Startups

Hippocratic is building a large language model for healthcare

Comment

illustration of telemedicine and online healthcare services
Image Credits: elenabs (opens in a new window) / Getty Images

AI, specifically generative AI, has the potential to transform healthcare.

At least, that’s the sales pitch from Hippocratic AI, which emerged from stealth today with a whopping $50 million in seed financing behind it and a valuation in the “triple-digit millions.” The tranche, co-led by General Catalyst and Andreessen Horowitz, is a big vote of confidence in Hippocratic’s technology, a text-generating model tuned specifically for healthcare applications.

Hippocratic — hatched out of General Catalyst — was founded by a group of physicians, hospital administrators, Medicare professionals and AI researchers from organizations including Johns Hopkins, Stanford, Google and Nvidia. After co-founder and CEO Munjal Shah sold his previous company, Like.com, a shopping comparison site, to Google in 2010, he spent the better part of the next decade building Hippocratic.

“Hippocratic has created the first safety-focused large language model (LLM) designed specifically for healthcare,” Shah told TechCrunch in an email interview. “The company mission is to develop the safest artificial health general intelligence in order to dramatically improve healthcare accessibility and health outcomes.”

AI in healthcare, historically, has been met with mixed success.

Babylon Health, an AI startup backed by the U.K.’s National Health Service, has found itself under repeated scrutiny for making claims that its disease-diagnosing tech can perform better than doctors. IBM was forced to sell its AI-focused Watson Health division at a loss after technical problems led major customer partnerships to deteriorate. Elsewhere, OpenAI’s GPT-3, the predecessor to GPT-4, urged at least one user to commit suicide.

Shah emphasized that Hippocratic isn’t focused on diagnosing. Rather, he says, the tech — which is consumer-facing — is aimed at use cases like explaining benefits and billing, providing dietary advice and medication reminders, answering pre-op questions, onboarding patients and delivering “negative” test results that indicate nothing’s wrong.

Hippocratic
Hippocratic’s benchmark results on a range of medical exams. Image Credits: Hippocratic

The dietary advice use case gave me pause, I must say, in light of the poor diet-related suggestions AI like OpenAI’s ChatGPT provides. But Shah claims that Hippocratic’s AI outperforms leading language models including GPT-4 and Claude on more than 100 healthcare certifications, including the NCLEX-RN for nursing, the American Board of Urology exam and the registered dietitian exam.

“The language models have to be safe,” Shah said. “That’s why we’re building a model just focused on safety, certifying it with healthcare professionals and partnering closely with the industry … This will help ensure that data retention and privacy policies will be consistent with the current norms of the healthcare industry.”

One of the ways Hippocratic aims to achieve this is by “detecting tone” and “communicating empathy” better than rival tech, Shah says — in part by “building in” good bedside manner (i.e. the elusive “human touch”). He makes the case that bedside manner — especially interactions that leave patients with a sense of hope, even in grim circumstances — can and do affect health outcomes.

To evaluate bedside manner, Hippocratic designed a benchmark to test the model for signs of humanism, if you will — things like “showing empathy” and “taking a personal interest in a patient’s life.” (Whether a single test can accurately capture subjects that nuanced is up for debate, of course.) Unsurprisingly given the source, Hippocratic’s model scored the highest across all categories of the models that Hippocratic tested, including GPT-4.

But can a language model really replace a healthcare worker? Hippocratic invites the question, arguing that its models were trained under the supervision of medical professionals and, thus, are highly capable.

“We’re only releasing each role — dietician, billing agent, genetic counselor, etc. — once the people who actually do that role today in real life agree the model is ready,” Shah said. “In the pandemic, labor costs went up 30% for most health systems, but revenue didn’t. Hence, most health systems in the country are financially struggling. Language models can help them reduce costs by filling their current large level of vacancies in a more cost-effective way.”

I’m not sure healthcare practitioners would agree — particularly considering the Hippocratic model’s low scores on some of the aforementioned certifications. According to Hippocratic, the model got a 71% on the certified professional coder exam, which covers knowledge of medical billing and coding, and 72.7% on a hospital safety training compliance quiz.

There’s the matter of potential bias, as well. Bias plagues the healthcare industry, and these effects trickle down to the models trained on biased medical records, studies and research. A 2019 study, for instance, found that an algorithm many hospitals were using to decide which patients needed care treated Black patients with less sensitivity than white patients.

In any case, one would hope Hippocratic makes it clear that its models aren’t infallible. In domains like healthcare, automation bias or the propensity for people to trust AI over other sources, even if they’re correct, comes with plainly high risks.

Those details are among the many that Hippocratic has yet to iron out. The company isn’t releasing details on its partners or customers, preferring instead to keep the focus on the funding. The model isn’t even available at present — nor information about what data it was trained on, or what data it might be trained on in the future. (Hippocratic would only say that it’ll use “de-identified” data for the model training.)

If it waits too long, Hippocratic runs the risk of falling behind rivals like Truveta and Latent — some of which have a major resource advantage. For example, Google recently began previewing Med-PaLM 2, which it claims was the first language model to perform at an expert level on dozens of medical exam questions. Like Hippocratic’s model, Med-PaLM 2 was evaluated by health professionals on its ability to answer medical questions accurately — and safely.

But Hemant Taneja, the managing director at General Catalyst, didn’t express concern.

