Machine Learning: The difference between knowledge and understanding

POSTED BY   Will Darbyshire
1st November 2017
machine learning

Machine Learning can, in theory, and increasingly in practice, compose music, recognise faces, identify smells, write prose; paint. But will it ever learn to understand the intricacies of the human condition?

Our sense of smell is extraordinary, it can identify, with incredible efficiency, the most specific of aromas. Blindfolded, we can identify the smell of butter as it melts in a pan, concrete after summer rain, the difference between raw onion and cooked, and whether a garden has been recently mowed.

There is scientific reasoning behind why things smell like they do, and even, to some extent, why certain smells appeal and appall certain noses. Aromas are created by molecules in specific chemical structures. And while two chemical structures can create very familiar smells, we are slowly learning to predict what certain chemical combinations and structures will smell like. When we can do that, we can teach machines to identify everyday smells.

But our sense of smell has a special curiosity, that which reacts instantly with memory. For example, I recently stayed in an Airbnb which, from the instant I walked in, hit me with a smell that threw me straight back to my Nan’s house. A strange house in Croatia took me back to the familiar safety of a home in York.

The smell was, in hindsight, a specific blend of potpourri that reacts with my brain and my memories. There are more simple versions of this, such as cloves smelling ‘Christmassy’.

machine learning


Music is composed through the mathematical placement of a sequence of notes that, from childhood, we are conditioned to recognise and appreciate. I say ‘condition’ because those of us who are born and raised in England, for example, are taught to appreciate a very different scale of notes to that which they do in, say, India. What does and doesn’t sound good to us depends on what we’re exposed to while growing up.

The Western Chromatic Scale encourages the use of chord progression and counterpoint, major and minor tonality. But Indian music uses neither progression nor counterpoint, instead employing a single melody line over a monotone drone. To many of us over here, it can be difficult to listen to.

Regardless, both musical styles are built upon the same mathematical rules of music, rules that can very simply be taught to a machine. As the machine gets older and composes more music, just as a with a human, their compositions will gradually increase in sophistication.

But music also has the ability to, wordlessly, make us cry; sometimes inexplicably. This too, much like smell, is our brain’s obsession with mixing sound and memory. Music is composed by an idiosyncratic composer. As such, it reflects an individual’s intent and emotion. And because the human experience is, in so many ways, universal, that individual’s expression resonates with others.


Machines can certainly write prose, especially when it comes to formulaic genres such as detective fiction. But can machines scare the shit out of us like Stephen King? Describe the indescribable sense of foreboding dread? Or is fear something so innate and deep within us that only a fellow human can exploit it?

As Machine Learning evolves and grows more sophisticated, it’s reasonable to imagine a world where people have their brains scanned while being scientifically frightened so that machines can begin to identify patterns in that which scares us. Broad strokes are possible; ghosts, for example, but the question is, will it be powerful enough to teach itself the subtleties of truly terrifying stories?

machine learning

Can a machine learn enough to exploit the minutia that lies behind our universal fear of death? If so, can it articulate it clearly? Can it invent unique metaphors; accurately describe the touch of another’s hand on your skin; or the warmth of the sun on your face?

I would argue that it can’t. Just as I would argue that a machine cannot compose worthwhile music because it lacks the memories that affect emotion and thus the emotions that affect the scoring.

And so I bring it back to smell, for I think it’s a useful way of questioning the benefits of Machine Learning in the creative arts. We will certainly, one day, have machines that can identify smells, even raw onion versus cooked. But what it won’t be able to do is understand the individual relationship each of us has with certain aromas, much like it can’t convey the abstract nature of ‘heartbreak’ via the medium of oils on canvas.

And what’s the point of that? It’s one thing to know something, another to understand it. The machine smells potpourri, but it doesn’t smell Nan’s house. And because machines can’t understand the human experience, it matters not how much they know because they’re completely missing the ultimate point of the arts;  to help us further understand who we are in relation to the world around us.  

Machine Learning: The difference between knowledge and understanding
Article Name
Machine Learning: The difference between knowledge and understanding
A discussion about the worthiness of using Machine Learning in the creation of the liberal arts, including music, literature, and painting.
Publisher Name
The Digital Marketing Bureau
Publisher Logo
Machine Learning: The difference between knowledge and understanding

Will Darbyshire

Will is Content Strategist with The Digital Marketing Bureau, writing on all aspects of tech. Will specialises in writing interviews and profiles, as well as all things PropTech.

Get in Touch With Will Darbyshire

You may also like

A Cruelty-Free Future With Machine Learning
Machine Learning In Retail – A Help Not A Hindrance? 
17th January 2018
Machine Learning
The Sunday AI Review | 18-24th September 2017