The top three news stories of the week, as chosen by our resident students. This week’s top stories are AI bias, odour nostalgia, and tying your shoelaces.
Article written by Justin Aluko.
In our image we made them
Artificial intelligence (AI) promises to be a game changer for mankind. Services like your spam filters, Netflix, or the plethora of virtual assistants found on any modern mobile device are making a big difference, even though the technology is barely in its infancy. During this time of exponentially increasing development into AI, there are moments when we have to be humbled by some of the results. A new study inspired by the implicit association test (IAT), a psychological tool used to indicate subconscious bias in humans, has found that this bias can manifest itself in machines as well.
The authors quantified this by using a machine analogue to the IAT, known as word-embedding association test (WEAT). The computer is provided with a set of established “word-embeddings”, which is effectively a computers definition of a word based on the context in which it appears. For example, “ice” and “steam” would have similar embeddings because they are more frequently associated with the word “water” rather than anything else. The computer then went on to analyse hundreds of billions of words on the internet and determined how strongly words correlate with other words. As a result, it found both gender and racial biases by associating female names with family more than career words, and African-American names such as Alonzo and Shaniqua were more associated with words such as “cancer” and “failure” than European-American names such as Adam or Stephanie.
This, however, doesn’t mean that computers are intrinsically biased, but rather it has simply picked these biases up from analysing words put into a particular context by humans and magnifying them in a similar way to Microsoft’s chatbot, Tay in 2016. While questionable applications could be derived from this behaviour as it stands, it does go some way to highlight the caveats of unsupervised machine learning, particularly when applied to decision-making, such as CV screening or applications for loans etc.
Will robots judge us based on our own bias?
That brings back memories
The area of the brain responsible for identifying smells is very close to the region responsible for maintaining long term memory. That’s why a particular smell can lead to a strong sense of nostalgia as your mind takes you back to a point of significance associated with that smell. In this vein, it is easy to overlook the fact that there are some smells that are moving ever closer to extinction, and with them that unique memory connection. Books are an example of this, due to their increasing digitisation and resulting reduction in prints. Luckily for those bibliophiles among us, a team of researchers from University College London have ensured that even after printed books are looked upon as fondly as floppy discs, their unique smell typically associated with libraries and museums, will live on.
Paper releases organic compounds as it decays, which are picked up by the nose and recognised as that “old book smell”. The authors managed to extract this simply by cutting up old paper into small pieces and effectively boiling it for a couple of hours after pre-treating with methanol. The remaining solution was then bottled and taken to several public spaces to assess the general public’s reaction to the smell. As a result, they derived a preliminary paper odour wheel, listing organic molecules associated with different categories of scent.
Understanding and extracting the compounds related to particular scents presents a novel way for both museums and libraries to document and preserve something that is so intrinsically linked to the visitor experience. Perhaps “scent galleries” are just around the corner!
Love the smell of an old book?
So you think you know how to tie your shoelaces?
All knots are equal, but some knots are more equal than others. This certainly rings true when it comes to your shoelaces. I’m sure we’ve all had pairs of shoes that, no matter how hard we tie them, they always manage to untie themselves, and as we bend down for the umpteenth time to retie them we curse under our breathe, “why the h*** is this happening!” Well, curse no more as engineers from the University of California have attempted to solve this most annoying of peeves.
While looking almost identical, it is already known that the “square knot” is more robust than the “granny knot” (I see you raise an eyebrow at this, check out an explanation here). This team of scientists tried to figure why this is the case by attaching accelerometers to the shoes of a runner on a treadmill to measure the g-force the knots experienced with every stride which, surprisingly, can be more than any rollercoaster can generate! By defining the average slip rate of the knots, they were able to characterise two regimes of knot failure, known as gradual loosening and acute failure. However, despite reconfirming the superiority of the square knot, they were still unable to explain exactly why this is the case.
Understanding and predicting when a knot will fail has implications beyond just tying shoelaces. Knots exist in all sorts of places that are perhaps not commonly considered, such as surgical stitches, on boats, in mountaineering and also on the molecular level, such as in DNA strands as a mechanism to regulate gene expression. “Why”, it seems, could be a question more profound than we thought.
Tying shoelaces like a pro!
Article written by Justin Aluko.