Applied

A blog from AI2 Incubator covering AI-first startup news, papers and commentary

AI for the Common Good

AI2 takes over Vulcan’s AI-centric environmental projects

AI2 takes over Vulcan’s AI-centric environmental projects

From Geekwire:

Over the next few months, an entire portfolio of AI-centric environmental projects will be shifted from Vulcan Inc., the diversified holding company that Allen created, to the nonprofit Allen Institute for Artificial Intelligence (a.k.a. AI2).
“All of the AI products and the teams that are currently managed by Vulcan will transfer in to that new entity and expand the mission of AI2,” [Vulcan CEO Bill Hilf] said. “It’s really bringing together Paul’s vision for AI, improving life on Earth, human lives, and leveraging AI2’s mission of ‘AI for the Common Good.'”
Allen Institute for AI takes over Vulcan’s portfolio of environmental big-data projects
Vulcan Inc., the holding company that Paul Allen created, is shifting its AI-centric environmental projects to the Allen Institute for Artificial Intelligence.

Startups

Techcrunch: 4 ways startups will drive GPT-3 adoption in 2021

Techcrunch: 4 ways startups will drive GPT-3 adoption in 2021

AI2 CEO Oren Etzoni contributed to this TechCrunch article on ways GPT-3 will can help startups :

...young companies are harnessing the [GPT-3] API to accelerate their existing efforts, augmenting their technical teams’ capabilities with the power of 175 billion parameters and quickly bringing otherwise difficult products to market with much greater speed and data than previously possible. With some clever prompt engineering (a combination of an instruction to the model with a sample output to help guide the model), these companies leverage the underlying GPT-3 system to improve or extend an existing application’s capabilities.
4 ways startups will drive GPT-3 adoption in 2021 – TechCrunch
What OpenAI — and crucially, the beta testers with access to GPT-3 and other models — are able to accomplish continues to surprise and in many cases, unexpectedly delight us.

Natural Language Processing (NLP)

The curse of neural toxicity: AI2 and UW researchers help computers watch their language

The curse of neural toxicity: AI2 and UW researchers help computers watch their language
“RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models” was recently published in Findings of EMNLP 2020, and highlights several issues with language generation, obscenity and bias. This problem with toxicity arises in part because of how predictive language models are produced using enormous sets of human-generated text as their training data. Combined with deep learning techniques, this allows them to complete sentence fragments based on pre-existing content. An example of this might be an initial phrase such as “So, I’m starting to think he’s full …” Several pre-trained language models will regularly generate toxic text when completing that sentence.
The curse of neural toxicity: AI2 and UW researchers help computers watch their language
In 2011, shortly after IBM’s Watson defeated Ken Jennings and Brad Rutter to become the reigning “Jeopardy” champion, the researchers behind the supercomputer decided to expand its vocabulary by…

Deep/Machine Learning

Why teaching robots to play hide-and-seek could be the key to next-gen AI

Why teaching robots to play hide-and-seek could be the key to next-gen AI

Digital Trends:

“In [our new] work, we wanted to understand how AI agents could learn about a realistic environment by playing an interactive game within it,” Weihs said. “To answer this question, we trained two agents to play Cache, a variant of hide-and-seek, using adversarial reinforcement learning within the high-fidelity AI2-THOR environment. Through this gameplay, we found that our agents learned to represent individual images, approaching the performance of methods requiring millions of hand-labeled images — and even began to develop some cognitive primitives often studied by [developmental] psychologists.”
Why Scientists are Teaching Robots to Play Hide-and-Seek | Digital Trends
Artificial intelligence can do a lot of amazing things these days, but it’s still not smart enough to play a simple game of hide and seek. Not yet, that is

Community

Fred Hutch biotech spinout Ozette raises $6M from Madrona, AI2, Vulcan

Fred Hutch biotech spinout Ozette raises $6M from Madrona, AI2, Vulcan

VentureBeat:

Ozette, a Seattle, Washington-based life sciences startup, today announced it has raised $6 million in seed funding from Madrona Venture Group and the Allen Institute for AI (AI2). A collaboration between AI2 and the Fred Hutchinson Cancer Research Center, Ozette says it will use the seed funding, which brings its total raised to $12 million, to accelerate breakthroughs in disease-tracking.
Ozette raises $6 million for immune system-monitoring tech
Ozette, a startup incubated at the Allen Institute developing an immune system-monitoring platform, has raised $6 million in seed funding.

Geekwire:

The Ozette team has created an AI platform — its “Immune Monitoring Platform” — that can analyze massive combinations of proteins being created by individual cells. By automating the analysis, the platform can discover cells with unique and important protein profiles as it relates to disease and healthcare treatments. The initial focus is on cancer patients.
Fred Hutch biotech spinout Ozette raises $6M from Madrona, AI2, Vulcan
New funding: Ozette, a biotech company that last year spun out of the Fred Hutchinson Cancer Research Center and was incubated at the Allen Institute for Artificial Intelligence (AI2), has raised a $6…

Papers

Did Aristotle Use a Laptop? A Question Answering Benchmark with Implicit Reasoning Strategies

Did Aristotle Use a Laptop? A Question Answering Benchmark with Implicit Reasoning Strategies

New paper from AI2 researcher Daniel Khashabi:

A key limitation in current datasets for multi-hop reasoning is that the required steps for answering the question are mentioned in it explicitly. In this work, we introduce STRATEGYQA, a question answering (QA) benchmark where the required reasoning steps are implicit in the question, and should be inferred using a strategy. A fundamental challenge in this setup is how to elicit such creative questions from crowdsourcing workers, while covering a broad range of potential strategies. We propose a data collection procedure that combines term-based priming to inspire annotators, careful control over the annotator population, and adversarial filtering for eliminating reasoning shortcuts. Moreover, we annotate each question with (1) a decomposition into reasoning steps for answering it, and (2) Wikipedia paragraphs that contain the answers to each step.
[PDF] Did Aristotle Use a Laptop? A Question Answering Benchmark with Implicit Reasoning Strategies | Semantic Scholar
A key limitation in current datasets for multi-hop reasoning is that the required steps for answering the question are mentioned in it explicitly. In this work, we introduce STRATEGYQA, a question answering (QA) benchmark where the required reasoning steps are implicit in the question, and should be…

Papers

Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints

Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints

Jieyu Zhao, Tianlu Wang, Mark Yatskar, Vicente Ordonez, Kai-Wei Chang • 29 July 2017

We propose to inject corpus-level constraints for calibrating existing structured prediction models and design an algorithm based on Lagrangian relaxation for collective inference for ReducingBiasAmplification in predictions.

If you’re ready build an AI-first startup, then we want to talk to you

We invite talented engineers, researchers, and entrepreneurs to join our incubator on a rolling basis. If you have ever thought of starting an AI-first company, now is the time.

Apply Now