‘Gen AI Zeitgeist’ On Display at AWS Summit

AWS SUMMIT — Experts are looking at generative AI from all sides at the AWS Summit in Washington D.C., this week. Most agree it will change our lives, but they point out the rapidly evolving technology must be handled responsibly for those changes to be positive.

An AWS Summit panel discussion of AI-ready infrastructure on Wednesday explored the need for sustainable infrastructure to support the growth of generative AI. Speakers said better designs of devices such as servers, storage, networking and semiconductors will help, but so can responsible human use of gen AI.

“We have not yet seen what gen AI can do to us,” said Sanjay Podder, global lead for technology sustainability and innovation at Accenture. “It's like the internet, and we know what happened with that — our lives changed. With gen AI, enterprises will reinvent themselves. Society will reinvent itself. All that will be possible because there is a lot of innovation happening in hardware and software. What we need to keep in mind is that none of us in this room knows what the world will look like. We should be keeping humans at the center of this transformation. We have to keep the environment and society in mind, so that AI does not disenfranchise communities and people. There will be a lot of innovations, but let's make sure that environment and sustainability and responsibility are central themes as we go through this journey.”

Neil Thompson, director of MIT’s FutureTech research project, also weighed the benefits versus the costs of gen AI.

“We have a bunch of costs that are associated with using AI,” he said. “But there are a bunch of benefits as well. Some of the modelling that my lab has done shows that AI can have huge benefits on the economy, and even more so in research and development to make discoveries. They may come with some cost, but then we can think carefully about what we are going to do mitigate the fact that these costs exist, and how we can deal with it."

The value of gen AI in research was a topic during a panel discussion on gen AI in health care.

“In five years, we’ll look back at 2024 as the very beginning of a huge transformation,” said Francisco Azuaje, director of bioinformatics at Genomics England. “We probably underestimate the opportunities, challenges and risks we face. The gen AI landscape will be radically different."

Adam Resnick, a director at Children’s Hospital of Philadelphia, said gen AI will transform health care “from being reactive to being preventive," he said. "We can set the stage for lifelong health care underpinned by data.”

But even the health care experts spoke of technical impediments.

“The biggest bottlenecks are hardware and power, and training models are still expensive,” said Dr. Praveen Meka of Dana-Farber Cancer Institute. “Once that’s overcome, it will be transformative in every aspect of health care.”

Thompson and Podder said more efficient use of generative AI by humans can help limit the power demands AI places on infrastructure.

“You absolutely have this larger appetite for resources that come in with AI,” Thompson said. “People are working hard to build [large language] models that are smaller so you can put them on your phone. You still have the energy used in your phone, but it’s not like creating a model that uses tens of thousands of GPUs.

“So I think that's one of the things that we can be thinking about here, is the efficiency. But the trade-off we have there is, if you make your models bigger, they become more powerful. And the big folks in AI are in the race to scale up these models. Scaling up does produce remarkable changes. Even the difference between, say, ChatGPT and GPT-4, there’s a big jump up. So that's quite exciting, but it does come with all the resource questions that we're having.”

Podder agreed that while large generative AI models provide more accuracy, smaller models are sufficient for many use cases.

“Most of the LLM providers today are giving you LLMs of different sizes,” he said. “The intelligent thing to do is select a model that is good enough for your business use case. You don't necessarily have to use the largest. That way, the energy generated is substantially lowered in the process.”

He said the same is true for inferencing, which takes up a great deal of energy.

“Again, simple techniques can be used to lower energy use,” Podder said. “If you batch your prompts, you can get your answers more quickly. So you don't have to go and query the large model again and again and again.”

During the main AWS Summit keynote, representatives of the CIA and U.S. Army joined AWS VP of worldwide public sector Dave Levy onstage to discuss their gen AI use.

“We were captured by the generative AI zeitgeist just like the rest of the world a few years back,” said Lakshmi Raman, CIA director of artificial intelligence.

CIA's Lakshmi Raman

She said the CIA uses gen AI for tasks such as search and discovery, writing assistance, and to generate ideas.

“It helps our analysts go through data quickly,” she said. “Imagine all the news stories that come in every minute of every day from around the world. We are neck deep [in data].”