Great share by Gideon Rosenblatt , and I’m sharing my comment on his thread as intro, too.
The truth of the article’s headline should be easy to grasp: all one has to do is read the weekly #SciSundayDigest that Mark Bruce has put out like clockwork on G+ for the last several years. I used to make it a point to read them every week and try to comment on at least a few, but I’ve drifted to other interests and don’t catch it like I used to. A summary of just 10 papers each week, for a total of 520/year, but the reading he must do to choose those 10 every week? Heck, I can’t even keep up with the research just in dentistry, much less medicine.
We certainly need better systems for choosing good potential research paths, summarizing and making available the research that’s been done, and DEFINITELY weeding out the crap. If AI could be used to help catch fraudulent papers, identify papers with faulty or poorly done analysis, etc, that alone would be hugely beneficial.
I do know that AI is being applied in the search for possible new compounds, materials, antibiotics, etc, but not how widespread or effective it is yet.
In the arena of personalized medicine, this is where I have some of the greatest hope; in particular, in diagnosing very rare genetic abnormalities. AI could search and make available information on rare cases with unusual signs and symptoms from around the world, of which humans might not be aware.
Science, technology, and medicine in general have surpassed humanity’s capacity to grasp, and we need help. Hopefully, AI can be that help.
Originally shared by Gideon Rosenblatt
Automating the Scientific Method
As science expands, the relative percentage of what is known that one scientist can retain in her or his head is diminishing. This is exacerbated by the growing specialization that our modern scientific establishment encourages.
In short, it is getting harder and harder for scientists to keep up with the rapidly expanding knowledge of the scientific community as a whole.
This begs the question, are there ways for artificial intelligence to assist with this process?
Francis Bacon (pictured), whose work Novum Organum was a progenitor of the modern scientific method, described his “new method” of induction as like a machine. This suggests that it is inherently algorithmic, and thus amenable to automation. The article talks about some possibilities here.
What interested me the most was its brief discussion of the work of Don Swanson in literature-based discovery (https://goo.gl/0GWVKH), which is essentially ways for navigating through existing scientific publications to connect scientific findings together in novel ways. Think of it as a kind of crawling of a massive scientific knowledge graph and making new linkages as a way to create new hypotheses. Hypotheses which might then be tested through automated experiments. Pretty cool, huh?
Need a reminder of the difference between deductive and inductive reasoning? Here’s a quick video: https://goo.gl/69h10N
A special thanks to Darius Gabriel Black for flagging this one for me. If you don’t know him already, Darius frequently comes up with interesting stuff.
https://aeon.co/ideas/science-has-outgrown-the-human-mind-and-its-limited-capacities
]]>