DRLearner is Released as Open-Source Code — Democratizing Public Access to State-of-the-Art Software for AI/Machine Learning

The 15th annual Artificial General Intelligence (AGI) Conference opens today at Seattle’s Crocodile Venue. Running from August 19-22, the AGI conference event includes in-person events, live streaming, and fee-based video access—and features a diverse set of presentations from accomplished leaders in AI research.

As the AGI community convenes, it continues to promote efforts to democratize AI access and benefits. To that end, several AGI-22 presentations will officially launch DRLearner—an open source project to broaden AI access and innovation by distributing AI/Machine Learning code that rivals or exceeds human intelligence across a diverse set of widely acknowledged benchmarks. (Within the AI research community these Arcade Learning Environment [ALE] benchmark tests are widely accepted as a proxy for situational intelligence.)

“Until now, tools at this level in ‘Deep Reinforcement Learning’ have been available only to the largest corporations and R&D labs,” said project lead Chris Poulin. “But with the open-source release of the DRLearner code, we are helping democratize access to state-of-the-art machine learning tools of high-performance reinforcement learning,” continued Poulin.

Ben Goertzel, Chairman of the AGI Society and AGI Conference Series, contextualized DRLearner as well-aligned with the goals of the AGI conference. “Democratizing AI has long been a central mission, both for me and for many colleagues. With AGI-22 we push this mission forward by fostering diversity in AGI architectures and approaches, beyond the narrower scope currently getting most of the focus in the Big Tech world,” Goertzel said.

DRLearner project presentations include:

“Open Source Deep Reinforcement Learning” General Interest Keynote presented by Chris Poulin, Project Lead. (Journalists Note: Poulin’s initial keynote is scheduled for Sunday, August 21. On this day the AGI-22 Conference is open to the general public.)

“Open Source Deep Reinforcement Learning: Deep Dive” Technical Keynote by Chris Poulin and co-principal author Phil Tabor. (Monday, August 22)

“Demo of Open Source DRLearner Tool” Code Demo by co-author Dzvinka Yarish (Monday, August 22)

Poulin also noted the importance of managing expectations on the benefits on what DRLearner will, and will not, provide in its initial Beta release: “Fully implementing this state-of-the-art ML capability requires considerable computational power on the cloud, so we advise implementors to maintain realistic expectations regarding any deployment”. DRLearner’s benefits could be substantial, however, for the numerous organizations who have substantial computing budgets: analytical insights, expanded research capability, and perhaps a competitive advantage. “And for those whose professional lives are focused on AGI, this is an exciting time, as DRLearner can enhance their neural network training efforts…” Poulin said.

Drawing on his working experience with both US and Ukrainian computer scientists and software developers, Poulin assembled an international team of expert developers to complete the open-source project. (See more about ‘DRLearner’s International Dev Team’ below.)

A final noteworthy addition, is that the work of Poulin et al was advised by Adria Puigdomenech Badia of DeepMind. “DRLearner provides a great implementation of reinforcement learning algorithms, specifically including the curiosity approach that we had pioneered at DeepMind,” said Puigdomenech Badia. Poulin likewise had high praise for the DeepMind’s prior “Agent 57” achievement: “Agent 57 was one of a limited number of implementations (at Deep Mind) that consistently beat human benchmarks. And due to the elegant simplicity of its particular design, and help of Adria, it was the best candidate to inspire our software implementation,” Poulin said.

ON ARTIFICIAL GENERAL INTELLIGENCE & THE AGI CONFERENCE GOALS

The original goal of the AI field was the construction of “thinking machines”–computer systems with human-like general intelligence. Given the difficulty of that challenge, however, AI researchers in recent decades have focused instead on “narrow AI”–systems displaying intelligence regarding specific, highly constrained tasks. But the AGI conference series never gave up on this field’s ambitious vision; and throughout its fifteen-year existence AGI has promoted the resurgence of broader research on “artificial intelligence”—in the original sense of that term.

And in recent years more and more researchers have recognized the necessity – and feasibility – of returning to the original goals of the field. Increasingly, there is a call for a transition back to confronting the more difficult issues of “human level intelligence” and “artificial general intelligence (AGI).” AGI leaders are committed to continuing the organization’s longstanding leadership role—by encouraging and exploring interdisciplinary research based on different understandings of intelligence.

Today, the AGI conference remains the only major conference series devoted wholly and specifically to the creation of AI systems possessing general intelligence at the human level, and ultimately beyond. By convening AI/ML researchers for presentations and discussions, AGI conferences accelerate progress toward our common general intelligence goal.

Other articles