Association for Academic Surgery (AAS)

Inspiring and Developing Young Academic Surgeons

What is your P(doom)?

by

Living in the Bay Area, it’s difficult to get through a day without hearing the letter “A” followed by “I”. Driving on the Highway 101, there is a billboard every half mile touting some sort of AI-enable something. Undoubtably, AI has changed your life, whether you know it or not. As AI becomes ever more powerful, the concept of AI safety has become an important topic. Amongst the experts and pundits, the concept of P(doom) has emerged in a tongue-in-cheek fashion. According to a peer-reviewed source¹, P(doom) refers to the probability that AI would lead to a catastrophic outcome (think Skynet). The concept of P(doom), and the value that people assign to it, is quite fascinating. From what I can tell, the most common P(doom) in the tech community is between 0.05 to 0.15. The actual value you assign to it probably depends on your level of risk acceptance (and also probably whether you have money in AI. 

We can also apply the concept of P(doom) on a smaller scale. How will AI and P(doom) affect your life as a clinician? Current AI models are best at predicting the next word in a sentence – that’s the basis for ChatGPT. If you’re not using ChatGPT, you should start (this blog is being written with the help of ChatGPT – I unfortunately have no stock in OpenAI). Predictive text, generating form letters, replying to patients – these are all things that are being used in hospitals across the world. We just sat through a presentation for an AI scribe. It was very impressive – P(doom for my human scribe) = 0.9. Soon, AI “agents” will be able to carry out tasks for you. Late night in the OR? Your AI “agent” will write the op note for you and order you a burrito (with extra guac, just the way you like it) and have it delivered to you by the time you get home. Is that cool or creepy? I guess that depends on your P(doom).  

But AI will get even better, and fast. We’ve been trying to build machine learning models for decades now to predict sepsis or an anastomotic leak. This will undoubtably come – P(doom for clinical acumen) = 0.4. We live in the information age – we live and breathe data. So does AI. Say goodbye to manual chart review – P(doom for my medical student doing manual chart review) = 0.9. Care coordination? Not only is AI more empathetic to patients², it also doesn’t miss the fact that Ms. Jones forgot to schedule her last surveillance scan – P(doom for my nurse coordinator) = 0.8. Once Ms. Jones gets that CT scan, the radiologist will soon lean on AI to find that small nodule that wasn’t there a year ago³ – P(doom for my radiologist colleagues) = 0.5. And once that nodule is biopsied, AI will suggest to my pathologist that it’s a metastasis⁴ – P(doom for my pathologist colleagues) = 0.5. 

But surgeons are safe right? AI is not going to be cutting that metastasis out, at least not during my career! That’s probably true, but it’s human nature to believe in self-preservation (or be in denial, I’m not psychologist [P(doom for my psychologist colleagues)=0.4]). Here’s something to consider – billions of dollars are being invested in AI-enhanced computational biology and drug discovery⁵. The natural history of surgery is that we do less and less surgery. There is no reason to believe that AI will not speed up this trend. Ms. Jones with the new lung metastasis will not need surgery at all – she will just get an oral medication developed by AI with a 100% response rate. AI will not replace surgeons; it will just make us less relevant. P(doom for me) = too personal to guess. 

So, what can we do about it? Since the release of ChatGPT, there has been a lot of debate on which jobs are safe and which jobs are P(doomed). I am firmly in the camp of “AI will not take your job, someone who understands AI will take your job”. The key, in my opinion, will be to keep up with the advances in AI, continue to make yourself relevant in this ever-evolving world of AI, and diversify your skillset. Is that good advice? I don’t know. I do know it’s vague enough that 20 years from now, it will probably still be applicable. 

References

  1. P(doom). Wikipedia. Accessed December 30, 2024. https://en.wikipedia.org/wiki/P(doom)
  2. Ayers JW, Poliak A, Dredze M, et al. Comparing Physician and Artificial Intelligence Chatbot Responses to Patient Questions Posted to a Public Social Media Forum. JAMA Intern Med. 2023;183(6):589-596. doi:10.1001/jamainternmed.2023.1838
  3. Cho HS, Hwang EJ, Yi J, Choi B, Park CM. Artificial intelligence system for identification of overlooked lung metastasis in abdominopelvic computed tomography scans of patients with malignancy. Diagnostic and Interventional Radiology. Published online September 9, 2024. doi:10.4274/dir.2024.242835
  4. Greeley C, Holder L, Nilsson EE, Skinner MK. Scalable deep learning artificial intelligence histopathology slide analysis and validation. Sci Rep. 2024;14(1):26748. doi:10.1038/s41598-024-76807-x
  5. Artificial Intelligence for Drug Discovery. Deep Pharma Intelligence. Accessed December 30, 2024. https://www.deep-pharma.tech/ai-in-dd-q1-2023-subscribe

 

 

Beiqun Zhao

Beiqun Zhao

Beiqun Mark Zhao, MD MAS is an Assistant Professor of Surgery at the UCSF. He is board-certified in general and colorectal surgery. He is also board-certified in clinical informatics. His research focuses on the intersection of informatics and implementation science. When he's not thinking about academic surgery, he spends time with his family and cheers on Arsenal FC.
Beiqun Zhao

Latest posts by Beiqun Zhao (see all)

Category:

About Beiqun Zhao

Beiqun Mark Zhao, MD MAS is an Assistant Professor of Surgery at the UCSF. He is board-certified in general and colorectal surgery. He is also board-certified in clinical informatics. His research focuses on the intersection of informatics and implementation science. When he’s not thinking about academic surgery, he spends time with his family and cheers on Arsenal FC.