The AI Revolution Is Meeting Reality | Commentary

A week with Amazon’s AI team and its partners revealed plenty of enduring optimism, but also a recognition that building effective generative applications takes work.

Nearly two years after ChatGPT’s debut, AI hype is giving way to reality. Companies are eager to build with generative AI, but they’re learning that doing so is hard. They’ve found that AI models are expensive, data conundrums abound, and “change management” isn’t so simple. To that end, only 21% of companies surveyed by Gartner earlier this year had GenAI in production, with the rest either “piloting” or “exploring” the technology, per data viewed by Big Technology.

Still, there’s almost unprecedented optimism around the tech. Every company with a pulse is examining how to integrate GenAI into their internal operations and external product. They’ve spent billions with Big Tech companies and consultancy firms as they race to figure it out. And they believe all this experimentation will eventually pay off. It better, because the economic future of the current AI wave depends on it. 

I spent a good chunk of the week discussing these ground-level realities with Amazon’s AI team and its partners and walked away with what might’ve been my best picture yet of what’s happening. I was struck by the measured tone of nearly everyone I spoke with. “It’s going to feel a lot more incremental than we’re probably used to,” Matt Wood, Amazon Web Services’ VP of AI products, told me, while insisting it will add up over time. And I learned about a few surprising products that expanded my view of the cutting edge. Here’s a breakdown of the core obstacles, and what surprised me on the product side.

Change management 

Despite estimates that 7 million+ people pay OpenAI $20 month for premium ChatGPT, generative AI will be most valuable for enterprises, at least in the near term. For businesses, the use cases are clearer, as are the returns. But for companies to find success deploying AI, employees will need to embrace new internal tools, changing workflows, and increasing automation. And leadership will have to apply the technology to clear problems vs. implement it for implementation’s sake.

“A year ago, you heard a lot of customers say, ‘Tell me more about generative AI, how can I use this?’ And they set aside budgets without having any idea of what they were going to use them for,” Ruba Borno, vice president of worldwide channels and alliances at Amazon, told me. “Without clarity of the ‘why,’ it becomes very difficult to move forward with the ‘what’ and the ‘how.’”

Valerie Henderson, president of AWS consulting partner Caylent, also spoke of the seriousness of change management challenge. “You cannot underestimate it,” she told me. “I had dinner with a customer last night, we were talking about this, and he said the biggest fear is you build this thing and it has one hallucination — it doesn’t give you the right output — and people ‘quiet quit’ using it.”

The significance of the change management issue drove my skepticism when endless press releases claimed AI would replace divisions or revolutionize companies last year. But now, some are starting to figure it out. Klarna CEO Sebastian Siemiatkowski, for one, recently convinced me he was able to turn a sizable amount of customer service over to LLMs (more on that soon). Successful deployments are rare today, but could be more common as the technology — and organizational readiness — progresses.

Model costs and ROI

AI applications can look great at prototype stage, but can be expensive to run when pushed to 100% of users. This is another issue holding back GenAI’s broader rollout — but it isn’t likely a long term-problem. Already, some companies are finding use cases with clear returns that justify the spend. One energy company, per Borno, built a GenAI tool to assess its tax needs in different countries and is already saving itself hundreds of millions of dollars from the output.

Broader utility will come as AI model providers lower costs, potentially bringing them close to nothing. “Everyone should be getting ready for the cost of intelligence to go to $0.00,” said Logan Kilpatrick, a current Google and former OpenAI employee. “It’s coming sooner than you would expect.” 

The data issue

Generative AI works best with well-organized data. That’s why some of the companies benefitting most today are from highly regulated industries like financial services, healthcare, and life sciences, which often have a ton of well-structured data. 

Amazon’s Matt Wood shared the example of a life insurance company with reams of policies for 90-year-olds that will likely soon pay out. “They’ve been scanned at some point, but no one’s ever read them,” he said. “And so they’re able to use generative AI to be able to piece that risk together and understand it more completely.”

This application of GenAI is admittedly more boring than the revolutionary, world-altering capabilities some in the tech industry are promising. But the productivity gains from turning this work over to the machines could be massive. “You actually want a lot of that boring work to be automated,” said Wood. “You want to be able to channel the boring work, which maybe inside some organizations is seen as a bit of a cost center, and to be able to turn that on its head and channel it into something which drives invention and drives growth.”

New use cases 

The most surprising product I heard about was a generative AI tool an automotive company built to help diagnose car issues using words, pictures, and sound. The AI solution they built ingested the documentation from the cars’ user manuals, including pictures, and also trained on noises cars make when they need service.

When the product ‘listens’ to the cars, it can suggest fixes to the service center. “This is working in production at three dealers,” Alan Chhabra, EVP of worldwide partners and international sales at MongoDB, which worked on the product, told me. He wouldn’t reveal the name of the company.

As the latest AI models get out in the world — with better intelligence, multimodality, and better economics — more applications like this are likely to emerge. And that will probably help drive up the 21% of companies with generative AI solutions in production. Even if it may be a bit longer than the narrative suggests. “The more we get customer stories where it’s worked in production,” said Chhabra, “the more likely customers will take the risk of going all the way and getting an ROI.”

This article is from Big Technology, a newsletter by Alex Kantrowitz.

The post The AI Revolution Is Meeting Reality | Commentary appeared first on TheWrap.