June 2022

AI2 Newsletter

A representation of the Unified-IO model showing icons of text, images, and bounding boxes on the left side in a field of magenta, and a field of similar icons on the right side on a field of blue, with the Unified-IO logo of a bicolor circle within a ring in the center. The magenta field is labeled 'input space' and the blue field is labeled 'output space'. The diagram is meant to convey how Unified-IO can successfully handle many different inputs and outpus.

Top story

A new general-purpose model with unprecedented breadth

How can we bring the power of transformer models to bear on complex data, like images? That's the challenge the PRIOR computer vision team at AI2 set out to meet with Unified-IO, a new general-purpose model that can perform dozens of AI tasks using a variety of inputs and outputs thanks to sequential processing.

Oren Etzioni reflects on the progress of AI, and his time at AI2

"AI can fool some of the people all of the time and all of the people some of the time, but that does not make it sentient or intelligent." AI2's CEO Oren Etzioni, who recently announced his departure from the institute this fall, spoke with Analytics India Magazine about the progress he has seen in the field of AI, where things are headed, and his time with Paul Allen and AI2.

Training your virtual robot just got way, way better

Our PRIOR computer vision team just announced ProcTHOR, a new framework that can procedurally generate many interactive 3D environments. This breakthrough enables the training of embodied AI agents to scale by orders of magnitude. One of our researchers describes ProcTHOR as feeling "like the equivalent of moving from MNIST to ImageNet for Embodied AI."

Semantic Scholar Academic Graph Updates

Semantic Scholar recently announced the availability of new datasets that give users access to the full range of data exposed in the Semantic Scholar Academic Graph (S2AG) API, over the entire corpus of papers in the system. These updates allow users to fetch data for a large number of papers according to their own criteria without any data discrepancies.

EarthRanger announces the 2nd annual Conservation Tech Award

AI2's EarthRanger program is committed to helping protected areas and conservation organizations safeguard wildlife and their habitats through the advancement of technology. This award seeks to provide two $15,000 grants to organizations that are deploying or developing technology to make a positive impact in conservation management.

Want to learn more about AI2's work or connect with an expert? Reach us at press@allenai.org.

Featured AI2er:
Sam Skjonsberg

Principal Software Engineer Sam Skjonsberg helps make AI researchers' work tangible via specialized tools and demos — and he's also got an incredibly adorable son and puppy. Learn about Sam's journey to where he is today on the blog. 

Learn more ➞

Understanding AI's Carbon Footprint

Advancements in AI can accelerate overall technological progress. However, they can also take significant computing power, which can result in significant carbon emissions. In a new paper by AI2's Jesse Dodge and Microsoft's Will Buchanan, the pair propose a new way of measuring and mitigating these emissions.

Learn more ➞

More from us

➞ Is this robot sentient? Q&A with AI expert on debate over Google’s chatbot software
➞ Understanding AI, Part 2: Performance Measurement

➞ Allen Institute launches new benchmark for general-purpose computer vision models
➞ Paul Allen Institute Helps South Africa Thwart Locust Swarms
Visit us: allenai.org | Follow us: @allen_ai | Work with us!

AI2 Color Logo