Dr Hannah Fry: There are limits to AI

After much hype surrounding artificial intelligence there is now a more measured approach being taken to its implementation by industries such as retail, which involves greater consideration of aspects like privacy.

According to Dr Hannah Fry, associate professor at University College London and presenter on TV and radio, AI is going through a classic Gartner Hype Cycle: “Phase one was unbounded excitement, phase two was concerns over privacy, and now we’re in phase three and there is a readjustment. Over the last 10 years people have been excited by AI but now they are being more sensible. People are reigning in their expectations and being more reasonable about what it can tell us. They recognise there are limits to it.”

One aspect requiring caution is the balance between volumes of data and its quality. Retail produces a great mass of data but Fry suggests that it is “much more valuable to have a small, but representative, sample of data to work with rather than a ton”.

“Too much data can be dangerous because you’ll definitely find something in it,” she says, citing the experiment involving a dead fish being placed in an MRI scanner in order to determine its reactions to being shown photographs of different human faces. “The scanner spits out such a vast amount of data that things [patterns] were found in there. It’s the combination of big volumes of data and uncertainty [that’s dangerous],” warns Fry.

Positive on uncertainty

Understanding uncertainty is important for retailers as they need to recognise that taking a data-driven route does not lead to “100% truths”: “Retailers need to be positive on uncertainty.” She recalls the often-quoted story of US-based retailer Target that was able to predict pregnant customers from purchase patterns.

“It looked incredible but there was lots of uncertainty. It was 60% right but they had no idea why some predictions were incorrect. When you hear about algorithms and data then it’s not facts, it is uncertainty,” says Fry.

For retailers to manage uncertainty and maximise the value from data she says they have to be able to answer three basic questions relating to their findings. Firstly, they need to know how to validate the data, to know when it’s right? Secondly, they need to know how often they are right? And thirdly, they need to know what happens if they are wrong?

The latter point is particularly important because the ramifications can be serious. Consider the Target example and also tools that use facial recognition technology to predict the likelihood that someone is a shoplifter when they enter a store.

Retailers also cannot forget that the data they gather, when accurate, is very powerful, and its use can have ramifications – the retail urban legend goes that a father discovered his teenage daughter was pregnant after being sent coupons from Target for discounted baby items.

Dr Hannah Fry’s three basic questions of data:

  • Do you know how to validate the data?
  • How often does the data make accurate predictions?
  • What happens when the data is wrong?

Data science talent

Thankfully retailers are being helped in their endeavours by an influx of suitably qualified personnel, according to Fry, who says the universities are “pumping out” people with the relevant qualifications who want to apply their knowledge in new domains including data science and also within the retail sector.

Whereas once they might have been attracted to those roles that paid the most money – notably financial services – she says this has changed. “When it comes to science graduates, who want to be data scientists, their primary motivation is to make a difference, not money. Some people will be put off if they are using their skills for profit alone,” says Fry.

She adds that greener, environmentally-conscious brands, and those with interesting data sets and projects have attractive characteristics that helps them to recruit high-quality people. Retailers definitely fall within this category and Fry says the supermarkets in particular are leaders in data analytics and so have attractive attributes in the job market.

“Their basket analysis is impressive. They are also leaders in tracking people and optimising the flow of customers,” she says, adding that what would make things interesting in stores and shopping centres would be combining this with behavioural data so that it would be possible to see if say certain floor tiles affected the movement of people.

What would also undoubtedly prove very useful in attracting the right people would be the ability to offer them time to work on “blue sky research”. Fry recommends all retailers allocate some resources to this area as it can lead to the discovery of some of the most valuable findings.

“If you can support blue sky research then it’s a positive. The freedom to just look for patterns and play with the data is very important rather than just trying to find solutions to problems,” she says.