Perhaps one of the most astonishing breakthroughs of 2022, when it comes to AI, was the release of the ChatGPT-3 platform launched by the OpenAI company last November. The company already secured a multi-billion-pound deal with Microsoft and generated over 1 million subscribers in less than 5 days. For comparison, it took Netflix 3.5 years to get as many subscribers, 10 months for Facebook, and 5 months for Spotify.
The applications of ChatGPT-3 are difficult to overestimate. It is probably sufficient to say that according to some reports, it managed to pass the US medical licensing exam.
For the purpose of this article, we will not be solely focusing on the company itself but will also look at the underlying technology that they use, called Natural Language Processing (NLP). If some of you have not heard about NLP, then you might have heard of AI-powered chatbots, for example. In fact, NLP is far more common than someone might think. Every time you use a Google search engine, it uses NLP to understand the intent behind your queries, while Facebook uses NLP to understand the context of the posts in the newsfeed and provide users with a more relevant and personalized experience.
So why is this such a significant milestone, and how is this relevant to the field of medicine? Well, I thought the most logical answer would be to ask the ChatGPT-3 itself. Yes, you’ve read it right! I asked ChatGPT-3 to explain the utility of NLP in medicine.
You can see the generated answer below. The algorithm has not just provided an answer but was also capable of including relevant references, adhere to a predefined word limit, and ensure that the text was original, i.e., not available in the same format from another source.
The exact request sounded as: ‘Write a professional article about the utility of natural language processing in medicine. It should contain relevant references and be around 300 words.’
I kept the generated text unchanged:
Natural language processing (NLP) is a rapidly growing field in healthcare technology and has the potential to revolutionise the way medical professionals process and analyse vast amounts of data. This technology helps in processing and understanding human language and speech, which can then be utilised for various purposes, including medical diagnosis and treatment.
One of the primary applications of NLP in medicine is to help doctors analyse vast amounts of medical records and identify trends and patterns that may help in the early detection and treatment of diseases. For instance, NLP can be used to analyse electronic health records (EHRs) and identify patients who may be at risk of developing certain conditions, such as diabetes or heart disease. This technology can also be used to identify patients who may benefit from early intervention, such as lifestyle changes, or those who may need to be referred to specialists.
Furthermore, NLP has been employed in the field of clinical trials to enhance the quality of data gathered and simplify analysis. This technology can assist in identifying and validating critical data, such as patient symptoms and demographics, that are essential in determining the effectiveness of new treatments. Additionally, NLP can enhance the efficiency of clinical trials by automating the data collection process, reducing the time required for manual data entry, and increasing accuracy.
In conclusion, NLP has the potential to transform the way medical professionals process and analyze vast amounts of data, from electronic health records to clinical trials. This technology can help to improve the precision and efficiency of medical diagnoses and treatments, increase the speed of clinical trials, and help identify new treatments and cures. With the increasing interest in NLP, it is only a matter of time before this technology becomes an indispensable tool in the field of medicine.
References:
Adzhubei, I., Sunyaev, S. (2017) Natural Language Processing in Medicine. Frontiers in Genetics, 8:222.
Agrawal, A. et al. (2017) Natural Language Processing for Electronic Health Records: A Review. Journal of the American Medical Informatics Association, 24(1), 13-24.
Shickel, B. et al. (2019) Natural Language Processing in Clinical Trials. Journal of Medical Systems, 43(4), 270.ChatGPT-3
Considering that it took less than one minute to generate such a comprehensive response, it is hardly surprising that ChatGPT-3 gained popularity within a short timeframe. Bloggers started using it to write articles, school pupils to help with their homework, and businesses to assist their customers. The OpenAI developers claim that their platform generates original content. What it means in practice is that an algorithm extracts data from multiple sources it has been trained on (books, articles, Twitter feeds, forums, etc.) and presents it in a written format not available from another individual source. Someone might argue that AI cannot replicate ‘original human thinking’ since an algorithm is only as good as the data it was trained on. Perhaps the counter argument to this would be – how much of the available or published information is genuinely original? Another way of looking at this dilemma is to consider how much ‘original thinking’ is required in our day-to-day lives. This can provide some idea of the proportion of tasks that could be automated using natural language processing technology.
The fact that ChatGPT-3 produces ‘unique’ content makes it challenging to distinguish whether the article is genuine or not. Fortunately, developers have already released software capable of distinguishing between text written by a human and text written by AI. The utility of this software is limited, however, since it is accurate in only 26% of cases.
Writing a subject specific article is only a small example of what the ChatGPT-3 platform is capable of. It can also answer various questions, solve complex tasks, generate customized computer code, translate coding languages, and much more. Here is a list of the primary tasks that GPT-3 can do:
- Text generation
- Summarisation
- Open domain question answering
- Paraphrasing
- Sentiment analysis
- Dataset generation
- Code generation
- Programming language conversion
So, why is the deployment of ChatGPT-3 and similar platforms a significant milestone? My personal view, is that the amount of accessible knowledge is growing exponentially and we need the tools to compartmentalise and effectively extract this information. According to the ‘Challenges and opportunities facing medical education’ article published in 2011, medical knowledge was doubling every 50 years in 1950, every 7 years in 1980 and 3,5 years in 2010. It was projected to double every 72 days in 2020 and it is safe to assume that things will only take a faster pace in the future. Could artificial intelligence, and particularly NLP, help solving this problem? I think it certainly has the potential for it!
At the time of writing this article OpenAI announced the release of a beta version of GPT-4 which has 100 Trillion machine learning parameters (compared to 175 Billion for GPT-3 version). It will be capable of processing multiple types of data including videos, images, sounds and numbers and you might be able to generate a fully illustrated 200 pages book in one day.
Looking at the historical evidence of innovative technologies, it is quite possible that ChatGPT will not necessarily be a leading company in this field in the future. Facebook, for example, was not the first or arguably not the best social network at the time, but it has the most number of active users out of all social media platforms. It is always worth paying more attention to the readiness of underlying technology rather than a company itself. Right timing is certainly a key factor for any innovation and OpenAI is currently a leading company in developing language models.
Hopefully this article has shed some light on what the future of medicine might entail. If you wish to stay informed about advancements in healthcare technology, please subscribe to our newsletters. If you have any inquiries, don’t hesitate to reach out to us.
Disclaimer: The above article regarding ChatGPT and natural language processing (NLP) is not intended to promote or recommend any particular company or product. ChatGPT is only used as an example of how NLP technologies have evolved over time. The opinions and views expressed in this article are solely those of the author and do not reflect the views or opinions of any particular company or organisation. The information contained in this article is for educational and informational purposes only.