A new report from Microsoft outlines the steps the company took to release responsible AI platforms last year. 

In its Responsible AI Transparency Report, which mainly covers 2023, Microsoft touts its achievements around safely deploying AI products. The annual AI transparency report is one of the commitments the company made after signing a voluntary agreement with the White House in July last year. Microsoft and other companies promised to establish responsible AI systems and commit to safety.

Microsoft says in the report that it created 30 responsible AI tools in the past year, grew its responsible AI team, and required teams making generative AI applications to measure and map risks throughout the development cycle. The company notes that it added Content Credentials to its image generation platforms, which puts a watermark on a photo, tagging it as made by an AI model. 

The company says it’s given Azure AI customers access to tools that detect problematic content like hate speech, sexual content, and self-harm, as well as tools to evaluate security risks. This includes new jailbreak detection methods, which were expanded in March this year to include indirect prompt injections where the malicious instructions are part of data ingested by the AI model.

It’s also expanding its red-teaming efforts, including both in-house red teams that deliberately try to bypass safety features in its AI models as well as red-teaming applications to allow third-party testing before releasing new models.

However, its red-teaming units have their work cut out for them. The company’s AI rollouts have not been immune to controversies.

When Bing AI first rolled out in February 2023, users found the chatbot confidently stating incorrect facts and, at one point, taught people ethnic slurs. In October, users of the Bing image generator found they could use the platform to generate photos of Mario (or other popular characters) flying a plane to the Twin Towers. Deepfaked nude images of celebrities like Taylor Swift made the rounds on X in January, which reportedly came from a group sharing images made with Microsoft Designer. Microsoft ended up closing the loophole that allowed for those pictures to be generated. At the time, Microsoft CEO Satya Nadella said the images were “alarming and terrible.”

Natasha Crampton, chief responsible AI officer at Microsoft, says in an email sent to The Verge that the company understands AI is still a work in progress and so is responsible AI. 

“Responsible AI has no finish line, so we’ll never consider our work under the Voluntary AI commitments done. But we have made strong progress since signing them and look forward to building on our momentum this year,” Crampton says. 

My breaking point with promotional emails and desktop alerts finally happened a few weeks ago. I woke up at 7AM to an automated Legacy.com email with my friend’s death-a-versary in the subject line. The email itself was annoying enough, but what it said made it a cold, thoughtless annoyance: “Being remembered matters. The flowers you sent last year were a comforting gesture of sympathy and support.” 

I didn’t send flowers. I planted a tree. That’s what my friend wanted. It was right there on the obituary guestbook Legacy.com was asking me to sign again.

Actually, the Legacy.com email was just the last straw. Things were kicked off a few months earlier by a Microsoft OneDrive notification. I had just switched from Google Drive, and instead of making a new email address, I used an ancient Hotmail account that’s been tied to my Xbox account for over a decade. If you had told me I had photos in that email’s cloud storage, I wouldn’t have believed you. I’d swear up and down I never used cloud storage under that email address. Yet, a day after I updated my subscription, an “On this day” memories alert popped up.

I clicked on it — and, oh my, was that a mistake. Microsoft OneDrive wanted me to remember one of the darkest times of my life by shoving photos of an abusive ex in my face — photos I had forgotten existed. In a rush of anger, I erased every single one of those photos from digital existence and canceled my OneDrive subscription. There are some things you don’t need to be reminded of because you’ll never forget them. 

Intrusive but necessary

Notifications can intrude into every moment of our lives, fighting for our attention without tact. Yes, we can turn them off or click “unsubscribe” on emails that never made it into the junk folder, but the point is they shouldn’t be happening in the first place. Would we be okay with a stranger holding up a sign that read, “Hey! Remember when your friend died?” as they ran up to us in a cemetery? Would we accept our ex-partner shouting, “It was supposed to be me!” in the middle of our wedding ceremony? Enough intrusive thoughts go through my head on a daily basis. I don’t need an algorithm reinforcing them because it mathematically concluded I want to see what it wants to show me. 

On the other hand, like all technology, notifications are tools. Receiving too many can distract and overwhelm us, but we might forget something important if we receive too few. And while we have some ability to adjust which notifications we get, the companies that create those apps don’t have much incentive to hand over control because they want us to use their products as much as possible. (Seriously, Duolingo, chill out. You don’t need to cry over my missed Klingon lesson.) It’s further complicated by having to figure out what buttons to tap in your settings to find the best middle ground between which notifications you want versus which deserve the silent treatment. 

I’ve now reined in my smartphone notifications (largely by buying a “dumb phone”), but emails and cloud storage alerts have stuck around like an endless game of Wacky Gator — even though I don’t remember signing up for most of them. And the moment they pop up, it’s easier for me to dismiss them than to figure out how to turn them off for good. I always intended to dig into my email and cloud drive settings, but a week became a month, then a month became a year — and now, I have 414 active email subscriptions and a cloud drive I never log in to because I dread its pop-ups.

