Scientists at the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) developed an algorithm that can determine which photos of faces, landscapes, or other objects people will find most memorable.
Aditya Khosla, a PhD student in computer science at the Massachusetts Institute of Technology, said that it is that first time computers have been taught to predict which photos people are more likely to remember.
There are other similar predictive algorithms that computers use, such as completing phrases people type into Google search, but the one developed by MIT is completely new.
According to Aude Oliva, principal research scientist at the MIT Computer Science and Artificial Intelligence Laboratory, people used to think that computer scientists would never manage to tackle high-level cognitive process like predicting human memory.
To test the MIT algorithm, the scientists worked with 5,000 people from across the globe – with whom they stayed in touch through a site called Mechanical Turk. The CSAIL team showed the people hundreds of images and recorded the memorable ones (each image was shown for only 600 milliseconds). Then, they presented the same images to the algorithm to see whether it could tell which ones humans found memorable.
The algorithm found patterns of memorability among different images – for instance what makes a photograph more forgettable or more memorable – but it could not explain how it found those patterns, according to Khosla.
Previous research found that photos of people usually stand out more than those of landscapes. Positive emotions are also less resonant, compared with negative emotions.
In 2014, the CSAIL team analysed 2.3 million photos using the algorithm, to determine which ones were more popular or less popular. The results showed that women in bikinis, and revolvers had the highest popularity, while laptops, spatulas, and plungers were among the least popular. Bright colours also appeared to draw a lot more attention.
Khosla said that image memorability and popularity could also have applications in education (besides social media or advertising). For instance, it could be used to determine the memorability of an image before placing it into text books. That way it would make it easier for students to remember the images.