Trust in enterprise AI starts way back

Being the first one to implement a Machine Learning algorithm for the first time in a company comes with good questions. Nontechnical folks will need to begin by understanding ML and the art of the possible. Although a great number of questions will come just before the go-live. They’ll need to test and be sure that the algorithm works and will perform as expected.

That’s the point where this kind of technology and tech start to intersect. People will be afraid of something they don’t understand and also about something that is non-deterministic (little changes in inputs can change the outputs). There are different ways of building trust in these algorithms, some are sexier than others (spoiler alert, you might want to start with the unsexy ones).

The first way to build trust is simple: test a lot. I mean a lot. Throw a lot of data to test and check the results. These will grow confidence in the nontechnical folks. The caveat is that this could take a lot of time, manual spot-checking and you might not have enough data available.

Other way is explainability. There is a whole AI field working on methods to explain the algorithms and be able to tell what went wrong when it outputs a wrong prediction. This is something worth checking out, but you might need to combine it with the first one.

The third and last way is data governance and data integration. A lot of companies have multiple sources of data (with duplicated data), nonexistent quality controls and zero governance. This means that data is not available, is hard to find and nobody is responsible for the data. If you are not able to trust the data you are feeding to the model, you will not be able to trust the results. Even for testing purposes, you need ground truth data, if it’s not grounded everything might be more difficult.

A lot of companies might want to run before they can crawl and that’s ok. I think ML projects are a great excuse to start doing some work on data governance and data integration. Just try not to forget about these two things. If you set the right foundation the future projects will come smoother.



One thought on “Trust in enterprise AI starts way back

  1. Pingback: Trust in enterprise AI starts way back – Full-Stack Feed

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s