Change the Path of AI Development

Who benefits from artificial intelligence? This fundamental question, especially relevant in recent years, was at the forefront of a conference at MIT, where speakers and attendees explored various dimensions of AI's impact. In one of the keynote addresses, journalist Karen Hao called for a shift in AI development, advocating for a move away from the massive scale-up of data use, data centers, and models associated with 'artificial general intelligence.' She emphasized that this scale is unnecessary to realize the benefits of AI.

Hao, who has become a prominent voice in AI discussions, stressed that if we genuinely want AI to be broadly beneficial, we urgently need to abandon the current approach. She highlighted the astonishing size of datasets used by major firms to develop large language models, as well as the trade-offs of such scaling, including massive energy consumption and emissions from hyper-scale data centers.

As an alternative path, Hao pointed to AlphaFold, the Nobel Prize-winning tool for identifying protein structures. This represents the idea of a 'small, task-specific AI model tackling a well-scoped problem that aligns with AI's computational strengths.' She noted that it is trained on highly curated datasets relevant only to the specific issue at hand.

In a second keynote address, scholar Paola Ricaurte underscored the need for purpose-driven AI approaches, outlining several conceptual keys for evaluating the usefulness of technologies. She remarked that there is no point in having technologies that do not respond to the needs of the communities that will use them.

The conference at MIT gathered over 300 people and included discussions and design sessions across various fields. Hao encouraged attendees to be active participants in AI discourse, emphasizing that the trajectory of technology is not yet fixed and that public interventions matter.

In conclusion, both Hao and Ricaurte urged attendees to take proactive roles in AI matters, noting that technologies will perform best when they address the pressing everyday needs of all citizens.