Sramana Mitra: In the context of what we are discussing, what is an example of something that is not doable without human intervention?
Peter Brodsky: Most things. Even if you take something as simple as looking at a form that has been filled out by hand. There is no clear way of knowing what it actually says. Handwriting recognition has the same challenges as speech recognition. Although you can get very good at it, you are going to make a number of mistakes.
When you operate in industries where mistakes aren’t tolerable, you really need to have extremely high confidence. For example in health service, mistakes cost lives. In financial services, it costs unimaginable amounts of money. In the government, you can upend someone’s life. This even includes the basic things like answering a question like, “What does it say on this page?” It’s not something that you want to leave in the hands of a machine without the ability for that machine to call for help. It just gets more and more complicated from there once you want action on that data. “Should we give this person a mortgage?”
There’re a lot of cases where it’s cut and dry, but in a lot of cases the documents submitted are about the assets of the person. That requires a certain amount of interpretation. Sometimes the machine is able to do that interpretation with extremely high confidence and at other times, it can’t. In that case, you want a person involved in every single thing that it does.
That said, the trick is to automate as much as possible. When you look at the impact that Hyperscience has today, we are able to automate roughly 95% of the work for our customers. For example, out of 100 people doing data entry, they will need our help in five. That’s huge labor savings.
Sramana Mitra: Let me see if I got this. Let’s say we are talking about a market application. Are you saying that your technology can reject market applications for 95%? The remaining 5% that should be considered for acceptance are the ones that get reviewed?
Peter Brodsky: It doesn’t break down like that. A much greater number of mortgage applications get accepted. 95% do not get rejected. It’s also not at the application level. It’s more in the unit of work level. For example, could we verify this person’s income? That’s a step in approving a mortgage. We verify this person’s assets. We appraise the value of the things they are buying. Those are steps in approving a mortgage. We automate 95% of the work.
We touch all the steps that we touch, but 5% of the time we say, “You know what? We actually cannot verify if this person makes this much amount of money or we can’t verify that this form says what we think it says. Did they check that checkbox or did they scratch it out? It’s too confusing, we need a person to look at it.” That happens about 5% of the time.
This segment is part 2 in the series : Thought Leaders in Artificial Intelligence: Peter Brodsky, CEO of Hyperscience
1 2 3 4