Generative AI, prompt-based artificial intelligence that can generate text, images, music, and other media in seconds, continues to advance at breakneck speed.
Less than half a year later, OpenAI released its generative AI tool, ChatGPT. Just four months later, the company released his GPT-4. This is a huge leap in processing speed, power and capability.
Every industry is paying attention. Especially healthcare. Observers were a little impressed when the original version of ChatGPT barely passed the US medical licensing exam. A few months later, his Med-PaLM 2 from Google passed the same test and scored in the “expert” category.
Despite growing evidence that generative AI is poised to revolutionize healthcare, patients are reluctant to embrace it. According to her recent Pew Research poll, six of her 10 American adults are “uncomfortable” with doctors relying on artificial intelligence to diagnose illnesses and suggest treatments. He says he feels
Today’s iteration of generative AI is not ready for widespread use in healthcare. They occasionally fail basic calculations, fabricate sources, “hallucinate”, and provide confident but counterfactual answers. The world is watching how quickly OpenAI, Google and Microsoft can fix these errors.
But these amendments alone cannot address the two biggest concerns of patients reported in the Pew survey.
- Technical risks such as security, privacy, and algorithmic bias.
- Ethical concerns about machine-human interaction.
This article examines the first series of fears. Next time on May 8th, we’ll cover ethical issues such as the impact of AI on doctor-patient relationships.
Is Patient Fear Valid?
Americans have long been skeptical of new technology. Remember, bank customers resisted using his ATM in the 1970s. In fact, cashpoint errors were common in the beginning. But once the banks made tweaks and the roots of people’s technology-driven fears stopped crystallizing, the fears themselves faded from consciousness. This process is known as habituation.
When it comes to the use of generative AI in medicine, people’s fears about security, privacy and bias are normal, legitimate and should be taken seriously. But the most important question to consider is, does AI pose greater risks to patients than the technology they already use in their daily lives?
Let’s examine this issue by looking at these technical issues one by one.
1. Security
Banks, government agencies and healthcare companies alike store sensitive information in large databases and are subject to extensive scrutiny. This is especially true now that more and more individual data moves to the cloud, the Internet-based servers on which these large databases run.
Cybersecurity is a major concern in the United States. A survey found that 9 out of 10 Americans worry that hackers will access their personal and financial information and use it for malicious purposes. But when it comes to her ChatGPT in healthcare, patients will want to know if generative AI increases the risk of future cyberattacks.
Contrary to patient expectations, a single doctor’s office is one of the least secure places to store personal medical data. Lone doctor practices lack the financial and technical resources to install state-of-the-art network security tools, making patient data highly vulnerable.
So why don’t cybercriminals hack into individual clinics more often? Because it’s worth it.
Ironically, making medical records more secure within a local clinic is the same as disabling the medical records system. Unless your doctor is part of a large medical group, your health record is not associated with (or available to) other clinics or nearby hospitals.
Government health officials have tried for decades to encourage secure patient record sharing, but today few providers have access to a “comprehensive” medical records system. As a result, even if local clinicians use the same digital records management platform, they do not have access to all the data about each patient they need to provide optimal care.
This is not only ineffective, but also dangerous.
If you arrive at the ER late at night when the doctor’s office is closed, the emergency doctor on staff will not be able to look up your medical history, current prescriptions, recent diagnostic tests, or any other vital information you may need to provide . the best possible care.
As such, security in healthcare is a double-edged sword. As patients, we want our medical data to be safe and off limits to malicious hackers. But it also needs to be comprehensive and readily available so that care can be received at any time of the day or day of the week.
Generative AI applications won’t solve this problem unless EHR manufacturers open up their application programming interfaces (APIs). The important thing is that the creation and operation of AI tools gain Security risks for people too.
Large financial institutions and EHR firms store vast amounts of digitized information behind hard-to-reach firewalls. We can assume that ChatGPT (and generative AI systems developed by Google and others) maintain at least equal protection. It is in their reputation and financial interests to ensure that.
2. Privacy
Large companies work hard to maximize data security, but their business models have long relied on violating user privacy.
A lawmaker once asked Mark Zuckerberg how Facebook could survive as a business without charging user fees. The CEO replied, “By advertising.” Zuckerberg meant his social networks profit from selling users’ personal information to third parties. And for decades, people using social media sites and search engines have effectively traded their personal information to advertisers in exchange for free access.
In medicine, it is illegal to extract and disclose data from medical records. However, it does not guarantee complete patient privacy. Recent news reports have revealed how hospitals and pharmacies engage in online data sharing with third parties without explicit permission from patients.
Additionally, when people search the internet for their symptoms or make online purchases to treat their health problems, that information is being used by companies to target ads. That’s why people receive diaper coupons within days of finding out they’re pregnant.
Patients may face the same privacy risks as generative AI companies. But like security, there is nothing about generative AI that magnifies user privacy risks. Similar to the entire existing social digital media landscape.
3. Prejudice
Unlike the areas of security and privacy, where Pew survey respondents expressed a high level of concern, most patients expected generative AI to be less medically biased in the future (51%). but do not expect it to be higher (15%).
Researchers continue to identify biases in algorithmic tools used in medicine. In some cases, these lopsided IT applications have been shown to exacerbate existing healthcare inequalities – quality-of-care gaps based on gender, race, ethnicity, income, etc.
As such, human concerns about bias in medical algorithms are valid. And the need for solutions remains urgent. But the fact that these errors are rarely generated by 0 and 1 glitches is lost in the headline. The truth is that computers and algorithms are not biased. Humans are. And when computers spit out biased recommendations, they do so because they are trained on human behavior and decisions.
These actions and decisions often reflect implicit bias. For example, in the United States, physicians recommend breast reconstruction less frequently for black patients after mastectomy than for white patients. They prescribe fewer pain medications to black and Hispanic patients than whites after surgery. And early in the pandemic, when black people were twice as likely to die from Covid-19, doctors tested black patients for the virus half as often as white patients with the same symptoms.
Therefore, when the creators of the AI tool input this data into their application, the algorithm assumes that black patients need breast reconstruction less often than whites, do not need pain medications, and do not need tests as often. increase.
Fortunately, generative AI has the potential to reduce the prevalence of bias in healthcare. This is because they include a much broader range of inputs than AI applications developed for specific medical functions (such as reading mammograms or managing diabetes). In contrast to these “narrow AI” applications, generative AI is built to answer an almost infinite number of questions and perform an unlimited number of functions. To facilitate its capabilities, the application has been pre-trained using massive datasets. While this dataset may unintentionally include bias in patient care, it also includes evidence-based research on the existence and dangers of bias in medicine.
So if a doctor “forgets” to suggest breast reconstruction, prescribe appropriate pain medication, or ask a black or Hispanic patient for a necessary test, ChatGPT will question the doctor’s decision. can present In doing so, we will move away from medical stigma and toward a more equitable form of care for all.
Generative AI: Hero or Villain?
I am optimistic that ChatGPT and future generative AI applications will empower patients and transform healthcare for the better. These tools help physicians and patients achieve superior clinical outcomes, make care more accessible, and reduce costs (partially due to medical error, health disparities, and effectiveness by reducing no medical care).
No application can completely avoid or prevent today’s security, privacy and bias issues, but the risks to users are no greater than those already experienced by people.
Finally, as Pew’s survey respondents noted, it’s not just technical concerns that plague the use of AI in medicine. As these tools become increasingly powerful and widely available, they also affect the doctor-patient relationship, raising highly controversial ethical questions. The next article in will focus on these very human issues.
Follow me please twitter or LinkedIn. check out My website or some of my other works can be found here.
