There is an old saying: “When all you have is a hammer, all of your problems look like nails.” In technology, there is a frequent habit of trying to use the latest, cool tech to solve all the problems we face. A few years ago, everyone was trying to use Blockchain, even if it was entirely inappropriate. Right now, everyone seems to want to use AI to solve whatever problem they face. Specifically, Gen AI is today’s technology hammer.

I don’t like the approach of saying, "How can we use AI to fix this?" I prefer to say, "We have a problem, and I think AI is the most appropriate solution.

Let me give you a specific example. One of the challenges I have faced in multiple roles is getting quality non-production data to test the applications that my team develops and supports. Many years ago, companies were willing to copy production data verbatim into the development environments and use that. This ensured, of course, that you have realistic data in terms of quantity and contents for your applications. It is a privacy nightmare and in breach of modern data compliance laws.

The result is that tests become less indicative of real-life data, and bugs that could have been caught before production deploy are only found afterwards.

It is possible for developers to write code that extracts, transforms, and loads production data into anonymized test data. However, this is a significant development effort upfront, and continuing maintenance overhead every time a database has changed needs to be reflected back in. Moreover, this method requires every test database to be the same size as production, and there are many scenarios where you want a subset.

Gen AI provides an attractive solution: A model can be developed that takes the schema and generates synthetic data which will reflect production data of appropriate quantity from a full-size dataset to a fraction of the size. The model can be rerun every time the production scheme changes, significantly reducing the developer maintenance and initial investment.

Gen.Ai is infamous for its hallucinations. Asked for the fastest marine mammal, one model answered it was a peregrine falcon: a creature that is neither marine nor a mammal. The reputational impact of associating your company with such gibberish is bad. With this use case, the output is internal and is by design fake.

There are commercial solutions that support this use case. This enables implementation of Gen AI to create synthetic data without needing to leverage your own development resources.

I think that with this use case we not only have a problem for which GEN.AI is a solution, but is the best solution.