More than 100 experts and risk assessors in the field of artificial intelligence issued an open letter warning that they do not have the necessary resources to test the safety of artificial intelligence technology.
Photo credit: Steve A Johnson/unsplash.com
The call aims to force core model providers to ensure that third-party AI evaluators have the necessary “scientific objectivity, transparency, independence and robust security” to do their jobs effectively and reliably.
“We are simply trying to demonstrate a common position on fundamental principles and ensure that independent oversight can be an effective tool for managing risks across the AI landscape,” stated in the interview CNBC Conrad Stosz, president of the Artificial Intelligence Evaluation Forum Alliance, which organized the letter.
Signatories of the letter include luminaries in the field of artificial intelligence such as British-Canadian scientist Geoffrey Hinton, as well as staff from organizations such as Johns Hopkins University, Stanford University and the nonprofit evaluation organization METR.
The niche evaluation community was thrust into the spotlight after Anthropic CEO Dario Amodei proposed over the weekend that independent experts have access to internal security procedures and incident reports to review and audit advanced reference models and their development processes.
Stosh said the consortium does not “advocate for any particular approach” to ensuring the safe development of AI models, but would like to see at least “breakthrough principles” and “a higher level of standardization” for evaluators and others working independently from large labs.
Stosh noted that Amodei’s proposal would provide greater access than evaluators have previously enjoyed. One scenario is where the base model creator allows outside experts access to company computers, allows communication with employees, and provides “Confidential internal material and unreleased systems.”
Experts also want to be protected from possible retaliation by the AI companies whose decisions they will evaluate.
Additionally, Storz said the third-party experts are not intended to replace the company’s internal model safety assessment process. “There are currently very few groups with the technical expertise, scale and capacity to do this kind of work,” – he added.
If you find an error, select it with your mouse and press CTRL+ENTER.
