It isn’t Magic; It’s Machine Learning — Including human input in data-driven processes
I still remember reading Harry Potter and the Philosopher’s Stone for the first time. As the Sorting Hat debated with himself whether to put Harry into Slytherin House or Gryffindor, my adolescent nerves rose. I internally pleaded with the Hat, “you don’t know the full story!”
I know what you’re thinking: yet another Millennial obsessed with Harry Potter, because reading the iconic book series is the only thing we’ve ever committed to. But bear with me here.
When I think of this scene today, I can’t help but draw comparisons to the Data Sources used for Machine Learning. If the Sorting Hat hadn’t worked Harry’s personal, verbal request of “not Slytherin” into his model, the data would have directed Harry into Slytherin House. From this comes my thesis:
The data may point to one outcome — but human perception and preference can change the story.
Behind every datapoint is the person who created it and, with it, a deeper narrative. Here’s what I mean:
The Data
In the Sorting Hat’s case, the data points he normally uses to draw recommendations and conclusions were clear. Harry showed dominant traits of: ambition, achievement, cunningness, cleverness, perseverance (Slytherin Qualities) and courage, bravery, and chivalry, with a daring and reckless nature (Gryffindor Qualities).
“Hmm…difficult, very difficult. Plenty of courage I see, not a bad mind, either. There’s talent, oh yes, and a thirst to prove yourself. But where to put you?”
The Deeper Narrative
The Sorting Hat isn’t aware some of the above traits are buried deep within Harry’s psyche due to his ‘linked’ relationship with Voldemort. In fact, it is utilizing a dataset that Harry may not even be cognitively aware of.
Harry’s request of ‘not slytherin’ is coming only from the social interactions he has had with Slytherin and Gryffindor house-members alike. Had he not interjected, we would have had a very different book on our hands:
“‘Not Slytherin, eh?’ said the small voice. ‘Are you sure? You could be great, you know, it’s all here in your head, and Slytherin will help you on the way to greatness, no doubt about that… no? Well, if you’re sure — better be, GRYFFINDOR!’
Here’s what I am getting at, real-world version:
If I am sharing my data with a company, my expectation is they will use it to create a better human experience for me. Although I strongly welcome the data and digital revolution, flexibility in human interference is paramount to giving me the experience I truly want. The data may say one thing (Slytherin), but I still want the chance to ask for another (not Slytherin).
We’re in an interesting age of data and customer experience wherein the collective value of one’s personal data is becoming increasingly critical for businesses to drive their operations effectively. Each transaction of data, between customer and business, creates a precise and equitable reflection of the presumptions and stories those datasets tell.
Through a lens of presumed customer preferences, we have the ability to drive next-generation human experiences for Albertans and the world. This opportunity comes at one ethereal cost: a current snapshot of all the personal preferences and tidbits of your identity you choose to share. And while unfiltered, unmodeled (and secured) data doesn’t amount to much unused, the opportunity for social implications begin to climb as soon as predictions or insights are formed based on that narrow data.
And, here is the trick I get stuck on: you deserve to know which pieces of your data are being used to drive Machine Learning recommendations or predictions in an easy-to-understand language and format. Standards of which, formed in the European Union’s General Data Protection Regulation (GDPR), are already starting to demand a customer/human-focused approach to data usage.
Humans are naturally evolving in each moment. Data-sets are a snapshot in time, potentially updating with the latest story (for example, transactions) — or, perhaps not. If you took my data last year compared to today, it would be clear that my interests, morals, values, and preferences have changed, grown, and diversified over time.
The Sorting Hat, notorious for refusing to admit it has made a mistake in its sorting of a student, could represent the way some companies may stand behind their Machine Learning products.
When AI is pointed at a particular dataset, and directed to suggest a particular outcome, it is important to remember its under-lining data will be pointed to yesterday’s ideal, not tomorrow’s. These insights, built on our past, are insightful and valuable for moving the needle forward with customer or digital experiences; however, one of the risks of looking at narrow, historical data sets, would be to assume the future will look like the past.
Often, when we think about social questions, policies, or even a customer experience, we don’t want our future to look like our past. We, sensibly, want to see improvement. So, what happens when the data or Machine Learning recommendations and our instinct don’t align?
The role of predictive analytics today is twofold: one, to accelerate business process faster and more efficiently than human computation can manage; and two, to provide insights a human may not have registered.
When a computer or machine addresses datasets, it is not aware of the social implications of its decision making. It isn’t using human emotions to dictate right or wrong, unless it was specifically programmed to mimic them. And, while this can serve as a strong data-driven backbone to the gut-feel of our past decision-making, it is important to remember it will act as a rearview mirror. Humans should feel empowered to disagree, challenge the math — not magic — and ask for what they really want, even if it goes against the data.
The outcomes of AI will inevitably hold a mirror to our missed opportunities of the past; the good and the bad. Today, only humans can move the needle on ‘right’, ‘wrong’, and ‘ethical’ for our future.
Ethics of Care is the process of continually revisiting, because what is right today may not be right tomorrow. And, while I am excited for companies to embrace these new technologies, for both their bottom lines and my user experience, people need to demand excellence from their companies and the AI/ML capabilities they use.
As technologies integrate further and deeper into our lives, I am often reminded of the skills and specialties which make humans unique. Empathy, deep intelligence and, perhaps the most fitting as an ATB team member, happiness are determining factors we use everyday in our decisions. As humans, we consider more than the narrow datasets in front of us.
If algorithms are simply opinions of success criteria hidden in code, I am eager to see how individuals challenge the companies they engage with to use a diverse success criteria. When ‘delight’, ‘happiness’ ‘or ‘human experience’ are just as important as the company’s bottom line, we have the opportunity to embrace the future, rather than reflect the past.
“It is our choices, Harry, that show what we
truly are, far more than our abilities”
You are more than a dataset. And, just as there was for Harry, there is more to your story.
#AIforGood #ExpectoPatronum
To learn more about AI in Alberta, visit www.atbalphabeta.com. Follow Chandra on Twitter @Chandra Rink.
