In this blog, I will discuss two fundamental approaches to using technology in businesses: One approach aims to replace the human in the process, as terminators are human looking machines, or to amplify humans, as Robocop is supported by technology.
Since this is a risk blog, let’s look at arguments to determine how a fully automated credit scoring system, represented by the Terminator, fares against the human hybrid version, represented by Robocop.
Let the boxing match begin!
Machines are emotionless: We, humans, get tired after long days of work. We might like or dislike a company based on our relationship with one person of that company. We make mistakes. Machines don’t make mistakes (or at least not these types of mistakes). Punch Terminator.
Machines pose systematic risks: Machines are in essence a set of rules. When one of the rules is wrong, then the machine is going to make the same mistake over and over again. The most prominent example of this category ought to be Knight Capital, a high frequency trading firm, who lost more than $400m in a single morning due to a software glitch. But even amongst commercial lenders, Wonga might have experienced similar problems due to lack of manual control. If your fully automated machine fails, then you will fail with it. Punch Robocop.
Machines are efficient: A machine can easily process a few thousand data points in less than a second. We, humans, can’t compete with that. After all, there is a business to run and nothing beats the machine’s cost per decision made. While the other arguments in this blog deal with the decision which is produced by the machine, we must not forget that machines enable lower cost products in shorter time frames. Punch Terminator.
Machines can’t think outside the box. All of today’s machines relate some structured input to some predefined output. Both input and output can be quite complex, as Google’s self-driving car demonstrates, but in the end, all possible outcomes had to be thought of by the designer beforehand. Coming back to the world of credit, machines don’t know how to investigate in case of inconsistencies (unstructured input) and as a result might misidentify a company. In terms of output, credit scores tend to be quite one dimensional, to lend or not to lend. Machines don’t propose solutions well, for example, giving the option to lend only against the pledge of additional security. Punch Robocop.
The round is over, how does the judge decide?
At MarketInvoice, we deeply believe in a hybrid approach to credit underwriting. We believe that credit scores provide the baseline for every decision made. But we also put considerable effort in making sure that the machine communicates the reasons for a given decision. This allows us to learn from and teach the machine – to become one entity, just like Robocop.