Artificial Intelligence is at the phase where federal and commercial technology leaders are amazed by novel model architectures. Each one can perform a function like data preprocessing, feature extraction, or prediction. The demonstrations are impressive; today we look at the data behind the dazzle.

Adam Kowalski from proximal has spent years helping organizations follow a data-centric AI approach. During the interview, he expands on having high-quality, relevant data has precedence over any model architecture.

One weakness of generative AI systems is they can be subject to bias. You may have insufficient data. Taking a data-centric approach can eliminate many of the errors we are seeing in AI results.

The results of varying from this precept can be disastrous as well as entertaining. In the commercial world, this can result in millions of dollars of loss. In the federal government, there can be much more grave consequences of ignoring the data.

It is not all serious. Adam Kowalski mentions an entertaining term from the AI community. If you ask a public language model a question about ancient Rome, you may discover a key player was Abraham Lincoln. This is called a “hallucination.”   

= = =

Want to leverage you next podcast appearance? www.podscorecard.com

Connect to John Gilroy on LinkedIn   https://www.linkedin.com/in/john-gilroy/

Want to listen to other episodes? www.Federaltechpodcast.com