Why is AI still dumb as a rock when it comes to understanding human motivation?
Thursday October 12, 2023.
Can AI ever fully grok a human?
Here’s a scary thought. There is a lot of agita about how the Powers that Be will see every problem as a nail suitable for the shiny, new AI hammer.
Here’s why we need more human babies…they’re smarter than AI in understanding human motivation.
I was shocked to learn that Artificial intelligence lags far behind a human infant in making the simplest and most basic of psychological inferences, according to a new study.
The researchers learned that an 11-month old infant can make basic inferences about the goals, intentions and preferences of other human beings, despite being offered the same stimulus, the AI was dumb as a rock.
What I found fascinating about this research was not the wonders of AI, but rather the humbling complexity of human-ish cognition.
How the study was conducted
The study required 11-month-old infants to watch a series of animated shapes moving around a screen. The movement of the shapes was designed to replicate the essential patterns of basic human decision-making behavior.
Both infants and the AI, which had been trained on literally thousands of views of the same animation, were tested on what they understood about the nature of the videos.
The results were summed up by Dr Moira Dillon, study co-author:
“Adults and even infants can easily make reliable inferences about what drives other people’s actions.
Current AI finds these inferences challenging to make.”
I was amazed that at the tender age of 11 months, the human babies clearly understood know how unseen human motivations may be found in different situations across environments (in other words, humans are goal-oriented beings which have the familiar capacity to grasp for which one does not yet have)…
The surprise paradigm
How do infants develop an understanding of emotions such as motivation? Scientists believe that a child’s understanding is expanded by what they call a ‘surprise paradigm’.
We’ve known for over 30 years that babies have a habit of looking longer at things which surprise them.
The AI, though, lacked the capacity to grasp could the underlying motivations enacted by the moving and shifting shapes in the video.
In other words, the Artificial Intelligence wasn’t ‘surprised’ when the unwritten laws of human behavior suddenly shifted, or the environment had changed.
Dr Brenden Lake, study co-author, said:
“If AI aims to build flexible, commonsense thinkers like human adults become, then machines should draw upon the same core abilities infants possess in detecting goals and preferences.”
Human thought is flexible, applicable to many situations and contexts, Dr Dillon said:
“A human infant’s foundational knowledge is limited, abstract, and reflects our evolutionary inheritance, yet it can accommodate any context or culture in which that infant might live and learn.”
AI ‘thought’, in contrast, is still very disappointing when it comes to a robust understanding deeper, more human-like human psychology. But it has great utility as a smartypants sidekick to just about any human endeavor.
But at this point, when it comes to actual, independent thinking with a human-like response to stimulus, it can’t out think an 11 month-old baby.
Be well, stay kind, and Godspeed.
RESEARCH:
Gala Stojnić, Kanishk Gandhi, Shannon Yasuda, Brenden M. Lake, Moira R. Dillon, Commonsense psychology in human infants and machines, Cognition, Volume 235, 2023, 105406, ISSN 0010-0277,
https://doi.org/10.1016/j.cognition.2023.105406.
