Mastering The Risk Factor: Would You Let AI Choose Your Spouse? (Anna Slodka-Turner) PlatoBlockchain Data Intelligence. Vertical Search. Ai.

Mastering The Risk Factor: Would You Let AI Choose Your Spouse? (Anna Slodka-Turner)

Artificial Intelligence (AI) is at the forefront of many conversations across industries. And why not? It’s brought us extensive solutions, saving humankind so much time. But like everything good, it has limitations, particularly general AI, which often
feels like a catch-all term for a general algorithm accessible via some speakerphone that can do anything.

When AI is hyped as a solution for so many things, it makes me think, how far can you drive that hype? A famous talk from ‘School of Life’ on ‘Why you will marry the wrong person’ inspired a question, ‘Would you
let AI choose who you marry?’ Could it help make the right marital choices?

While AI can’t fully answer extremely complex relationship matters, it can get you significantly closer to finding the answer. We come across this quite often in the financial world. Is AI able to predict the next deal? The answer is no, that’s not yet possible.

However, AI can be used to build models with enhanced analytical and forecasting capabilities, providing much deeper insights and uncovering patterns to provide a clearer idea of what’s coming.

Applying AI to Decisions

Let’s consider this in the context of decision-making. In a simple way we have two types of decisions:

 – Ones we make frequently, and therefore with plenty of feedback loops. Eg: Buying milk. It took my family a few months to discover we need four bottles a week that is unless it’s cold and, on the weekend, when everyone needs a few extra ‘warm cuppas’.
AI could have potentially solved it for us sooner, as long we as fed it the weather data to spot the pattern.

-The second type of decision is the ones we make infrequently. Possibly, only once in a lifetime with little chance to make a correction based on the outcome of our decision. Eg: Choosing a profession, university degree, first job, or
LOL, deciding to get married.

Of course, we live with the consequence of our choices, but the opportunities to learn from them and make other decisions are limited and often costly.

A parenting book I read holds a caveat along these lines: “While we support the parenting advice in the following chapters, we acknowledge it is not possible to try different parenting methods on a child and compare the outcomes”. Simply put, there is no
way to try different decisions and compare outcomes. Just another thing that shows parenting is hard.

And it illustrates how important it is to have enough data to see patterns.

Machine Learning Challenges

Machine learning, a popular form of AI, has for a while been seen as a ‘magical solution’ to complex problems. The attraction of it being able to absorb plenty of data and try to find sense in it has a certain appeal. Why wouldn’t it? The promise of technology
taking something complex and coming up with the best solution would appeal to any decision maker.

The challenge of machine learning solutions is helping make a simple decision from complex input information; incredible amounts of data, internal and external, and then how  the  output is communicated. . In the above examples of two types of decisions,
machine learning algorithms would hopefully solve the milk buying question rather quickly.

Assuming that we provide the data on the quantities purchased and the weather outside – the model would create a good forecast going forward. Organizations like tourist destinations, restaurant chains, airlines, logistics companies and many more receive
analytics that can be used to predict daily, weekly, and seasonal volume based on the weather, and even recommend how many resources they might need to meet that demand. Additional variables add more complexity to the model and creates potential additional
need to answer other questions and add more variables (e.g., weeks that the cleaner comes vs not).

Back to the core question of allowing AI to decide whom you marry. Surely, there are plenty of data points – hundreds of millions or billions of marriages. The relevant inputs have been studied for centuries both by researchers and matchmakers. There are
plenty of outputs.

So, what’s the problem?

  1. While there are many data points, each unique decision maker will have their unique preferences – so in the modelling world, we would need to create a different algorithm for each person who needs to be matched for matrimony. This is complex, but possible
    in the future. Consider how recommendation engines like Apple Music and Pandora continue to evolve the types of music they suggest to you based on your reactions.  Such solutions where each decision is made by a uniquely optimised model are already deployed
    in the business world.
  2. Secondly, we need to capture the right and relevant data points and reduce the ‘noise’. While some may prefer blue-eyed brunettes or brown-eyed blondes, there is little to prove marriages based on “preferred types” are more successful than others. Dating
    apps continue to hone their algorithms in hopes of finding the right formula for such matches. Still, you have to go on the dates and see.
  3. Lastly, the cost of making the wrong decision is high. While leaving it to the individuals making decisions may not yield the best outcomes, an expert team building a machine learning solution may not want the responsibility for making these decisions.
    There is a career liability risk that needs to be worked out. In the business context – it may be better to let the experts decide than to insist the ‘black box’ knows best.

Avoiding Blind Trust

So, back to the challenges of matrimony. The famous speech by School of Life simply states that we will of course marry a person that is in some ways wrong for us. “The person who is best suited to us is not the person who shares our every taste (they don’t
exist), but the person who can negotiate differences in taste intelligently — the person who is good at disagreement.

Rather than some notional idea of perfect complementarity, it is the capacity to tolerate differences with generosity that is the true marker of the ‘not overly wrong’ person. Compatibility is an achievement of love; it must not be its precondition.”

Moving to a broader general context, in the language of machine learning – pretty much none of the standard variables we know ahead of time about a potential candidate could help us predict if the decision is incorrect. We are far away from ‘feeding the
machine lots of data’ and expecting it to make sense of it. In fact, it may never happen without human intervention. We feel safer when the pilot switches the autopilot off during turbulence, and for a good reason.

While machine learning and AI can make our lives easier, it’s safe to say we wouldn’t blindly trust these technologies to make life-changing decisions for us. Taking from that, what can we say to industry experts making important business decisions? Use
AI and ML to take you halfway to your goal – but hold on to your experts to analyze the data and use their best judgment with context to guide you in the final steps. We sure are working on it.

Time Stamp:

More from Fintextra