Tim Bernard

If you're still not convinced that all AI should be labeled "Handle With Care" then you might want to read the recently released New York state bar guidelines on AI use (there's a link in this Reuters story). Basically, it's a 90-page introduction to AI and its implications for a particular profession (law). But also a good starting point for any organization thinking of using AI, given that it explains a wide range of factors to consider before/after doing so. I will also put a couple of other relevant links in the comments.

2

4 Comments

Big New Report: Today, my colleagues and I at Stanford Internet Observatory, led by Shelby Grossman, are publishing a comprehensive report about the CyberTipline, the pipeline by which platforms report child sex abuse material (CSAM) they find on their services. The report analyzes what works and what doesn’t, and makes recommendations - some technical, some political - for improving a 26-year-old system that has become indispensable in the fight against online child sexual exploitation. We based the report on dozens of interviews we’ve conducted since last summer with stakeholders from across the reporting ecosystem: platforms, law enforcement, prosecutors, criminal defense attorneys, child safety groups, and NCMEC, the organization that operates the CyberTipline. Much has been written about the CyberTipline, but we think our report is unprecedented. In January, NCMEC hosted us for a 3-day site visit at their HQ that gave us a level of access to their work that we’re pretty sure has never before been granted to outside researchers. We are grateful to NCMEC, as well as the 60-plus other individuals who generously gave us their time for interviews over the last 9 months. This report would not have been possible without all of them. Read the report: https://lnkd.in/gzpnQaGm Join us for a launch webinar today at 9am PT: https://lnkd.in/gKkUrvYA

75

1 Comment

Julia Angwin’s piece out today in the NYT sharply lays out the backside of the hype cycle for AI—escalating costs for marginal improvements, a hunger for more data that poses corporate risks and explains OpenAI’s appification as we all get recruited to train the models (and privacy and data ownership are obliterated as a sunk cost), unfixably biased outcomes and accuracy issues, and the climbing environmental price tag. https://lnkd.in/eXU8J-mV

9

Day 2 of "Pride Under Attack", a collaborative investigation done by DDIA, Univision, and Agência Lupa, in Spanish, English, and Portuguese. If you understand how important sex education is, please share this post. #LGBT #disinformation‍

3

Interesting essay on how to distribute the benefits from AI –– "Predistribution over Redistribution" -- by Saffron Huang and Sam Manning. It's published on The Collective Intelligence Project's blog: https://lnkd.in/eKn_tRzm Advanced AI systems could concentrate power and economic value in the hands of a small group of AI companies. To address this concentration of wealth and power, you might want to take a primarily redistributive approach. This essay argues that *re*distribution should be complemented by *pre*distribution.

22

2 Comments

A nice write up about my new case study on Federal AI use rules and how they are suppressing the ability of agencies to manage AI risks/benefits in their contexts. Specifically, I discuss the federal governments deep translation capacity problems which are yielding life-impacting service , intelligence , and emergency response failures. Machine translation comes with error but could also fill gaps when the alternative is no translation at all. Unfortunately, one-size-fits all federal AI minimum "risk management" standards inhibit the ability to weigh risks and benefits in the context of a program/agency. The result: any benefits are lost and the many, many negatives of exsisting capacity failures locked in. https://lnkd.in/e-_s5H3q

6

1 Comment

Delighted to see this important blog post from Public Knowledge Policy Analyst for content moderation and information systems, Morgan Wilsmann!

8

My thoughts on the Senate AI Working Group's new AI Framework: “We have to wait and see what bills and roles are proposed based off of a lot of this but a lot of it is indeed promising. There's a strong emphasis on R&D, there's a strong emphasis on regulatory clarity, for example, and making sure that our existing laws are clear so that we can ensure easy, confident adoption of these technologies. There was clearly a thoughtfulness that went into this,” Mittelsteadt said. I'd estimate around 75% of the package is headed in the right direction: emphasizing engineering-based solutions, clearing up laws, and making targeted spending bets on talent and other discrete priorities. That said, elements such as AI export controls solutions in search of a problem and a recipe for policy failure. Too much is on the line to overzealously regulate. While action is needed on many fronts, with regard to export controls: “They've been there a broken record since at least 2016 about social media regulation, if not earlier. And when you cry wolf for going on 10 years about technology regulation, people stopped believing you,” said Matt Mittelsteadt, a research fellow at George Mason University’s Mercatus Center. “That's exactly where Congress is, and I think the most likely thing that we can expect is nothing."

