Countless tech headlines warn us of – or celebrate – the rise in artificial intelligence (AI). AI is taking over our jobs, say some, while others extol AI’s ability to enhance the workplace and increase business opportunities.
What is undisputed, though, is the intelligence of artificial intelligence. The very basic of AI – artificial narrow intelligence, or weak AI – is known to be better than humans at one particular thing at any one time. Take it one level up, and artificial general intelligence may combine near-human-like thinking with impressive computational prowess. A feat that humans likely won’t be able to accomplish.
Incredibly, while AI seemingly has infinite possibilities, there is one arena in which humans can outsmart it (and no, we don’t mean emotional intelligence). As we’ll explain below, we haven’t yet come across an artificial intelligence that can decipher optical illusions as cleverly as humans can. Take this optical illusions quiz to test your mettle, and read on for reasons why humans are better at optical illusions than AI.
Optical illusions improve our visual literacy, but AI can’t compute
Visual literacy refers to the brain’s ability to interpret and derive meaning from visual information. When we look at an image, the brain picks up on visual cues – light, shadow, height, etc. – to provide context to what it’s seeing.
This context helps the brain understand the image it’s seeing and anticipate what that image logically means (for example, in the famous Kanizsa Triangle illusion, our brain can see a complete triangle despite no such shape actually being present). This logic is based on connecting the image we see to images we’ve seen in the past and predicting what the ‘probable’ image might be. Optical illusions deliberately mislead the brain by manipulating the all-important ‘context’ of the image. Thus, regularly viewing optical illusions can help improve our visual literacy as well as make it harder for us to be ‘caught out’ by the deception that illusions try to create.
In an effort to learn more about how optical illusions impact the human brain, researchers Robert Williams and Roman Yampolskiy from the University of Louisville (UofL) turned to AI. They wanted to task AI with creating new optical illusions they could test on humans. They pumped their AI’s neural network with over 6,000 optical illusions, hoping it could in turn create some new illusions. Alas, the AI simply couldn’t figure out how to create new illusions from the data given to it.
Our brains are better at understanding optical illusions than AI
Unfortunately for the UofL researchers, neural networks – or the same machine-learning systems that can identify faces and create realistic new ones – simply can’t recognise optical illusions. According to the researchers, there are only a few thousand static optical illusions in existence. This meant their neutral network didn’t have enough information to work with. Currently, we don’t have an advanced enough machine-learning system that can learn from such a small dataset.
The key, it seemed, was that humans can actually see the illusions whereas machine-learning systems can’t do the same. They can’t understand the principles behind the illusions – i.e. the context that human brains can quickly pick up on – and lack the same complex structure of the human visual system that optical illusions aim to play upon and manipulate.
For the moment, at least, AI is not advanced enough to be able to interpret and decipher optical illusions. Humans, on the other hand, can not only decipher illusions and understand how they work, they can also overcome them, learning – through practice – to become immune to the manipulation that illusions rely so heavily on. So, at least where optical illusions are concerned, humans outsmart AI.