But there’s something more nefarious about using your own photos and memories — even the good ones — to capture your attention. It’s convenient to store them in the cloud, even if you disable all “On this day” notifications. But that cloud is a server owned by a tech company that can lock you out of your memories if you cancel your subscription. What then?

An extreme solution by today’s standards is to store everything on an external drive where no one has access to them but you. You lose the convenience of accessing them from any device anywhere at any time, but you gain something much better in return: privacy. So, that’s where I will store all my photos from now on. I’ve avoided getting a NAS because it seems like too much work, but it’d be nice to still be able to access my stuff from anywhere. I’m done dealing with emotionally unaware algorithms and automated emails feigning sympathy to get me to engage with websites. My memories are not marketing tools.

A new bill seeking to track security issues by mandating the creation of a database recording all breaches of AI systems has been filed in the Senate. 

The Secure Artificial Intelligence Act, introduced by Sens. Mark Warner (D-VA) and Thom Tillis (R-NC), would establish an Artificial Intelligence Security Center at the National Security Agency. This center would lead research into what the bill calls “counter-AI,” or techniques to learn how to manipulate AI systems. This center would also develop guidance for preventing counter-AI measures. 

The bill will also require the National Institute of Standards and Technology (NIST) and the Cybersecurity and Infrastructure Security Agency to create a database of AI breaches, including “near-misses.” 

Warner and Tillis’ proposed bill focuses on techniques to counter AI and classifies them as data poisoning, evasion attacks, privacy-based attacks, and abuse attacks. Data poisoning refers to a method where code is inserted into data scraped by an AI model, corrupting the model’s output. It emerged as a popular method to prevent AI image generators from copying art on the internet. Evasion attacks change data studied by AI models to the point the model gets confused. 

AI safety was one of the key items in the Biden administration’s AI executive order, which directed NIST to establish “red-teaming” guidelines and required AI developers to submit safety reports. Red teaming is when developers intentionally attempt to get AI models to respond to prompts they’re not supposed to. 

Ideally, developers of powerful AI models test the platforms for safety and have them undergo extensive red teaming before being released to the public. Some companies, like Microsoft, have created tools to help make adding safety guardrails to AI projects easier. 

The Secure Artificial Intelligence Act will have to go through a committee before it can be taken up by the larger Senate.

LinkedIn is now in the gaming business. Starting today, users on the LinkedIn mobile app or on desktop can play one of three different games — Pinpoint, Queens, and Crossclimb. You’ll be able to play each game once per day, and after your daily session, you’ll get access to all kinds of metrics including your high score and daily streak, different leaderboards, and who in your networks has also played. The games are available here under the LinkedIn News and My Network section on desktop or the My Network tab on mobile.

Here’s a brief rundown of the three games.

Pinpoint is a word association game. The game will unveil five different words, and your job is to guess the category the words fit into. The words will reveal themselves on a timer with the objective being to guess the category in as few words as possible.

Crossclimb combines trivia with clever wordplay. You’ll be given a clue for a word, and with that word as a starting point, you’ll create a ladder of words with each subsequent entry being just one letter off from the one before. Arranging the words in the correct order will reveal the clue to guess the locked entries on the ladder to win the game. It’s probably better to see it in action:

Finally, Queens is the most interesting game as it’s merely sudoku without numbers. Place queens on a grid such that no queens touch each other and there is a single queen in each row and column.

LinkedIn’s decision to get into puzzle games shouldn’t be surprising. Digital content businesses are struggling to make money with ad revenue shrinking and Google doing its level best to ensure you never click on a valuable link again. Adding a slate of “gaming” content, then, has proven hugely valuable. It offers a unique way for businesses to capture new users and engage older ones before eventually getting both to spend money they otherwise wouldn’t.

According to Axios, New York Times games were played over 8 billion times last year with more than half of those plays belonging to Wordle, the guess-the-word game the Times acquired in 2022. The Times offers subscription packages for its games by themselves or in a more expensive All Access package that bundles games with other New York Times content. In an interview with Digiday, the publisher’s head of games, Jonathan Knight, explained how gaming subscriptions have helped the Times grow and keep users. “If you’re a subscriber, and on any given week, you engage with both news and games, the likelihood that you’re going to retain [your subscription] over a long period of time is much higher.” 

Other publications are following similar paths. Late last year, Hearst, which publishes magazines and newspapers across the country, acquired Puzzmo, a puzzle game platform that includes games like SpellTower and Really Bad Chess. And the trend isn’t limited to digital news organizations as Netflix’s gaming endeavor continues to grow, adding exclusive mobile versions of popular games like Hades and Sonic Mania Plus.

LinkedIn isn’t charging for its games yet. Rather, they seem to be a way to keep users engaged on the platform. In addition to displaying a person’s high score and daily streak, LinkedIn will show who in a person’s connections has also played as well as school and company leaderboards. Lakshman Somasundaram, LinkedIn’s product director, said in the press release, “It’s time we turn over a new leaf in how we deepen and reignite relationships at work, and put fun at the heart of it.”