Sean Malvey

Gartner predicts that by 2024, more than 

 of AI model training data will be synthetic. 

Some might think this bring us one step closer to the Matrix, but I remain cautiously optimistic, and you should too.

Synthetic data, if managed correctly, can positively impact industries that struggle to collect organic data. 

For example, healthcare workers will be able to augment real patient data for training doctors, and finance professionals can improve fraud detection by expanding their data models and testing for abnormal transactions. 

But what about IT? If you’re tasked with managing your company’s digital experience, should you incorporate synthetic data?

I think if you’re looking to develop and improve your own internal applications, synthetic testing might be for you. 

Instead of bugging employees to test out features in your app, you can use a synthetic-fed AI tool and not risk jeopardizing those relationships.

And before you release the application into the wild, you can run thousands of simulations verifying that its functions are performing correctly and that its code is flawless.

Most importantly, synthetic testing permits you to stress test 

 capacity and performance. Instead of trialing how one application performs for one employee, you can exercise how it performs across thousands of stand-in employees. 

It remains to be seen how synthetic datasets will inform large language models. Ultimately, IT leaders need to remember that what they’re looking at isn't real.

You should also be cognizant about the inherent biases that come with this innovation. AI and large language models aren’t great at correcting for their own 

Ultimately, IT leaders should aim to collect and understand real human sentiment in the context of the digital workplace, in addition to the hard data they obtain.

You don’t want to unintentionally hide behind a blanket of false security. 

For example, imagine you’ve just rolled out a digital transformation project and all of the hard data (network connections, RAM, boot time, etc.) comes back positive. Should you declare victory? What about the employees interacting with the tool? How would they perceive your definition of success?

These are questions that synthetic data, at the moment, cannot answer. I’m hopeful, however, that one day we’ll get there, but until then, I’ll stay cautiously optimistic. 

Gartner predicts that by 2024, more than \n60%\n of AI model training data will be synthetic. \nSome might think this bring us one step closer to the Matrix, but I remain cautiously optimistic, and you should too.\nSynthetic data, if managed correctly, can positively impact industries that struggle to collect organic data. \nFor example, healthcare workers will be able to augment real patient data for training doctors, and finance professionals can improve fraud detection by expanding their data models and testing for abnormal transactions. \nBut what about IT? If you’re tasked with managing your company’s digital experience, should you incorporate synthetic data?\nSynthetic Testing for App Development\nI think if you’re looking to develop and improve your own internal applications, synthetic testing might be for you. \nInstead of bugging employees to test out features in your app, you can use a synthetic-fed AI tool and not risk jeopardizing those relationships.\nAnd before you release the application into the wild, you can run thousands of simulations verifying that its functions are performing correctly and that its code is flawless.\nMost importantly, synthetic testing permits you to stress test \nload\n capacity and performance. Instead of trialing how one application performs for one employee, you can exercise how it performs across thousands of stand-in employees. \nOkay, But Pump the Brakes \nIt remains to be seen how synthetic datasets will inform large language models. Ultimately, IT leaders need to remember that what they’re looking at isn't real.\nYou should also be cognizant about the inherent biases that come with this innovation. AI and large language models aren’t great at correcting for their own \nisms\n (racism, sexism, ageism, etc.). \nUltimately, IT leaders should aim to collect and understand real human sentiment in the context of the digital workplace, in addition to the hard data they obtain.\nYou don’t want to unintentionally hide behind a blanket of false security. \nFor example, imagine you’ve just rolled out a digital transformation project and all of the hard data (network connections, RAM, boot time, etc.) comes back positive. Should you declare victory? What about the employees interacting with the tool? How would they perceive your definition of success?\nThese are questions that synthetic data, at the moment, cannot answer. I’m hopeful, however, that one day we’ll get there, but until then, I’ll stay cautiously optimistic. 

Synthetic Data? I’d Tread Carefully

Synthetic Testing for App Development

Okay, But Pump the Brakes

Request a Nexthink Demo