Billionaire tech mogul Elon Musk took to his social media platform X last week to ask users of his artificial intelligence chatbot Grok to upload medical images and generate diagnoses.
“Why not send your X-ray, PET, MRI, or other medical images to Grok for analysis,” Musk suggested on Oct. 29, announcing the X platform, which gives subscribers access to AI. Suggested.
Uploading medical data through social media platforms raises privacy questions.
Some in the medical industry are already testing large-scale language models (controversial for mining social media data as part of their training) and have provided mixed reviews regarding their accuracy.
Meanwhile, Grok had already urged European privacy regulators to question possible violations of the General Data Protection Regulation.
There are pros and cons regarding accuracy.
“This is still in its early stages, but it's already pretty accurate and will be very good,” Musk declared in a post on X.
Dr. Deliya Unutmaz, who studies human T-cell mechanisms at the Jackson Laboratory, a nonprofit scientific research institute in Connecticut, was one of the many people who examined Glok with X-rays last week and found that adjustments needed to be made. He pointed out. Prompt asking for correct answer:
“I used this prompt, and this time Grok answered correctly: 'You are an emergency physician. This patient was brought to the emergency room complaining of shortness of breath. There were no obvious injuries. What is the diagnosis? ?’ for my next x-ray,” he replied to Musk.
“Apparently it needed some context,” he added. Anyway, I was very impressed with how polite the response was!
Dr. Laura Heacock, a breast radiologist, deep learning researcher, and associate professor at New York University's School of Imaging, was less impressed, according to her response.
Mr. Heacock used the same breast mammograms, ultrasounds and MRIs previously used to test GPT4 as benchmarks, and in a series of 12 social media posts provided Mr. Glok's answers and her opinions on those answers. Explained in detail.
“So how did Mr. Grok do when it came to breast radiology? He did a little better than GPT4v, but he didn't get a single correct diagnosis,” she said, considering Mr. Musk's statements about accuracy. He then pointed out that he had expected much better performance.
Grok was able to identify the image as a mammogram or ultrasound, but answered Heacock's short questions with the wrong body part or potentially erroneous findings. On a mammogram with a known malignant tumor, Grok suggested that what Heacock called “an apparent large left-sided squamous carcinoma” could be calcifications.
“Nothing,” she said.
Heacock said he would try Grok again in the coming months, but said, “For now, non-generative AI methods continue to be superior in medical image processing.”
Social data in model training
xAI began laying the foundations for the “Counter-Awakening” Grok-1 AI model in November 2023.
“Grok is designed to answer questions with a bit of wit and has a rebellious streak, so don't use it if you don't like humor,” xAI said in an announcement last year.
“Grok's unique and fundamental advantage is that it has real-time knowledge of the world through the X Platform, and can answer difficult questions that most other AI systems are denied.” the company added.
On March 11, Musk announced on his social media platforms that xAI would be making the Grok algorithm, a direct competitor to ChatGPT, accessible as open source software.
The company then announced in May that it had raised $6 billion after releasing the Grok-1.5 model, which has improved reasoning and problem-solving capabilities, and the Grok-1.5V, which added the ability to process visual information. . Includes documents, diagrams, charts, screenshots, and photos.
Dr. Sai Balasubramanian, who was appointed Director of Public Health Informatics Data and Exchange at the Texas Department of State Health Services earlier this year, recognized the potential to disrupt healthcare and incorporated Grok into daily care as a foundational model. He told Forbes that it could be done. Workflow for radiology treatment, etc.
Like other AI models, Grok can be used to increase task automation by improving clinical workflow and productivity, but it is likely to face competition from industry veterans in the enterprise space and balance Bramanian said.
However, the head of public health informatics pointed out in the article that there is a huge opportunity for Musk to combine xAI with his medical device company Neuralink.
In 2020, Musk's Neuralink demonstrated brain computer implants in pigs. In September, the neurological technology startup announced that its Blindsight implant, which aims to restore vision, had received breakthrough device designation from the U.S. Food and Drug Administration.
The CEO of Tesla and SpaceX once joined the movement to stop what he called the “AI apocalypse” and spoke with Facebook CEO Mark Zuckerberg about its potentially destructive power. We have exchanged words.
The latter criticized Musk's ideas about the doom of AI and was optimistic about its potential for enhancing diagnostic capabilities and innovation in drug discovery. Musk has previously said that an AI-controlled future could reduce humans to “pets”, but countered that Zuckerberg's “understanding of this issue is limited”. .
Fighting privacy regulations
Currently, X Premium and Premium+ users have access to two new models, Grok-2 and Grok-2 mini, which integrate real-time information from X social media platforms.
The Grok-2 beta release in August allowed users to create images (which, as researchers were able to prove this summer, were often highly inappropriate due to a lack of filters) from text prompts, Obtained the ability to publish to social platforms. And by connecting real-time data from X, Grok will be able to respond to prompts about current events as they unfold.
Government watchdogs responded by citing the laxity and general disdain for safe AI and social media standards that permeate Grok. In addition to the European Union's requisition for potential violations of digital safety laws, nine states have sent letters regarding misinformation about the 2024 election created using the platform.
In an Aug. 5 letter, Minnesota Secretary of State Steve Simon wrote, “We want to ensure that all voters who use your platform receive guidance that reflects true and accurate information about their constitutional right to vote. “We have a responsibility to ensure access.”
“We urge X to immediately adopt a policy that directs Grok users to CanIVote.org when asked about the US elections.”
When to collect data and apply HIPAA?
At least one privacy expert says HIPAA-covered entities' participation in efforts to improve the accuracy of Grok's medical data via social media could compromise patient privacy in the United States. expressed concern that there is.
European government leaders have also previously questioned developer xAI over possible GDPR violations. On August 6, the Irish Data Protection Commission asked the Irish High Court to stop the social media network from collecting account holders' data to train Grok LLM.
However, if a user asks Grok something, it will not be automatically posted to the X social media platform. For now, users must post directly, said Lee Kim, senior principal of cybersecurity and privacy at Healthcare IT's parent company HIMSS. news.
“Currently can be toggled on and off – “We'll show you not only your posts, but also your interactions with Grok, inputs, and results for use in training and fine-tuning. You can also choose to delete your conversation history,” Kim wrote on Monday with a screenshot. He said while sharing.
The company criticized DPC's efforts to lock out Grok on social media platforms, calling it “unfair, broad and identifying ” he swore.
In August, the company said, “Unlike the rest of the AI industry, we are giving all X users a simple “We decided to provide control.”
Users can make some choices according to xAI's privacy policy, but “users need to be aware of how their data will be used and treated,” Kim said. She explained that HIPAA can apply to data shared on social media platforms if a covered entity discloses protected patient information.
“It all depends on who discloses the information,” Kim said. “If you are a covered entity, business partner, or organized health care arrangement, HIPAA applies.”
Andrea Fox is a senior editor at Healthcare IT News.
Email: afox@himss.org
Healthcare IT News is a publication of HIMSS Media.
