“Trust the metal head” Dr House advicing patient to trust a computer based solution
Introduction
Computational thinking is computers doing human thinking load [1]. Thus, if you cannot make the difference between a computer solution from a human’s, this is computational thinking. As artificial intelligence evolves, we may improve this definition, but for now it is enough in layman terms. Mathematical models applied to biology would never have been possible if it were not for computers. Computers make it possible to consider nonlinear models. Artificial intelligence is even more important: one can consider decisions models, called by Kahneman and colleagues [2] as mechanical judgment; in contrast to clinical judgement. Furthermore, they showed that mechanical judgement can overcome humans even when the models is a simple regression model. That leads us to consider seriously AI decision models; or any variation such as chatGPT (a language model).
ChatGPT (Chat Generative Pre-trained Transformer) [3] is a language model trained using deep learning, to replicate the human’s language abilities. Different from other chatbots, it is a general-purpose language [1]. It can be used on unknown texts since it learns patterns, not specific information. It has been calling attention for the texts it creates, very close to human’s; it can also code [4]. Recently, chatSonic[1] claims to have improved the flaws from chatGPT, e.g., its training dataset stops at 2021. Imagine what we could do if we use the same technology focused on mathematical models applied to medicine. What takes four years of a PhD to create, those bots could create in a blink.
On this short essay, we shall talk about chatGPT and medicine. Would chatGPT be a problem or a companion to medical doctors? What about bioinformatics researchers, could they count on chatGPT for a hand?
See that we shall use the name chatGPT, but consider as synonymous of what may come next, since in the tech world, things chance so fast that we cannot even breathe [5].
A brief look at the literature
ChatGPT passes exam for law and business school [6]. This shows that even human-knowledge related to academic degrees is possible to automate. On the case of a medical exam, it gets close [9]. This is impressive anyway. This is an exam that “usually requires some 300 to 400 hours of preparation” [9, based on 7], this is not to mention the training time of a medical doctor. It took others to question the medical education system [16]; I have been questioning the engineering education system, math in general. My guess is that as chatGPT and alike gain momentum, we will need to review what is important to teach at universities, or even, require from people on their daily-work life. Once people needed to make calculations from head, once we created calculators and computers, no one talks about it anymore as something to endure on professionals. What may happen is what Kahneman called attention to [2]: even for models much less advanced, people are generally resistant, even been proven to be better than humans. See the case of automatically-driven cars: even though we may hear of accidents [8], statistically speaking, humans kill much more daily at traffic. Once I asked what is really important on the engineering formation for possibly creating an AI: it seems, it is not that hard to create an artificial intelligence-based engineer.
ChatGPT may release scientists from boring-writing tasks [9]. I have already somehow guessed that even before chatGPT, since other models less successfully already tried that, with less success. Writing scientific papers is a mechanical and repetitive task.
In a survey on Nature readers, about 80% of their readers have used chatbots [10], and about 50% use them for “creative fun”. I see as a sign that even though we may have local resistance, those chatbots may enter medical doctors daily-lives: it can be used to start say a diagnose, especially if it is rare. We know how hard it can be to come up with ideas on pressure, and several professionals use formally or informally brainstorm sections.
The good news, as I see, we have discussions going on regarding its place on scientific research [11–14], which could easily contaminate also its role on medical practices. One question I have heard on another context: if a algorithm make a mistake, would the medical doctor be responsible? I guess we need to consider that from a law perspective.
Discussion
It is very easy to be deceived by new achievements in artificial intelligence, especially if you are not from the field. They look like “magic” when released. Many people that use computational models are not from mathematics or even computer science: openAI even has that as their aim, AI for nonexperts. What we should keep in mind: all the endeavors related to mathematical biology will continue. We still need virtual pancreas and humans; white-box models are still imperative.
What may happen, and I really hope, it will be easier to create innovative models in computational biology, see [17]. Moreover, it may even be easier to create equations from human-like language. What we do as modelers is to translate scientific articles into equations, and chatGPT can summarize texts, quite well, it may be the onset of creating equations. It already has the ability to create codes from human’s commands.
As I see it: chatGPT could help to make mathematical modeling more accessible for non-mathematics experts. It already has mathematical abilities: it can solve integrals. With some extra work, even easier than we think since it is a general-purpose, it can replicate what a PhD student must do for creating a model, as I did, and getting a PhD. Mathematical biology already gained a lot when tools such as Matlab started to gain popularity, others such as Simulink. Now imagine if we could create mathematical models just with words-lines of commands. Innovating with biomathematics [17] will become even more accessible. At the current state, it can create codes even advanced from word-like commands.
