So the researchers created a computational model that can mimic humans’ ability to predict emotions
Anticipating human emotions is our ability as human beings. Will artificial systems be able to do it too? It seems to be, according to the latest scientific research, the new frontier of the so-called Cognitive Artificial Intelligence.
On social media there is a poster that concerns ChatGpt and that is making the rounds on the web. A billboard asks the chatbot to “build a building”. Obviously it’s all a provocation, to remind people that there are some things that AI just can’t do, like build buildings. There is a big one however: a limit in the organic nature of what ChatGpt can access concerns only the current times, or at least, what we know of the developments in robotics.
Realities like DeepMind have already shown us the possibility for robots to bring their capabilities closer to those of humans and a future in which machines will build a building, enhanced by the successors of ChatGpt, is not that far away. The tipping point could be the possibility for algorithms to get emotional, to predict how they will feel after an event. In a nutshell, getting even closer to man and his ability to anticipate human emotions. And on this point we are already on the right track.
Anticipate human emotions, research
MIT neuroscientists have designed a computational model that can predict other people’s emotions – including joy, gratitude, confusion, regret and embarrassment. The model was designed to predict the emotions of the testers involved in a situation based on the prisoner’s dilemma, a classic game theory scenario in which two people have to decide whether to cooperate with their partner or to cheat on him (a game in which one understands well what means “anticipating human emotions”). To build the model, the researchers incorporated several factors that were hypothesized to influence emotional reactions, including that person’s desires, their expectations in a particular situation, and the possibility that someone was observing their actions.
“These are very common basic insights,” he explained Rebecca Saxe, the John W. Jarve Professor of Brain and Cognitive Sciences, a member of MIT’s McGovern Institute for Brain Research and an author of the study. “We wondered if we could integrate that basic grammar to create a model that learns to predict people’s emotions, according to some parameters.”
While a great deal of research has gone into training models to anticipate human emotions, or rather infer a person’s emotional state based on facial expressions, this is not the most important aspect of human emotional intelligence.
“Much more important is the ability to predict someone’s emotional response to events before they happen,” the study said. “Anticipating what others will feel before it happens belongs to humans, not machines. If all of our emotional intelligence were reactive, it would be a catastrophe».
Le Golden Balls
To try to model how human observers make these predictions, the researchers used scenarios from a British TV game show called Golden Balls. On the show, contestants play in pairs to try and take home a $100,000 pot. After negotiating with their partner, each contestant secretly decides whether to split the pool or try to steal it. If they both decide to split, they each receive $50,000. If one splits and one steals, the thief gets the whole pot. If both try to steal, nobody gets anything.
Depending on the outcome, contestants may feel a range of emotions: joy and relief if both split, surprise and anger if one’s opponent steals the pot, and perhaps guilt mixed with excitement if one successfully steals.
To create a computational model that can predict these emotions and simulate how humans are able to anticipate human emotions, the researchers designed three separate modules. The former is trained to infer a person’s preferences and beliefs based on their action, through a process called “reverse planning.” It is based on what happens when we see only a small part of someone’s behavior, from which we can then probabilistically infer how a situation could develop. Using this approach, the first module can predict the motivations of the contestants based on their actions in the game. For example, if someone decides to split in an attempt to share the dish, it can be inferred that they expect the other person to split as well. If someone decides to steal, it could be because they expect the other person to steal and therefore don’t want to be cheated. Or, you expect the other person to separate and decide to try to take advantage of it. The model can also integrate knowledge of specific players, such as the competitor’s occupation, to help him infer the most likely motivation.
The second module compares the outcome of the game with what each player wanted and expected to happen. Then, a third module predicts what emotions the contestants might feel, based on the outcome and what was known about their expectations. The latter was trained to predict emotions based on human observers’ predictions of how contestants would feel after a particular outcome.
The authors point out that it is a model of human social intelligence, designed to mimic the way observers causally reason about each other’s emotions.
«From the data, the model learns that what it means, for example, to feel a lot of joy in this situation, is to get what you wanted, to do it correctly and without taking advantage of it», Saxe’s words.
Fundamental insights to anticipate human emotions (artificially)
Once the three modules were up and running, the researchers used them on a new dataset from the game show to determine how the models’ emotional predictions compare to predictions made by human observers. The result, according to the researchers, led to conclusions much closer to those obtained so far, with previous experiments but less “referenced” from the point of view of the models.
Success comes from its incorporation of key factors that the human brain also uses to predict how someone else will react to a given situation. For example, calculations about how a person will evaluate and emotionally react to a situation, based on their own desires and expectations, which affect not only material gain but also how we will be perceived by others.
“Our model incorporates human metrics, which are not only based on obtaining an economic reward but also on being considered honest by others, without being cheated or passing for simpletons”, continue from MIT.
The researchers managed to build a deeper understanding of how emotions help determine our actions, and then, by flipping their model, to explain how we can use people’s actions to infer their underlying emotions. A line that helps to consider emotions not only as “feelings”, but as a crucial and subtle role in human behavior.
breaking latest news © (Article protected by copyright)