Latest News



The Value of Algorithms in Artificial Intelligence

Algorithms by themselves aren’t valuable. Their ability to solve a particular business problem is. The two most important questions to ask when assessing the value of algorithms in artificial intelligence are: 1) Can they solve the business problem? 2) Do people actually use the right conditions under which to perform their analysis?

Does their use of algorithms make a difference?

For example: Job candidates are often scored by algorithms in order to help human business leaders responsible for making hiring decisions. Sometimes recruiters who receive this machine-generated output don’t actually follow the recommendations provided by the algorithms. It’s as if a tree falls alone in the forest and no one is around to hear it. One could argue that it didn’t happen, or at least that the algorithms (represented by the tree falling) didn’t matter. If someone didn’t hear the algorithm’s output or listen to its recommendations, the algorithm by itself didn’t change the business decision at hand. Did the algorithms really do the analysis in the first place?

The value of algorithms in artificial intelligence depends upon their practical use.

Both of my fellow panelists mention the importance of closing the loop, and collecting data at all stages of the customer lifecycle, even when communicating about topics as complex as cancer medication and lifestyle. Doing a retrospective analysis and predicting an outcome is important, but it is simply the first step in ensuring algorithms used in artificial intelligence produce value.

I will give you an example of where over-reliance on machine-generated output–without a human-in-the-loop–fails. At RapportBoost, our team of data scientists performed an analysis on a customer with millions of chats between their live chat sales team and visitors to their website. Our algorithms determined a high correlation between an agent apologizing and a low customer satisfaction score. An unsupervised deep learning algorithm would assume causation and recommend that the live chat sales agent never apologize. Common sense, human experience and an understanding of our clients’ online retail business tell us, however, that many apologies took place because a customer was not happy. The customer would have reported a low score anyway. Algorithms are more valuable to artificial intelligence systems when humans can help interpret causation.

When one gathers data and does A/B testing to determine causal imprints instead of a backward looking quick analysis you get really deep and understand people on a much deeper level. This is where value is created by algorithms for use in artificial intelligence.

Learn more about the live chat agent training solutions from the team at RapportBoost.

The excerpt summarized above is from Dr. Michael Housman, Chief Data Scientist and Co-Founder of RapportBoost.AI, discussing the value of algorithms at the JMP Securities Conference panel on AI in February 2017.[/vc_column_text]