A study released last October has revealed that deep learning could be used to help predict first impressions.
The study, called “Predicting First Impressions with Deep Learning”, looks at participants’ first impressions of a number of people’s photographs based on dominance, trustworthiness, age and IQ.
Researchers then used this data to help train their machine-learning algorithm and see whether it could make the same first impressions as humans would.
To conduct the study, Mel McCurrie, Fernando Beletti, Lucas Parzianello, Allen Westendorp, Samuel Anthony, and Walter J. Scheirer used brain test website TestMyBrain.org to ask participants how they perceived a number of people from black and white photographs.
Participants were asked to rate images from a dataset of 6,300, based on how dominant and trustworthy they thought the person was, as well as their IQ and age.
Each face was analysed by an average of 32 people for dominance and trustworthiness, and 15 for IQ and age.
An average was then created from each photo’s results and a score was generated – the researchers used 6,000 images to help train their algorithm, an additional 200 to fine-tune, and a further 100 images to test the technology.
The report explains: “Rather than analyse and model actual trustworthiness, dominance, IQ, and age, we choose to study people’s described perceptions of the aforementioned traits.
“For example, our dataset does not include actual ages, instead the images are annotated by a consensus score — aggregate statistics of what many people said about the ages of the subjects in the images.”
By doing this, the team were able to create a machine algorithm that can make judgements based on people’s faces in a similar way to humans.
The researchers also used the algorithm to test how trustworthy and dominant Edward Snowden and Julian Assange were, and compared this to how the technology viewed the actors (Joseph Gordon-Levitt and Benedict Cumberbatch) who played them in films.
Results show that the real-life photographs and those of the actors within their roles scored similarly on trustworthiness and dominance.
Scheirer said: “With just a glance, our first impression of a face may lead us to believe that a person is smart, worthy of our trust, and perhaps even our admiration – regardless of the underlying truth behind such attributes.
“In general, we are modelling very low-level cues that people recognise instantaneously and do not consciously think about.
“Our models are quite accurate and have remarkably good correlation with how a group of people will rate faces.”
Find out more about this study here.