4

I've finished my initial analysis of Google's Site Reputation Abuse Penalty/Update (seems more penalty than update right now). Major, targeted hits to reputable sites hosting not-so-reputable third-party content, especially coupon content (but also gambling content, payday loans, etc.). Impacted sites were hit hard, with coupon subdirectories and subfolders on major brands taking total visibility losses in many cases. This is definitely a punitive action, and not a core-style update. Not seeing a lot of reversal yet, but I am hearing rumors. Seeing a couple of odd cases where sites were deindexed recently and now are ranking again, so Google must be reprocessing. https://lnkd.in/dbcFJaKR

24

3 Comments

I'm giving a lightning talk at the Trust and Safety Research Conference today around 2:45pm PT! If you're at Stanford for the event I'd love to find time to chat. My talk is on my team's work last year to break down the Parliament's draft of the AI Act and analyze how companies' policies and practices aligned with its transparency-related provisions. Since publishing this work in June 2023, our thinking has evolved significantly (as has the AI Act), so I'm excited for today's talk as an opportunity to reflect. To assess this, we took a three step approach at the time: 1️⃣ We looked at 12 provisions in the draft proposal that specifically apply to foundation models and develop rubrics 2️⃣ We graded ten major providers and their flagship foundation models on each provision in the draft 3️⃣ We compared results across providers to gauge trends across the industry and potential gaps in the proposal We found that every model provider had room for improvement, but that it was feasible to adhere to the draft proposal as evidenced by some providers doing well on each provision. Now that a different version has entered into force, my takeaway remains that companies can learn from each other to align better with the AI Act's transparency requirements. Thanks to Rishi Bommasani, Daniel Zhang, and Percy Liang for co-leading this project, and to Stanford Institute for Human-Centered Artificial Intelligence (HAI) for its support!

125

7 Comments

Two years ago, I noticed that kids were primarily connecting through platforms like Discord, Google Hangouts, and Messenger. Unfortunately, none of these platforms encouraged in-person interaction, and their open nature posed serious safety risks. Strangers could easily reach out, and bots engaged children in endless conversations. Many kids, lacking the necessary self-control, ended up spending over 9 hours a day on these digital platforms. Despite continued mis-steps by big tech companies like Meta on platforms like Instagram, parents often forgave them because there seemed to be no better alternative. That’s when I decided to go back to the drawing board. I asked myself: How can we create something that helps kids feel connected, encourages outdoor play, and doesn’t become addictive? A product isn’t addictive when it’s need-based, not designed to trigger dopamine-fueled behaviors. With that vision, I built Play Connect, a platform specifically for middle schoolers, whether they have a phone or not. It’s designed to do just one thing: help kids coordinate play outdoors. There are strict safeguards in place—no adults are allowed, and it’s not open to the world. It is non addictive and there is no media. With Vivek Murthy highlighting the issues of social isolation and the harms of social media, Play Connect is the platform you want to bring to your kid community and the only app for connection you should have. Kids want social connection not social media. We should end trying to bring the same addictive technologies adults use for kids. Thankyou Nir Eyal for inspiration and Jonathan Haidt for bringing this voice to the limelight and Kamal Hassan for believing in us and all those parents who have voiced their frustrations *Also if you are a marketeer or a developer and really want to make the business of connection right and end this social isolation come work for us* if you want to bring Play Connect it to your school /community head to https://lnkd.in/ghc5TSZn If you believe in making this change and can share this post!! #MentalHealthMatters #SocialMedia #MentalHealth #socialmedia #parenting #screentime #mentalhealth

61

9 Comments

Wearing my personal hat, just published this in Tech Policy Press: AI will soon be everywhere. In the wake of announcements last week by Google about AI integration at the top of search, and the widely publicized product demo of ChatGPT4.o by OpenAI (and that company's announcement of a pending deal with Apple), we can see how competitive pressures are pushing companies to launch AI products aggressively towards consumers. Sure, it may be convenient, but it is also true that these AI companies need our data to improve their models even as they face diminishing returns. The full implications of this pivot from the subscription model to product and internet integration is being lost in some of the coverage of the public relations debacle that was the OpenAI demo—which, because it was aided by special NVIDIA chips, made it just a demo in the truest sense. Some of the decision to release 4.o, which focused on upgrades to consumer-facing capabilities, appears intended to distract us from ongoing technical challenges with accuracy, bias and transparency—flaws that will require actual fixes for AI decision-making to matter, but are difficult to solve given current designs. Combined with the public departures of almost the entire “safety” team at OpenAI over the past few months, culminating in the resignations of Ilya Sutskever and Peter Leike last week, something appears rotten in the heart of Silicon Valley. While the product demo for OpenAI in the fall of 2023 was led by CEO Sam Altman, as a display of normalcy following his ouster and reinstatement, the cringey demo this past week featured CTO Mira Murati, whose presence was clearly intended to give sanction to the otherwise sexist circus show… https://lnkd.in/e42h_Dck

