Open Access to Training Data Vital for AI Safety and Innovation: Expert
The argument found support among other speakers, including Senator Maria Cantwell.
WASHINGTON, August 23, 2023 – An open ecosystem providing public access to artificial intelligence data is vital for the development of a safe and innovative AI system, am expert said at a forum on Monday.
Instead of the current “black box” approach to AI training, developers should adopt a transparent “glass box” approach, where they provide not only the data but also the models and step-by-step guidance for model replication, said Ali Farhadi, CEO of Allen Institute for Artificial Intelligence. This approach would enable developers to learn from each other’s mistakes, thus reducing the occurrence of repeated errors and associated costs, he explained.
The accessible dataset also serves as a critical “traceability” factor to assist lawmakers in crafting legal frameworks and safeguards against a multitude of risks posed by AI, ranging from misinformation, deep fakes to child safety concerns and workforce-related challenges.
“Looking back at the history of how software has been developed, whenever we actually opened up a piece of technology, the progress outpaced the malicious acts,” he added.
His argument found support among other speakers, including Senate Commerce Committee Chairwoman Maria Cantwell, D-Washington, who agreed that an “open architecture” has the potential to encourage a “public-private partnership” that could facilitate further advancements in AI development.
“We’ve been really working since the 2020 bill on understanding ways that we can accelerate our process to come to faster resolution of some of the issues that come to the table,” said Cantwell, who spearheaded the “The Future of AI” Act to convene leaders across academia, federal, and the private sectors to examine the opportunities and consequences of AI technology.
“I believe the government must continue to partner with industry and academia,” she added. “And public private partnership is the right direction for us to keep going.”
Hosted by Sen. Cantwell, the forum joined other lawmakers’ efforts to gain a deeper understanding of AI. The White House announced in August a competition with prizes up to 20 million as an incentive for developers to bolster the capabilities of AI systems. In late July, the administration also secured commitments from leading AI companies to oversee the safe and transparent development of the technology.
These initiatives are part of Washington’s effort to take the lead in the development of AI and maintain its technological competitiveness, especially as counterparts in Brussels and Beijing have been racing ahead in terms of regulations.
