New technologies such as the rapidly growing deep learning models have given rise to increasingly complex artificial intelligence (AI) models. With expectations ranging from autonomous vehicles (land, air, and sea) to highly specialized information retrieval and rendering like ChatGPT, the possibilities seem endless. However, there are potential pitfalls such as material and energy issues as well as layoffs and privacy issues.

Every action a computer performs corresponds to electrical signals that run through its hardware and consume energy. Deep Jariwala, associate professor of electrical and systems engineering in the School of Engineering and Applied Sciences, and Benjamin S. Lee, professor of electrical and systems engineering and computer and information science, spoke with Penn Today about the impact of increased reliance on artificial intelligence computing. As infrastructure evolves to facilitate ever-increasing needs.

What sets AI and its current applications apart from other iterations of computing?

Jariwala: A completely new paradigm in terms of function. Consider the first computer we had at Penn, the Electric Digital Integrator and Computer (ENIAC). It was designed to do math that would take a lot of time to calculate by hand, and was mostly used to calculate ballistic trajectories, so there was a clear logic behind it: for example, the 10 digits entered were manually added, subtracted, multiplied, and divided.

Lee: There are three main parts of computing for AI. One is data preprocessing, which means organizing it before doing anything with a large data set. This may involve tagging or clearing data, but basically you’re just trying to create a struct within it.

After preprocessing, you can start “training” the AI; like teaching him to interpret data. Then we can do what we call artificial intelligence that runs the model in response to user requests.

Jariwala: AI is more about using complex algorithms and machine learning to learn and adapt to new information or situations, rather than calculating raw numbers. It’s not just about manually entering a value, as it can pull information from larger datasets like the Internet.

The ability to gather data from different places, use probability models to assess fit for the task at hand, integrate that information, and then deliver an incredibly human-like outcome in most cases is what sets it apart from traditional computing. Major language models like ChatGPT show this new set of actions when you ask him a question and put together a particular answer. It takes the basic premise of a search engine but increases its speed.