11

3 Comments

Last week David Inserra, Emma Hopp, and I filed comments in the FCC proceeding on AI in Election Ads discussing the potential speech and administrative law issues in this approach you can read them here: https://lnkd.in/di8xXva3 #AI #TechPolicy #OnlineSpeech

26

1 Comment

In this blog, I delve into AI as a public good. Can it be regulated better to serve the democratic institutions, keep the welfare of people at the center, take into account the necessity to protect democratic and civil institutions, and check on the power imbalance that ensues with unhindered and unregulated use of technology by the masses? Do read- https://lnkd.in/gyTberh3  #AI #AIinDemocracy #IndianElections2024 #Elections #ArtificialIntelligence #MachineLearning #Democracy #humanlanguage #Misinformation #Disinformation

3

🤔 Predatory or revolutionary to use establish framework for a veryyyy different purpose? 🤔 (+ a Milennial’s Tumblr shoutout) 👇🏻 ➡ Tl;dr - User-generated (& often inaccurate) content is being surfaced in the format of news alerts/ notifications to the >1BN global users of TikTok... ➡ Come on, TikTok. Barely an oversight, this is just lazy development - tag the user-generated content with a flag? Filter it out of notifications? Present the notis in a different format? ➡ A success story here is my dear darling beloved Tumblr. They notify me of user-generation content in this format, but with UX copy + presentation format that's clear & different to the "news alert" sentiment.

1

Interesting to see how widespread the boycott of brands over the Gaza conflict has become. At Anavizio Data Solutions & Technologies we’ve been tracking the issue on social media since October and - while the effectiveness of the boycott may be debated - it shows no signs of letting up anytime soon. That is because public anger over what is happening in the Palestinian Territories is very real and, in the face of an otherwise helpless situation, boycotting is one of the only things people can do. If nothing else it goes to show just how important social media has become in shaping brand perceptions and impacting purchase decisions!

12

Sharing the latest #NIST publication on guidance for federal agencies and organizations (large and small) to develop and manage cybersecurity training programs. You will find information on broadly accepted standards, regulations, legislation, assessments metrics and best practices. #humancenteredcybersecurity

3

Excellent piece by Electronic Frontier Foundation (EFF) about Surveillance Pricing and the latest (and most necessary) investigations by the Federal Trade Commission (FTC) into this dangerous practice. What is surveillance Pricing? Surveillance pricing analyzes massive troves of your personal information (location, demographics, credit history, and browsing or shopping history) to predict the price you would be willing to pay for an item—and charge you accordingly. Retailers can charge a higher price when they think you can afford to spend more, or when you need something the most. Why is this practice possible? Due to the lack of protection of the right to privacy rights in the US companies can often surveil us with impunity. Unregulated data brokerages buy and sell enormous amounts of information, while it is quite difficult for people to protect themselves from these practices. What are the consequences? Surveillance models establish what are often inaccurate personal profiles. The algorithms and data behind those models are frequently opaque and lack governance regulations that allow to appeal unfair practices. Worse, surveillance pricing will often have disparate impacts on people of color and those living in poverty, worsening historical social and economic models of disparity. So, is there any good news? Yes! The FTC is investigating surveillance pricing practices. It recently issued orders to eight companies to provide information about products and services that use personal data, including finances and browser history, to set individualized prices for the same goods or services. The orders are seeking information on four major areas: (i) Types of products and services being offered; (ii) Data collection and inputs; (iii) Customer and sales information; (iv) Impacts on consumers and prices. This investigation is key to better understand how surveillance pricing is affecting consumers and market practices. Find the full article here 👉 https://shorturl.at/RFx0L #surveillance #privacy #US #competition #digitalrights

7