Meta, the parent company of Facebook and Instagram, said it is developing a personal artificial intelligence (AI) agent.
Meta Chief Executive Officer Mark Zuckerberg said on the 29th (local time) during a conference call after the second-quarter earnings release, "I think it is extremely unlikely that five years from now billions of people will not have a personal agent that understands their goals and works on their behalf around the clock to achieve them."
Zuckerberg said personal AI agents will be used in every area, including health, hobbies, personal finance, productivity improvement, relationship building, and career development, and predicted, "It will be a very important and massive market."
However, he stressed that consumer AI agents must be different from the developer agents currently in development. He said, "Developers have technical knowledge, so they are willing to invest time to get an agent to work (according to their intent)," adding, "But if you want to build a product used by billions of people, that product should just work." He explained that consumer-facing products should work as users expect from the start without going through a complicated setup process.
He said, "I think corporations that can provide personal agents that work well on their own will seize the next opportunity in AI." Meta operates the world's largest social media (SNS) platform, represented by Facebook, Instagram, and WhatsApp, so it believes it has strengths in building popular products and scaling them to large numbers of people, and it plans to target the personal AI agent market going forward.
On open AI models, which have emerged as a hot topic in Silicon Valley in the United States recently led by Anthropic, he said, "We will develop and operate open-weight (public-weights) models and closed models together." He said, "We see open models as an important part of the AI ecosystem and believe they have benefits."
Until last year, Meta had been developing the open-weight model Llama, but when its performance fell short of expectations, it reorganized the relevant unit, established a superintelligence lab, and recently shifted to developing closed models such as Muse Spark.