Humans vs Algorithms

Since the conception of the first computer operating on algorithms, there has been a struggle between humans and machines – when will computers get so good at doing our job that they should replace humans?

Two most interesting articles shed light on this question. One was written by Garry Kasparov¹, the chess grandmaster who battled the famous Deep Blue, a supercomputer programmed by a full team of humans to play chess. Kasparov’s career spans a pivotal period when computers progressed from a weak chess player to an unbeatable one. But the most interesting finding of all was the fact that when weak human players team up with a few average computers, they are superior to chess grandmasters or the best supercomputers alone.

The second article by Kahneman² explains why this is by characterizing how human intuitions work. First, good intuitions take years to attain – a study by Chase et al³ showed that chess players take 10 years of dedicated study and competition to possess a good mental collection of board patterns that allow players to identify a good move without calculating all possibilities.

Second, intuitions are easily affected by biases or the way information is presented. One example is the anchoring phenomenon. When people are asked “Is the average price of German cars more or less than $100,000?” before giving an estimate of the average price of German cars, participants will “anchor” around Mercedes and high-end cars when estimating. On the contrary, when another group of respondents are asked a different anchoring question “Is the average price of German cars more or less than $30,000?” they anchor around cheaper cars and give a lower estimate.

Third, human intuitions are inconsistent. A study by Goldberg⁴ created simple diagnostic algorithms based on the criteria used by 29 psychologists to distinguish neurotic from psychotic patients. These 29 psychologists then compete with algorithms built from their knowledge to distinguish new sets of patients. Researchers found that algorithms differentiated neurotic from psychotic patients more accurately than psychologists from whom the models were derived. Kahneman believed that this is because human judgments are inconsistent.

Last and most importantly, intuitions only work in a limited environment that provides good cues and rapid feedback. This explains how a team of weak human players and average computers becomes so powerful. In this setup, humans use accurate cues and rapid feedback provided by computer calculations to make decisions. The key is a good process with humans and computers playing to each other’s strengths – computers are better at processing information, while humans are better at strategic planning.

The advancement of computers creates an impending shift in how we practice medicine. As we strive to limit human errors and stretch limited resources to expand health care access to underserved areas, computers and their algorithms play an important role. To reduce medical errors, evidence-based guidelines can help physicians concentrate on strategic thinking instead of recalling medical knowledge from memory. To increase access to care, we can rely on a team of health workers aided by computer algorithms for simpler medical problems, while referring to human physicians and their intuition to override preformed models for more complicated cases.

Overall, considering the fact that common medical conditions are common, the role of physician experts will change as we rely more on evidence-based algorithms and mid-level health professionals. This shift should free doctors to fulfill a bigger, more complicated role in health care that has yet to be determined. Until then, it is important to recognize that intuitions are limited. Using algorithms to aid in diagnosis is not just an option – it is a must to improve patient care.

References

1 Kasparov G. The Chess Master and the Computer. The New York Review of Books. 2010:57:2

2 Kahneman D, Klein G. Conditions for intuitive expertise: A failure to disagree. American Psychologist. 2009:64:515-26

3 Chase. The mind’s eye in chess. In: Chase WG. Visual information processing. New York: Academic Press, 1973:215–81.

4 Goldberg LR. Man versus model of man: A rationale, plus some evidence, for a method of improving on clinical inferences. Psychological Bulletin. 1970: 422-32.