What is human intelligence? A formal definition is ‘the ability to acquire and apply knowledge and skills’.
It turns out an all encompassing definition for intelligence is not easy to find as it can be defined in many different ways. There was an attempt to put forward 8 types of intelligence by American Psychologist Howard Gardner in his book ‘Frames of Mind: The Theory of Multiple Intelligences’. These types of intelligence were; Visual-Spatial, Linguistic-Verbal, Interpersonal (social), Intrapersonal (self awareness), Logical-Mathematical, Musical, Bodily-Kinesthetic, and Naturalistic. However these were criticized as being abilities or personality traits rather than strictly intelligence.
This wider definition has its advantages though, especially for those who possess one of these multiple intelligences in lieu of traditional definitions. Applying a narrow view of intelligence could lead to missed opportunities in the development of ‘intelligent’ machines. So maybe looking at this question of what is human intelligence again, is worth revisiting.
There are traditional types of intelligence that are familiar to us. For example, IQ tests, emotional intelligence and pattern recognition are commonly used as signals of intelligence in humans. But what about the type we use in AI?
There are types of artificial intelligence we are getting good at. Machine learning gives software the ability to acquire new knowledge and apply it. The most powerful applications in use include learning to understand natural language and recognize images. AI’s key feature is that it can learn but this learning is usually restricted to a narrow purpose. The newly acquired knowledge allows it to get very, very good at a specific thing i.e. recognizing faces. It doesn’t expand beyond a specific task unless we change the learning models and retrain for the new task.
The type of intelligence which we think of in humans is more of a general type of intelligence. Artificial General Intelligence (AGI) is the holy grail of AI. Even the most cutting edge examples like Open AI’s GPT-3, slip up in AGI, failing to recognize common sense questions.
We could look at Maslow’s Hierarchy of Needs to give us some clues to what constitutes a human intelligence. Maslow was a psychologist who was part of the movement for a humanistic psychology. The hierarchy was based on the idea that humans are innately good and want to improve themselves. Before this, the two schools of psychology were behaviorists who only considered the external behaviors not internal ones, and psychoanalysis that focused on unconscious behavior. Maslow (and others) put forward more humane theories. This included the famous hierarchy that showed human needs like self fulfillment and esteem above basic needs like food and safety. Humanistic psychology is closer to what is practiced today. It is usually analyzed through case studies and qualitative data and is referred to as the ‘third force’. We could say this third force movement is echoed in today’s desire for a more human type of AI or machine learning. Perhaps we are entering an era of a new humanistic force of AI.
Very Human Traits
Are there more human traits that could be argued to be additional indicators of intelligence? Here are some proposed ‘higher levels’ of intelligence. They could be labelled abilities, traits or needs but in summary are all very uniquely human. And they are often related to success. It could be argued they contribute to intelligence.
Why is the trait ‘unpredictable’ at the top? Because this is one of the most human things. There is an ever present chance that we could go against all logical and rational thinking. If I was to design a new Turing style test that would be on the list. Humans get tired, they get lazy and they quite often do things that are completely unpredictable. Sometimes due to love or anger or without explanation. So that’s what I would wait for – the inevitable human error that a machine was unlikely to do. While an error may lead to dismay or regret, it sometimes lead to breakthroughs too. Accidents have lead to new discoveries and innovations. Creativity and being irrational can go hand in hand. What seems unlikely or unpredictable can be the key to something new and why it’s such an important human trait. Orson Wells went so far as to say accidents were a fundamental part of film making process.
‘A director is someone who presides over a series of accidents’
We need the accidents to happen if we want creativity.
One way to deduce if these traits are really human is to see if they exist in animals. This is more difficult to assess than you would think. In fact, research shows traits we once thought were only in humans, have actually been found in animals. So we can never know for sure. Perhaps there are undetected subtle traits we are yet to made aware of in the animal kingdom. However we know there are some things we see a lot more and very openly in humans that we don’t see in animals. This BBC article from a few years ago was about a series on this very subject – the uniqueness of humans. Here’s a quote:
“When you pull together our unparalleled language skills, our ability to infer others’ mental states and our instinct for cooperation, you have something unprecedented. Us.”
Which traits and types of intelligence are useful for AI and Machines?
Today we apply a very narrow definition for intelligence in machines. It’s mainly pattern recognition, recalling of large knowledge bases and fast processing. Is that enough? Even if we expand to include emotion sensing, moralistic and more empathic systems (which is happening) there is still plenty missing in our list of human intelligence. A better question to ask might be – What if they just taught machines to know about human traits and intelligence, not try to replicate them? After all if some of these are missing in machines, it probably would not affect the ability to perform the task but it may help to have an understanding (and predict) why humans react the way they do. Lots of initiatives and orgs are trying to make AI more human e.g. Center for Humane Technology, Stanford Human Centered AI. This includes trying to include humanities in the design of AI like philosophy for ethics and psychology for behavior. There is also a philosophy of humanistic AI by Siri co-founder Tom Gruber. It feels like humans are recognizing the need to think carefully about the intelligence we install in machines.
Ultimately we will need to decide which of these human traits and intelligent abilities are most relevant to AI (or any machine). Either to have them display these themselves or at the very least be aware of their existence in humans. However it’s done, the prospect of going deeper into the question of what makes humans intelligence, is sure to be a worthy pursuit.