AANS Neurosurgeon | Volume 28, Number 2, 2019

Advertisement

Healthcare at Your Fingertips

Print Friendly, PDF & Email

Many aspects of artificial intelligence (AI) have the potential to impact medicine. For example, AI making medical decisions is thought to be unattainable in the near future. Given today’s access to “big data,” opportunities to extract new knowledge is probable, making AI medical decisions increasingly probable. Hundreds of terabytes (1012 bytes) are collected by NASA every hour from the universe,1 while zettabyte (1021 bytes) is the volume of healthcare data that is expected to be collected. Making sense of this enormous data becomes a challenge when using traditional logical programming. This is where machine learning, a type of AI, can shine.

Machine Learning Turns it Up

Machine learning (ML) is a branch of AI that focuses on the concept of learning with certainty.2 By learning from labeled observations, the algorithm can recognize unseen variable-variable relationships. ML excels in:

  • Image recognition;
  • Natural language processing;
  • Voice recognition; and
  • Many other tasks.

All the previous algorithms provided opportunities for better human-machine interactions through creating smart applications and embedding them into devices; thus, providing a symbiotic Human-AI interaction. ML turns up the heat, because now the smart-devices can listen, see, understand and appropriately respond. The applications’ responses can thus be tailored based on our collected human domain knowledge. By way of poignant example, ML impressively delivered within the field of game theory. The ancient game of Go is a contest of wisdom every Chinese scholar was expected to master. An AI winning against an advanced player was thought to be impossible; yet, AlphaGo beat the champion, Ke Jie. Soon after, AlphaGo Zero, which learned without human knowledge, won over the new champion AlphaGo 100:0.3,4 It seems the potential of ML is infinite.

The essential combination of significant improvement in computation power and hardware development, along with the rich environment of personal devices, provides a rich environment for ML algorithms to run and advance. The abundance of sensors (including audio, visual and touch) and actuators in personal devices provide opportunities for even stronger Human-AI interaction. Now, ML algorithms can run inside personal devices independent from the internet, while relying solely on device hardware.  

ChatBots Emerge

Unsurprisingly, natural language processing appears as a quintessential component of the Human-AI interaction. This leads to the emergence of general domain and domain-specific chatbots (chat robot). Chatbot is a conversation robot, which relies on the voice recognition and natural language capabilities of ML.5 A chatbot utilizes a specific type of ML called sequence-to-sequence ML. The words to the chatbot are a series of letters or words where patterns can be recognized. Much like a teenager just awakened from sleep, when that meaningful connection happens, it will start generating sensible messaging.

Logically, with their strong infrastructure, a chatbot can live in a personal device. While chatbots can still run on the web, personal devices provide an optimal environment for users. The few medical chatbots developed in the last few years are mostly centered around triaging systems and making simple diagnoses.6 While these technologies have not yet impacted the neurosurgeon directly, it is likely they will soon, due to the ongoing development of ML algorithms. If an AI can recognize sequences of the medical literature phrasing and become able to communicate with humans through a chatbot, then it is likely that we can develop a general domain “PhysicianBot.” For neurosurgery, one can imagine a near future when a chatbot could radically and positively impact:

  1. Patient compliance with medications, treatment regimens, office appointments and pre-operative measures.
  2. Answering patient questions about a wide spectrum of pre- and post-operative care, issues and possible complications.
  3. Provision of comprehensive educational material, including informed consent, in an interactive, rather than static, format.

There are many imaginable and tangible positives from such systems. Think about a wide-awake and happy voice answering questions about constipation after spine surgery! Or, consider few or no surgical cancellations because of patient misunderstanding. The possibilities really are endless.

ML and the Human Spirit

Despite many fears, we are not embarking on a world such as that described poignantly by George Orwell in 1984. Despite how “smart” an AI can become, the compassion that a human carries in their daily interactions will remain far beyond reach by any form of AI. This interaction is what drives us as physicians to continuously check imaging and labs on critical patients. This compassion is what underscores a great physician and will always be lacking in the above technologies. In an interview between two AI experts, Dr. Sebastian Thrun and Dr. Kai-Fu Lee, both expressed the difficulties with the existence of general domain AIs. They also outlined the innate desire of humans to interact with humans, as opposed to robots. Despite all the technologies offered by “Big Data”, ML and personal devices, human emotions and innate human responses will always be an essential partner in adapting a “PhysicianBot.”

 

References

1. NASA. (2013, October 17). Managing the Deluge of ‘Big Data’ From Space. Retrieved from https://www.jpl.nasa.gov/news/news.php?release=2013-299

2. Goodfellow, I., Bengio, Y., & Courville, A. (2017). Deep learning. Cambridge, MA: The MIT Press.

3. Silver, D., Schrittwieser, J., Simonyan, K., Antonoglou, I., Huang, A., Guez, A., . . . Hassabis, D. (2017). Mastering the game of Go without human knowledge. Nature, 550(7676), 354-359.

4. Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., Driessche, G. V., . . . Hassabis, D. (2016). Mastering the game of Go with deep neural networks and tree search. Nature, 529(7587), 484-489.

5. Nguyen, H., Morales, D., and Chin, T. (2017). A Neural Chatbot with Personality. Retrieved from https://web.stanford.edu/class/cs224n/reports/2761115.pdf

6. Sennaar K. (2018). Chatbots for Healthcare – Comparing 5 Current Applications. Retrieved from https://www.techemergence.com/chatbots-for-healthcare-comparison/

Calendar/Courses

NeuroSafe 2019 Symposium
Aug. 8-9, 2019; Minneapolis

SNSA Congress 2019
Aug. 8-11, 2019; Cape Town, South Africa

2019 Managing Coding and Reimbursement Challenges
Aug. 22-24, 2019; Rosemont, Ill.

Leave a Reply

Be the first to reply using the above form.