DataDownload: Machine trust 101… Welcome to the world of Explainable AI (XAI)

7 min readMar 7, 2020

DataDownload: Machine trust 101… Welcome to the world of Explainable AI (XAI) A weekly summary of all things Media, Data, Emerging Tech View this email in your browser

Robots are learning how to explain themselves. The Church is looking to have limits set to protect people from unethical AI. Google is working to remove gender bias from its AI. And AI is tracking COVID-19.

In the connected world, the outbreak is making clear just how much our networks, supply chains, and systems are critically tied around the globe. Can technology help to solve problems as we look to build new solutions to emerging global issues? Let’s hope so. I’m certainly disappointed to be missing all of you at SXSW, but hopeful that we’ll find a vaccine that slows the spread and soon enough we’ll be returning to festivals and conferences in the RW. Until then, we go virtual.

As always, enjoy hearing from you, so don’t be shy. Steve@NYCMediaLab.org.

Cheers,

Steven Rosenbaum
Managing Director
The NYC Media Lab Must-Read How Robots Explain Themselves Matters More Than You Might Think

Explainability in AI systems is a crucial area of research for Darpa. On their Explainable AI (XAI) project page, they note that explainable machine learning “will be essential if future warfighters are to understand, appropriately trust, and effectively manage an emerging generation of [AI] partners.” (They also mention commercial applications).

Darpa’s XAI program recently funded work at UCLA’s Center for Vision, Cognition, Learning, and Autonomy where researchers are exploring the factors that make machines more trustworthy — you can read the full paper here. In this Fast Company guest post, two of the paper’s authors go over the paper’s experiment: their Baxter robot was equipped with a tool to explain the machine’s planning system in action, rather than being added after-the-fact.

The team tested the tool’s trustworthiness on 150 participants, finding that “the symbolic explanation… made people trust the robot most” (the middle row in the image shows the symbolic component).

5 min read

Read More The Catholic Church Proposes AI Regulations That ‘Protect People’ It’s an interesting time when the Catholic Church is on the same spectrum (not necessarily wavelength) as the Pentagon regarding AI ethics. The Pentagon recently deployed new ethical guidelines for AI systems in combat and non-combat applications, advising “appropriate levels of judgment and care,” and that systems should be traceable and governable.

Meanwhile, Vatican officials issued a Rome Call for AI Ethics, which lays out six principals around safer AI creation: “transparency, inclusion, responsibility, impartiality, reliability, and security and privacy.” IBM and Microsoft were the first c0mpanies to sign.

2 min read Read More For the Media Salesforce’s AI Navigates Wikipedia to Find Answers to Complex Questions

To better scour Wikipedia’s 6B pages for answers to complexly-worded questions, Salesforce researchers built a state-of-the-art framework that sequentially retrieves paragraphs from the site to extract plausible answers. Natural language processing is a crucial component of Salesforce’s Einstein AI prediction platform (Einstein Language specifically analyzes unstructured text).

The company also released CTRL last year — the largest open-source language model to date, surpassing even OpenAI’s enormous GPT-2 model in size.

2 min read

Read More Google AI Tool Will No Longer Use Gendered Labels Like ‘Woman’ or ‘Man’ in Photos of People Google recently emailed developers that their Cloud Vision API will not identify people as “man” or “woman” in images anymore, but as a “person”, as part of an effort to mitigate bias in its algorithms: “Given that a person’s gender cannot be inferred by appearance, we have decided to remove these labels in order to align with the Artificial Intelligence Principles at Google, specifically Principle #2: Avoid creating or reinforcing unfair bias.”

1 min read Read More Alphabet’s X Launches Tidal to Automatically Track and Log Fish Behavior

Alphabet’s moonshot factory X has launched Tidal, which will start off providing fish farmers with tools to “run and grow their operations.” Tidal project lead Neil Davé noted in his blog post that “humanity is pushing the ocean past its breaking point, but we can’t protect what we don’t understand.” Davé said the project’s initial focus will be developing tech to “bring greater visibility and understanding of what’s happening under the water.”

Davé’s team has been consulting with fish farmers for the past three years. The current method of assessing the health of thousands of fish is taking a small sample and observing them manually — something “unreliable and impossible to scale.” X developed an underwater camera system and machine perception tools that “log fish behaviors like eating, and collect environmental information like temperature and oxygen levels.”

3 min read

Bloomberg features Boston Children’s Hospital chief innovation officer John Brownstein, who discusses the hospital’s new coronavirus tracking tool, which leverages AI to notice small signals that are difficult for governments to detect using traditional methods.

6 min watch

Watch Now Jobs & Events Event: Data Science Day 2020
Date: March 31, 9AM-5PM
The Data Science Institute’s flagship annual event will host presentations from leading voices in data-driven innovation, lightning talks from Columbia University faculty, and a pavilion of interactive technology demonstrations and research posters. Eric Schmidt, Technical Advisor to the Board, Alphabet Inc., and Pat Bajari, Chief Economist and Vice President of Core AI, Amazon, will deliver this year’s keynote addresses. Register Here.

Event: Climate Data in a Warming World
Date: March 10, 8:30AM-10AM
Join the Building Energy Exchange and the ASHRAE New York Chapter for their second installment of the joint ASHRAE Standard technical series — providing a deep dive for their development, industry impact, and relationship to local energy efficiency and climate action objectives. Register Here.

Event: The Future of Local News
Date: March 10, 6:30PM-8PM
For more than a decade, too many local news outlets have been downsized or shuttered leaving a real news vacuum. In response to challenging times, new publications have emerged and existing news operations have been transformed, all finding creative ways to connect with audiences. Hear from New York’s savviest news innovators about where the opportunities are in journalism’s brave new world. Register Here. A Deeper Look Reinforcement-Learning AIs Are Vulnerable to a New Kind of Attack

Supervised learning has notoriously been prone to disruption from adversarial attacks where bananas become toasters and turtles appear to be guns. Adversarial attacks on reinforcement learning systems have been studied for several years, but UC Berkeley researcher Adam Gleave and his colleagues have found a previously unrecognized threat model: changing the environment around the AI, aka adversarial policy. Gleave and Co. trained bots to play two-player games to demonstrate adversarial policy in action:

“The adversaries learned to win not by becoming better players but by performing actions that broke their opponents’ policies. In the soccer game and the running game, the adversary sometimes never even stands up. This makes the victim collapse into a contorted heap or wriggle around in circles. What’s more, the victims actually performed far better when they were ‘masked’ and unable to see their adversary at all.”

4 min read

While Google, IBM, Rigetti, and others have settled on transmon-based quantum systems, Honeywell has gone a different route for their new quantum computer, using a technology called “ion trap.” Honeywell’s president of Quantum Solutions, Tony Uttley, believes the system can “scale rapidly,” with additional qubits added without fundamental changes to architecture.

Based on a benchmark defined by IBM called quantum volume, Honeywell is calling itself “the world’s most powerful quantum computer.” Ars Technica examines the technology and the claim in-depth.