Databricks CEO Ali Ghodsi argues that AGI has already been achieved, but most companies lag in adoption due to context, security, cost, and IP lock-in. He presents a pragmatic vision for enterprise AI centered on an organizational ontology that understands people, data, and meetings. Ghodsi dismisses the existential superintelligence risk from recursive self-improvement, outlining four conditions that are not yet met, and focuses on the tangible cyber risks and real-world applications like drug discovery.
Summarized by Podsumo
Ghodsi claims AGI is already here based on the original definition from his Berkeley lab days, but says most enterprises only use AI as a 'glorified Google search' or basic coding assistant.
He introduces 'Genie,' Databricks' internal AI that builds an organizational ontology from meetings, emails, Slack, and docs to give CEOs ground truth—replacing detective work with data.
The CEO identifies four barriers to deep enterprise AI adoption: capturing full context, solving cybersecurity, controlling exponential token costs, and avoiding vendor lock-in.
Ghodsi presents four conditions needed for recursive self-improvement (RSI) that are not yet fulfilled: faster training, fewer resources, broad accuracy gains, and sustained iterative loops.
He provides a concrete example: by correlating employee survey data with manager performance, Genie debunked the assumption that long hours cause low engagement, revealing a nuanced organizational insight.
"I actually think we already have artificial general intelligence. I've been saying it since last year... We have all the intelligence we need to completely transform the planet."
"Just because 98% of the code is being written by AI, we don't have RSI. If it automates the last two percent, who cares?"
"You need to really tackle all four of those [context, security, cost, lock-in]. Only then can you start getting something out of it."