that these sorts of machinations are not all performed by the same type of programming language , type of programmer or even type of machine . Indeed , artificial intelligence is a term used so broadly that it can occasionally be difficult to decipher whether something even is artificial or intelligent at all .
Whether or not the general public should receive education on advanced computing is a separate issue , but I do think a cursory understanding may be a de facto requirement for medical practitioners within the next few decades . Central to this education will be what AI is and what it can and cannot do ( Figure 2 ). As illustrated , almost all of the interest-generating methodologies discussed in both the media and scientific papers utilize a narrow subset of artificial intelligence , albeit very complex . Many of the titles use terms like “ deep learning ” or “ convolutional neural network .” The transition to machine learning is straightforward conceptually . As opposed to a calculator ’ s software , which is stagnant , the software in machine learning changes as more information is provided . Deep learning quite literally adds computing layers to these analyses . Instead of processing a single time , images may pass through eight or more layers before the final output is generated . At the less intensive end of the spectrum , most basic statistical software packages come with linear regression capabilities , which is the prototypical example of supervised machine learning processes taught in entry level computer science courses .
Think now on the last time you saw a Methods section without a mention of regression of some variety . The ubiquity of the test is likely related to easy access , as mentioned above , but also because we have become familiar with it as a profession . To illustrate this , I have run a regression myself and forced the statistical analysis software I use to generate a multivariable linear regression on severely flawed data ( Figure 3 ). In this example , some of the issues are obvious in the graph . But if I just reported some of the data , for example that variable x carried a p value of 0.23 in the model I ran , the reader would not have the luxury of much scrutiny . Please keep in mind
( continued on page 16 )
Figure 1
Figure 2
Figure 3
ARTIFICIAL INTELLIGENCE
APRIL 2022 15