Training AI in Social Intelligence

SOLVVE continues to highlight the development in the area of Machine Learning and Artificial Intelligence. Previously we discussed how this technology can help with predictions and automation. However, there is more to it. Namely, AI is also handy when it comes to social intelligence or emotional intelligence.

We interact with AI-powered products and services every day and this interaction will only grow closer over time. That is why SOLVVE decided to discuss what has been made to train AI in social intelligence and how these achievements can later be applied to solving issues related to the interaction of humans and machines.

What skills in social intelligence AI already has?

Assessing human age

Researchers from Ruhr-Universität Bochum (RUB) have published a report in the Journal of Machine Learning Research about their achievements in age recognition by photo using AI. Their algorithm is capable of assessing people’s age with only a 3.5-year error from an actual number. If you think that this number is too big, it is worth mentioning that compared to human assessment AI is much more precise. And this is despite the fact that humans have way more experience in dealing with age assessment simply due to the nature of our daily existence as social beings.

Interestingly, not even researchers themselves are sure about how it works and what exact factors the algorithm takes into account. That is because researches have conducted their study in a reverse manner.

The usual way for similar researches would be to feed the algorithm with a data set of pictures and corresponding ages and see what intermediate steps the algorithm would create when assessing the ages. However, RUB researches compiled data the other way around. They fed sets of photos already sorted by age and let the algorithm figure out the features for each age.

That is why it is hard to tell which features are integral to each age and which were discarded as non-essential. But it is safe to assume that algorithms still rely on such factors as wrinkles or blemishes the way humans do. However, it is more accurate than humans already at this step.

Assessing emotions by how one walks

ProxEmo is a project that resulted in a scientific paper from the Unversity of Maryland where researchers taught a machine to understand human emotions by gait and act accordingly. After looking at a person in motion, the robot will distance itself from an angry person and walk closely to someone who is sad and offer help.

To be more precise the algorithm understands perceived features of a certain mood because currently not even a human being can be absolutely sure about how another person is feeling. Nevertheless, same to the perception of age, perception of mood is a social skill that is crucial to our daily interactions.

Understanding behavior patterns

Close to the case above but not the same. AI can recognize movement patterns that are typical to certain types of behavior that rather universal in a given community. One example could be a detection of shoplifting cases where people behave in a coherent manner if trying to sneak items. By knowing how people usually move in a given space, the algorithm can spot anomalies and inform security. In other cases, AI can help to monitor the distance between people to fight COVID-19 spread. Some warehouses deploy Amazon’s AI-powered trackers to help workers observe social distancing.

Why these results are important?

Automation as a part of Industry 4.0 draws a future where people will be living and interacting with robots and systems even more often than today at home, at work, during the commute, etc. Taking the pandemic situation into account, automated systems within healthcare become crucial as well.

To smoothen this interaction it is important to make our technologies as human-like as possible: to understand our feelings, intentions, and expectations. That is why developing AI that is trained in social intelligence is important to maintain our daily routines where we feel comfortable and safe.

If you have any questions or ideas related to Machine Learning or Artificial Intelligence, do not hesitate to contact us. SOLVVE experts will find the best solution for your goals and budget. Let us make this happen!