Friday Five: machine learning myths and data set diversity

Each week Letty Key picks the five best stories on digital trends, experiences and technologies...

Granary Square

1. Second chance at first impression for Duplex

As regular Friday Five readers will know, Google Duplex's demo in May was met with reactions ranging from awe to anger. I was impressed with the quality, but it also generated a collective sense of unease. Some critics suggested the AI phone assistant was duplicitous while others thought the whole thing could have been staged. 

Google set out to convert the non-believers this week by organising an immersive demo at a Thai restaurant, allowing attendees the chance to take Duplex's calls. Initial reports have been positive, with this article offering an interesting insight into Google's test and learn strategy, which includes adapting voice and personality to see what is received best by those who encounter Duplex.


2. Debunking the myths of machine learning 

Nearly everyone will be familiar with the phrase 'machine learning', but what does it actually mean? It's a many-headed beast in the minds of both consumers and businesses, which is why this article is useful in finding a middle ground between the mathematical nuts and bolts of it all and dystopian fantasies of machines taking over.

As the author points out, the term 'artificial intelligence' is more of a hindrance than anything, as AI has become this opaque monolith (à la 2001) that defies analysis. It's a great palate-cleansing analysis that offers enhanced clarity on what is possible and strategic use cases.


3. Under the influence(r)  

The credibility of influencers has come under scrutiny this week, with ripples being felt as far away as Cannes. Unilever CEO Keith Weed has railed against 'influencer fraud' and then there's the LA mural that was only accessible to verified influencers and those with more than 20,000 followers, which later turned out to be a PR stunt. 

The latter is an interesting lens to view the topic through, in that calling out the issue of numbers in a real-life space made it all look a bit callow and naff. Influencer authenticity is important, but given that until now the primary focus of brands has often been breadth of reach, you cannot entirely blame influencers for chasing those numbers.


4. Microsoft (finally) embraces diversity in data

Diversity isn’t just about giving all people opportunities, it’s actually vital to making fundamentally good things – whether those are businesses, products or decisions. And the need for diversity in tech was highlighted spectacularly by Microsoft's facial recognition software that turned out to be massively racist.

The program had near-perfect accuracy for light skinned men, but an error rate of 20% for dark skinned women. Microsoft has now returned a with more representative offering, which allows for the somewhat underwhelming claim that the software is “no longer racist” and slightly less sexist. But it’s a definite win for showing that algorithm bias can be analysed and addressed.


5. Mighty oaks from little acorns grow  

While many traditional brands are concerned with translating real-life experiences to digital, fintech company Acorns has been grappling with the opposite as it attempts to convert its message of mindful digital saving into a physical service object.

The result is the company's first debit card, made of a heavy metal that has been designed to reflect the significance of spending. Digital-first services are increasingly looking to carry their concepts over into the physical world, which makes psychological research into experience design and the ramifications it has on how much we value products all the more interesting.

Like what you see? Sign up here to get the Friday Five delivered directly to your inbox every week!