Generative AI’s Potential and Challenges Discussed at Reuters NEXT Conference

In the year since the launch of ChatGPT, a sensation in the realm of generative AI, its impact on industry has been limited and largely experimental, according to top business, government, and civil society leaders speaking at the Reuters NEXT conference in New York. While the AI text generator has captivated users with its ability to produce everything from Shakespearean sonnets to academic essays, its tendency to generate inaccurate information, or “hallucinate,” has so far hindered its widespread application in industry.

“One key takeaway is the significant gap between being able to achieve something at a basic level and doing it well enough for a specific use,” said Anthony Aguirre, founder and executive director of the Future of Life Institute, a nonprofit dedicated to mitigating the potential catastrophic risks posed by advanced AI. He drew a comparison with self-driving cars, a technology that is still struggling to reach full deployment despite some level of functionality. “They’re not yet reliable enough to replace humans. This has proven to be far more challenging than initially anticipated.”

Sherry Marcus, director of applied science at Amazon’s AWS, noted that customers are at varying stages in their journey with generative AI. “There are numerous generative AI applications currently in production, while others are just embarking on their journey.”

One area where generative AI has seen wide-scale deployment is in writing computer code. Microsoft’s Github, an online code repository, has seen about half of its programming written with the help of an AI tool called Copilot that automatically suggests lines of code, according to Microsoft Corporate Vice President Lili Cheng. “Developers report increased productivity with Copilot,” said Cheng. “It’s an excellent example of how a generative model can leverage data within GitHub to enhance efficiency and make programming more accessible.” She also mentioned AI-generated summaries of meeting transcripts as another instance where the technology is demonstrating its value.

Financial professionals also told Reuters they are actively utilizing AI models in their businesses for tasks such as coding, generating documentation, and optimizing capital deployment. However, they are proceeding with caution due to the heavily regulated nature of financial services.

Gary Marcus, a professor at New York University, acknowledged that generative AI, like the AI images generator, is prone to errors in coding as in other areas, but this issue is less problematic in the tech sector where programmers are equipped to troubleshoot. “Coding is the area where it’s really making waves because coders know how to rectify the errors these systems generate,” said Marcus. “However, in almost any other type of business, these hallucinations pose a serious problem.”

Executives emphasized that companies should be slow and methodical when integrating the technology into applications where accuracy is paramount. Vijoy Pandey from Cisco believes that AI has shown its value for “the low-hanging fruit,” or applications where “the cost of being wrong has been relatively low.” The next challenge, he said, is transitioning the technology to more sensitive “business-critical use cases,” such as legal and security.

“We should anticipate that people will make mistakes,” said Pandey. He stressed the importance of developing technology, guidelines, and frameworks “to safeguard everyone against ill-advised actions” in the coming years.

In the latest AI news, it’s clear that while generative AI, including AI video generator tools, has shown promise, it remains largely experimental and its impact on industries is still limited. However, as AI tools continue to evolve and improve, their potential to revolutionize various sectors cannot be underestimated.