Some questions are more specific, like “Share any relevant details about the ceremony venue, date and time, or special guidelines,” and pertain to different aspects of the person’s life, like whether they served in the military, what people often said about them, their proudest accomplishments and your favorite memories together.

Many of the questions don’t have to be answered, and responses can range in length from a few words to several paragraphs. Gura says that the flow was modeled on the obituary writing services commonly offered by funeral homes and professional obituary writing companies.

“With Finding Words, Empathy empowers individuals by helping them work on this process themselves — and offers the service for free,” he added. “The tool helps people understand what is typically included in an obituary, and prompts users to consider the sort of details, memories, and anecdotes that are essential in drafting a personalized obituary, ultimately crafting the details inputted into cohesive text.”

Finding Words’ obits might not win awards, but they were better than I expected, frankly (certainly compared to ChatGPT’s attempts). While I kept answers to the prompts relatively nonspecific in my test, the AI managed to craft them into something coherent — if a bit formulaic. (To be fair, most obits are formulaic — to the point that a cursory Google search yields dozens of templates.) If I hadn’t been told, I doubt I’d suspect AI had a hand in the writing process.

Generative AI, including the type of text-generating AI underpinning Finding Words, has a tendency to generate untrue or otherwise problematic text. I didn’t observe any in my testing. But in the interest of thoroughness, I asked Gura what preventative steps Empathy took, if any.

“Finding Words is powered by an AI algorithm trained and refined by Empathy’s team of developers, writers and grief professionals and is based on insights from thousands of sample obituaries … Our AI model has been trained to generate a cohesive outcome that accurately reflects whatever details users input,” Gura said. “We take care to inform users that the text generated by Finding Words is fully automated and advise them to review the text thoroughly in order to verify that all information is correct.”

Will Finding Words make obit-writing services obsolete? I doubt it — those services tend to be more bespoke. But while I was tempted to dismiss it out of hand, I can’t say it wasn’t serviceable in my brief test. (Sorry to be the bearer of bad news, career obit writers.) With some fine-tuning, the results could be quite good, in fact — and definitely on par with some of the templates out there (and Wired’s 2016 AI-written obituary for Marvin Minsky).

Given generative AI’s plagiaristic proclivities, I am wary, though, of how Empathy is training the language algorithm that powers Finding Words. Gura didn’t disclose where the aforementioned sample obits came from, and also didn’t say whether Empathy uses any user data to fine-tune them. (I’ve asked him to clarify.) In any case, whether or not the creators of the training data are being fairly compensated (and properly informed), Empathy — which is venture-backed, with $43 million raised to date — is no doubt under pressure from investors to monetize. I wouldn’t be surprised in the slightest to see a fee attached to Finding Words in the future, at which point the tool will warrant higher scrutiny.

“We trained the algorithms by using hundreds of obituaries previously written by our team of professional writers. Based on the obituaries our team had manually written, we built up an understanding of the relevant questions commonly used when drafting an obituary and used these prompts to develop the custom model for writing first drafts,” Gura said in a follow-up email. “We don’t plan to use data from Finding Words to validate and train our algorithm. However, we will use general feedback from users to iterate on the product and offering as a whole, expanding it to encompass and support more scenarios and situations.”