Amazon’s generative AI assistant, Amazon Q, is under scrutiny. Reports indicate hallucinations and data leaks, sparking debates on its readiness for corporate usage.
As concerns mount, experts emphasize the importance of thorough testing, potential regulations, and Amazon’s role in navigating these challenges.
Hallucinations and privacy issues emerge
Leaked documents reported by The Platformer reveal that Amazon Q is grappling with inaccuracies, including hallucinations and data leaks. The studies highlight the inaccuracy of large language models (LLMs) when connected to corporate databases. Analysts tracking the industry suggest that these issues render Amazon Q unsuitable for decision-making in a corporate setting.
Pareekh Jain, CEO of EIIRTrend & Pareekh Consulting, points out the limitations, stating, “If hallucinations are present, you cannot use them for decision-making in a corporate setting.” While Amazon positions Q as a work companion for millions, analysts question its readiness for widespread corporate usage.
Testing challenges and the importance of internal trials
To address these issues, experts stress the need for extensive internal testing before the generative AI assistant is ready for commercial release. Jain emphasizes the significance of evaluating data and algorithms to pinpoint the root cause of inaccuracies.
“I think they need to do more testing with internal employees first,” Jain added. “They have to see if it’s an issue with the data or the algorithm.” Amazon Q leverages 17 years of AWS’ data and development proficiency, emphasizing the stakes involved for Amazon in the rapidly evolving AI landscape.
Training and steps towards improvement
While hallucinations pose challenges, Sharath Srinivasamurthy, associate vice president at IDC, highlights steps to improve the use of generative AI. Srinivasamurthy suggests training models on higher-quality data, prompt augmentation, ongoing fine-tuning on organization-specific data, and incorporating human checks for suspicious responses.
“Training the models on better quality data, ongoing fine-tuning the models on the organization or industry-specific data and policies, and augmenting a layer of human check in case the response is suspicious are some of the steps that need to be taken to make the best use of this emerging technology,” says Srinivasamurthy.
Regulatory concerns and the call for responsible AI
Reports of hallucinations prompt discussions about the need for regulations, but Sanchit Vir Gogia, chief analyst and CEO at Greyhound Research, cautions against potential counterproductivity. Gogia suggests that overly stringent regulations could impede the exchange and utilization of data, pointing to the success of OpenAI’s GPT as an example of the benefits of a less regulated industry.
Jain echoes this sentiment, emphasizing the importance of self-regulation. “Regulations may exist, but the focus is primarily on self-regulation,” Jain explains. “The emphasis should be on responsible AI, where the logic can be explained to customers instead of creating ‘black box’ systems.”
As Amazon enters the generative AI space, all eyes are on the tech giant to address these challenges, especially considering its late entry compared to industry leaders like Microsoft and Google. Jain notes that AWS is a laggard, raising expectations and scrutiny regarding technologies like chatbots.