تقسیم کار میان انسان و کامپیوتر در حضور مشاوره سیستم های پشتیبانی تصمیم گیری

Prior research suggests that decision support system (DSS) provide model advice and display non-modeled information for decision makers [4,13]. We investigate whether decision makers (1) delegate the processing of the modeled information to the model, (2) cognitively process the non-modeled information, and (3) decide based on the model's advice adjusted for the non-modeled information. Experimentally, decision makers were no more likely to execute normative strategies when they had requisite knowledge for the strategy than when they did not have the requisite knowledge. We observed alternative processing, including ignoring the advice altogether, and evaluating the advice. Our findings suggest that DSS builders must encourage decision strategies that capitalize on the relative strengths of human and computer in using those features.

When a human decision maker is offered advice by a decision support system (DSS), how does the decision maker divide labor between human and machine in processing available information? Research in psychology and information systems has documented the relative strengths and weaknesses of human judgment and intuition compared to mechanical processing of information and the results are clear: formulas, algorithms, and decision rules are almost always superior to humans when processing the same information as humans, but humans can recognize when such mechanical advice should be adjusted because of additional information or extenuating circumstances (see Ref. [17] for a review). Hoch and Schkade [13] recommended a psychological approach to DSS that supports these relative strengths and compensates for these relative weaknesses of human and machine by providing model advice but also displaying potentially informative non-modeled information to the decision maker.

The divide-and-conquer strategy requires that the decision maker delegate processing modeled information by accepting (at least conditionally) the model's advice. This delegation cannot occur if the decision maker ignores the model's advice. Thus, in our search for decision makers who executed the divide-and-conquer strategy, we partitioned out those who routinely ignored the advice. To accomplish this, we analyzed the process data from each of 630 trials (90 participants with DSS advice available multiplied by seven experimental trials per participant) to search for clear cases of ignoring the advice. In 19.0% of these trials, the advice was never acquired by clicking the mouse to open the advice cell on the computer screen, and so we coded these trials as ignoring the advice. In another 16.3% of these trails, the decision maker obtained advice that disagreed with the final choice, but receiving this advice led to no further data acquisitions (i.e., the “disconfirming” advice was first acquired as the final data acquisition in the trial). In these trials, we assume the advice must have had no substantive influence, otherwise the decision maker would have obtained additional information while considering the advice. We coded these trials as ignoring the advice. When these two categories were combined, 35.3% of trials were coded as ignoring the advice.