A study of computer-generated facial composites raises new questions about the credibility of eyewitnesses in criminal cases (Jl. of Applied Psychology 1997, Vol. 82, No. 2, 235-246).
Researchers tested the identification efficacy of the computer-generated facial composite system, Mac-a-Mug Pro. In one study, college freshmen, from memory, prepared composites of other students and faculty from their former high schools. Other students who had attended the high schools could not recognize either students or faculty when the ten composites of individuals known to them were mixed with composites of forty strangers. The researchers also found that viewers had confidence that they could recognize the faces depicted when they could not. Neither preparer-familiarity with the target, preparer-assessed composite quality, nor viewer-familiarity predicted composite recognition.
In the second study, participants rated the match of a composite to a set of five photographs, including the target individual and four foils. Participants were told that each photograph lineup might or might not contain a photograph of the target face depicted in the facial composite. Participants could not reliably differentiate the target photograph from the foil/distracter photographs. The researchers note that their memory preparation-memory identification study model mimicked the real-life situation where a witness memorizes composites disseminated through posters and television, and then identifies a subject from memory of the composites.
Despite these disappointing results, computerized facial composites still help law enforcement elicit leads and tips. Mac-a-Mug Pro can produce 98 times the number of combinations of facial features than some versions of the non-computerized Identikit and Photokit. Otherwise, their utility maybe limited.
Like IBM chess king Deep Blue and voice-recognition computer software, computer composite information requires further development that incorporates subtle natural tendencies of memory management. Memory distortion will always make translation of exact visual memory difficult. As we still understand little of how this takes place, computerized composite development will likely advance slowly.