“Munjal and I hatched this company on the belief that healthcare needs its own language model built specifically for healthcare applications — one that is fair, unbiased, secure and beneficial to society,” he said via email. “We set forth to create a high-integrity AI application that is fed a ‘healthy’ data diet and includes a training approach that seeks to incorporate extensive human feedback from medical experts for each specialized task. In healthcare, we simply can’t afford to ‘move fast and break things.’”

Shah says that the bulk of the $50 million seed tranche will be put toward investing in talent, compute data and partnerships.

More TechCrunch

Looking Glass makes trippy-looking mixed-reality screens that make things look 3D without the need of special glasses. Today, it launches a pair of new displays, including a 16-inch mode that…

Looking Glass launches new 3D displays

Replacing Sutskever is Jakub Pachocki, OpenAI’s director of research.

Ilya Sutskever, OpenAI co-founder and longtime chief scientist, departs

Intuitive Machines made history when it became the first private company to land a spacecraft on the moon, so it makes sense to adapt that tech for Mars.

Intuitive Machines wants to help NASA return samples from Mars

As Google revamps itself for the AI era, offering AI overviews within its search results, the company is introducing a new way to filter for just text-based links. With the…

Google adds ‘Web’ search filter for showing old-school text links as AI rolls out

Blue Origin’s New Shepard rocket will take a crew to suborbital space for the first time in nearly two years later this month, the company announced on Tuesday.  The NS-25…

Blue Origin to resume crewed New Shepard launches on May 19

This will enable developers to use the on-device model to power their own AI features.

Google is building its Gemini Nano AI model into Chrome on the desktop

It ran 110 minutes, but Google managed to reference AI a whopping 121 times during Google I/O 2024 (by its own count). CEO Sundar Pichai referenced the figure to wrap…

Google mentioned ‘AI’ 120+ times during its I/O keynote

Firebase Genkit is an open source framework that enables developers to quickly build AI into new and existing applications.

Google launches Firebase Genkit, a new open source framework for building AI-powered apps

In the coming months, Google says it will open up the Gemini Nano model to more developers.

Patreon and Grammarly are already experimenting with Gemini Nano, says Google

As part of the update, Reddit also launched a dedicated AMA tab within the web post composer.

Reddit introduces new tools for ‘Ask Me Anything,’ its Q&A feature

Here are quick hits of the biggest news from the keynote as they are announced.

Google I/O 2024: Here’s everything Google just announced

LearnLM is already powering features across Google products, including in YouTube, Google’s Gemini apps, Google Search and Google Classroom.

LearnLM is Google’s new family of AI models for education

The official launch comes almost a year after YouTube began experimenting with AI-generated quizzes on its mobile app. 

Google is bringing AI-generated quizzes to academic videos on YouTube

Around 550 employees across autonomous vehicle company Motional have been laid off, according to information taken from WARN notice filings and sources at the company.  Earlier this week, TechCrunch reported…

Motional cut about 550 employees, around 40%, in recent restructuring, sources say

The keynote kicks off at 10 a.m. PT on Tuesday and will offer glimpses into the latest versions of Android, Wear OS and Android TV.

Google I/O 2024: Watch all of the AI, Android reveals

Google Play has a new discovery feature for apps, new ways to acquire users, updates to Play Points, and other enhancements to developer-facing tools.

Google Play preps a new full-screen app discovery feature and adds more developer tools

Soon, Android users will be able to drag and drop AI-generated images directly into their Gmail, Google Messages and other apps.

Gemini on Android becomes more capable and works with Gmail, Messages, YouTube and more

Veo can capture different visual and cinematic styles, including shots of landscapes and timelapses, and make edits and adjustments to already-generated footage.

Google Veo, a serious swing at AI-generated video, debuts at Google I/O 2024

In addition to the body of the emails themselves, the feature will also be able to analyze attachments, like PDFs.

Gemini comes to Gmail to summarize, draft emails, and more

The summaries are created based on Gemini’s analysis of insights from Google Maps’ community of more than 300 million contributors.

Google is bringing Gemini capabilities to Google Maps Platform

Google says that over 100,000 developers already tried the service.

Project IDX, Google’s next-gen IDE, is now in open beta

The system effectively listens for “conversation patterns commonly associated with scams” in-real time. 

Google will use Gemini to detect scams during calls

The standard Gemma models were only available in 2 billion and 7 billion parameter versions, making this quite a step up.

Google announces Gemma 2, a 27B-parameter version of its open model, launching in June

This is a great example of a company using generative AI to open its software to more users.

Google TalkBack will use Gemini to describe images for blind people

Google’s Circle to Search feature will now be able to solve more complex problems across psychics and math word problems. 

Circle to Search is now a better homework helper

People can now search using a video they upload combined with a text query to get an AI overview of the answers they need.

Google experiments with using video to search, thanks to Gemini AI

A search results page based on generative AI as its ranking mechanism will have wide-reaching consequences for online publishers.

Google will soon start using GenAI to organize some search results pages

Google has built a custom Gemini model for search to combine real-time information, Google’s ranking, long context and multimodal features.

Google is adding more AI to its search results

At its Google I/O developer conference, Google on Tuesday announced the next generation of its Tensor Processing Units (TPU) AI chips.

Google’s next-gen TPUs promise a 4.7x performance boost

Google is upgrading Gemini, its AI-powered chatbot, with features aimed at making the experience more ambient and contextually useful.

Google’s Gemini updates: How Project Astra is powering some of I/O’s big reveals