Will Models Rule the World? Data Science Salon Miami, Nov 6-7
This post is excerpted from the thoughts of Data Science Salon Miami speakers on the future of model-based decision-making.
There’s no doubt that we live in a world that is quantified and powered by models. And according to our panel of experts, this is largely a good thing. "A model should fit the data and the problem one is trying to answer,” said Colleen Farrelly, Data Scientist at Graham Holdings. “When models fit projects, the potential for real insight into difficult problems exists, particularly within economics and finance data." Lloyd Reshard, CEO of Cognitive Big Data Systems, agrees that models are an inextricable part of the modern economy. “Models are an important key to decision making, especially in the research, engineering and business fields, and especially when you are building models from experimental, test, or marketing campaigns or financial data.”
But why have models become a go-to strategy for business answers? “Modern businesses produce an immense amount of accessible data and it only makes sense to complete the feedback loop using decisions from data driven models,” said Sreya Ghosh, Data Science Manager at Restaurant Brands International. “This quantitative feedback loop ends up evaluating the decision itself, allowing timely edits in the decision.” Madhav Khurana, Senior Data Scientist at Careem, agrees, “Human judgement is highly biased and can lead to decisions which won't necessarily work in all situations. Even if we do manage to overcome these inefficiencies of human judgement, it is also important to make decisions in real time when there are high number of transactions - be it predicting fraud in banks, cancellations in bookings or the next move of a player in a game.”
But despite the power and wide adoption of models, there are still some limitations. “Models are limited by the data that is chosen, the quality of the data, and how well it represents the range of values for each input,” according to Reshard. Ghosh goes further, “Biases in the data consequently bias models and decisions. And it can’t disrupt - creative decisions still need the human brain.”
And models can also be extremely resource intensive as well. Khurana says, “The most obvious limit of model-based decision making is that it needs data and a lot of it.
My viewing behavior on Netflix may lead a model to decide that I like comedies, but because I have never watched a Korean action thriller, neither me nor the model will ever know if I'd like it. One skill that humans have but a machine learning model doesn't is the skill of improvisation - taking action when something cannot be predicted.”
In the end, Khurana says, maybe it’s best to live by this quote from Jim Barksdale:
“If we have data, let’s look at data. If all we have are opinions, let’s go with mine.”
Where can I learn more?
Make sure to catch all of our speakers at the Data Science Salon Miami on Tuesday, November 6 and Wednesday, November 7, 2018. Tickets are almost sold out, so pick yours up today! Follow this link for 20% off tickets, exclusive to KDnuggets readers.