Artificial intelligence already can do quite advanced assistance to mathematicians [18]. Thus, using chatGPT to make the interaction with humans does not seem that hard. Remember, the goal is advance even more the usage of models by life scientists.
The imperative question: can it replace medical doctors?
First of all, apps for mimicking medical doctors are not new [19]. In some apps, you answer a set of questions, and it can make a diagnosis.
I do not think that it will replace humans, not on a foreseeable future. However, it may work alongside. For instance, you can ask for information on a medical condition, and use the information to brainstorm, as a starting point. If we remove the possible human’s pride, this is a quite powerful tool! Already helping people to be creative.
References
1. Pires, JG (2023). Computational Thinking: How computers think, decide and learn, when human limits start and computers champ, vol. 1. https://www.ideacodinglab.com/computationalthinking
2. Daniel Kahneman, Olivier Sibony, Cass R Sunstein. Noise: A Flaw in Human Judgment. Little, Brown Spark (18 maio 2021)
3. Wikipedia contributors. (2023, February 20). ChatGPT. In Wikipedia, The Free Encyclopedia. Retrieved 19:03, February 21, 2023, from https://en.wikipedia.org/w/index.php?title=ChatGPT&oldid=1140590947
4. Pires, JG. (2023). Just asked chatGPT to code in Angular Angular Material, and Velo (Wix). Published in Computational Thinking: How computers think, decide and learn. https://medium.com/computational-thinking-how-computers-think-decide/just-asked-chatgpt-to-code-in-angular-and-angular-material-6566936f6e19
5. Peter H Diamandis, Steven Kotler. The Future Is Faster Than You Think: How Converging Technologies Are Transforming Business, Industries, and Our Lives. Simon & Schuster; 1ª edição (28 janeiro 2020)
6. Samantha Murphy Kelly,. ChatGPT passes exams from law and business schools. https://edition.cnn.com/2023/01/26/tech/chatgpt-passes-exams/index.html
7. Kung et al (2023). Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models. February 9, 2023 https://doi.org/10.1371/journal.pdig.0000198
8. THE ASSOCIATED PRESS. Nearly 400 car crashes in 11 months involved automated tech, companies tell regulators. https://www.npr.org/2022/06/15/1105252793/nearly-400-car-crashes-in-11-months-involved-automated-tech-companies-tell-regul
9. John Tregoning. AI writing tools could hand scientists the ‘gift of time’. CAREER COLUMN 22 February 2023. https://www.nature.com/articles/d41586-023-00528-w
10. Brian Owens. How Nature readers are using ChatGPT. NEWS 20 February 2023. https://www.nature.com/articles/d41586-023-00500-8
11. Victoria Corless. ChatGPT is making waves in the scientific literature. Advanced Science News. https://www.advancedsciencenews.com/where-and-how-should-chatgpt-be-used-in-the-scientific-literature/
12. Chris Stokel-Walker. ChatGPT listed as author on research papers: many scientists disapprove. NEWS 18 January 2023. https://www.nature.com/articles/d41586-023-00107-z
13. Holly Else. Abstracts written by ChatGPT fool scientists. NEWS 12 January 2023. https://www.nature.com/articles/d41586-023-00056-7
14. Brian Lucey; Michael Dowling. ChatGPT: our study shows AI can produce academic papers good enough for journals — just as some ban it. https://theconversation.com/chatgpt-our-study-shows-ai-can-produce-academic-papers-good-enough-for-journals-just-as-some-ban-it-197762
15. DAVID NIELD. ChatGPT Can Almost Pass The US Medical Licensing Exam. https://www.sciencealert.com/chatgpt-can-almost-pass-the-us-medical-licensing-exam
16. Mbakwe et al . ChatGPT passing USMLE shines a spotlight on the flaws of medical education. February 9, 2023 https://doi.org/10.1371/journal.pdig.0000205
17. Pires, J.G. (2022). Innovating with Biomathematics: the challenge of building user-friendly interfaces for computational biology. Academia Letters, Article 5792. https://doi.org/10.20935/AL5792
18. Castelvecchi, Davide. How will AI change mathematics? Rise of chatbots highlights discussion. NEWS 17 February 2023. https://www.nature.com/articles/d41586-023-00487-2
19. Pires JG. Alguns insights em Startups um novo paradigma para a tríplice aliança ciência, tecnologia e inovação: a novel paradigm for understanding the triple alliance of Science, Technology and Innovation. Rev. G&S [Internet]. 1º de fevereiro de 2020 [citado 22º de fevereiro de 2023];11(1):38–54. Disponível em: https://periodicos.unb.br/index.php/rgs/article/view/28626
[1] https://writesonic.com/chat