AI Needs Humans

AI Still Needs Humans In The Loop

Whose turn is it to sit in the front? If you’re a parent you’ll know the way that particular school run conversation tends to pan out at 8am every Monday to Friday.

And who hasn’t thought how amazing it would be to pop the kids into driverless cars, programme in the school address and have a second cup of coffee?

We are on the cusp of a science fiction world where AI rules.

But does it?

Artificial intelligence still needs human help” was the title of a recent article by Dmitry Matskevich in Forbes magazine and I agree with him 100%. In our industry no-where is that more apparent than in the development of affordability technology. Our Categorisation as a Service (CaaS) engine uses an open banking platform to categorise transactional information. It has been “trained” by processing millions of anonymised transactions so that it can accurately categorise and sub-categorise each individual item of spending and income into over 155 categories. But CaaS, like other AI engines doesn’t do its learning all by itself. There are millions of human hours involved, correcting the mistakes AI makes until the algorithms have been trained to get it right first time. Technology journalists have coined a phrase for this and call it “human-in-the-loop”.

At Castlight we have recruited some of the best “humans in the loop” on the planet and they have an extraordinary relationship with CaaS. Recently they have invested thousands of hours to teach CaaS a number of new functionalities. One of them is the ability to fine tune the information drawn from transactional data on salary. CaaS can now identity not just regular salary credits but also irregular bonus payments. That’s not to say that every broker will decide to factor in the bonus information, when making a lending decision but it gives him or her the choice. Different brokers have different risk appetites, different lending organisations have different risk policies. But when you are making a decision on lending hundreds of thousands of pounds, its a big advantage to have as much financial insight into your customer as possible.

And in just the same way that CaaS algorithms are only as good as the “human-in-the-loop” data scientists who train them, the CaaS insights are only as good a the “human-in-the-loop” brokers who decide what to do with them.

I believe its a brave new world for brokers. AI is providing new levels of financial insight but its the best brokers with the most knowledge and experience that will be able to use the insights to make the best decisions.

And by the way, if driverless cars don’t come around fast enough for you, the secret of school run sanity is a full week in the front seat and then shift around.