What should be done about AI safety
Last Tuesday, Anthropic researcher Jacob Coxon said he was quitting over fears that his AI company and its competitors were building self-improving tools that could destroy humanity. A team lead at Anthropic, Evan Hubinger, wrote on X that he thought there was a more than 10 percent chance AI could kill all humans in the next decade.
On Saturday, the leaders of four of the biggest AI companies — Dario Amodei of Anthropic, Sam Altman of OpenAI, Demis Hassabis of Google DeepMind, and Elon Musk of SpaceXAI — each agreed that they need to slow development of the technology, says The Wall Street Journal.
Here is Amodei’s essay on the need for each major AI company to place third-party evaluators in their firms to verify safety practices (he’s pledging that his company will do this, as has Altman) and coordinate with other companies on common safety standards and limits on unchecked AI progress. He also says the United States should cooperate with other governments.
A New York Times article on why it’s difficult for tech companies to rein in AI says: “Right now, because AI works so fast, researchers need AI to monitor it. But that doesn’t always work, because the AI monitors can appear to be more sympathetic to other AI systems than to the humans setting the rules.”
And a second and more difficult issue is “alignment,” an industry term that basically means making sure AI does what’s best for humans, says the Times.
On Sunday, President Trump seemed dismissive of the calls to slow AI development, says The Washington Post. “We’re leading China in AI, we’re the most sophisticated country in the world, and frankly I want to keep it that way,” he said.
After Trump signed an executive order last year rolling back Biden-era policies that took a more cautious approach to AI, his sons invested in the industry.
The top Republican and Democrat in the House agreed on Sunday that Congress should put guardrails in place, but the House is returning to Washington on Monday for what’s scheduled to be its final week in session before the November midterm elections, says the Times.
Here is Heather Cox Richardson on the AI issue.
And here is Robert Reich.
How Iran hard-liners sabotaged the push by more moderate pragmatists to reach a peace deal with U.S.
Trump touches a nerve while in Ireland, saying unified Republic of Ireland and UK’s Northern Ireland would be ‘fantastic’
Election officials are preparing for threats to midterm elections from Trump, his agencies and allies
AP: DOJ failed to get felony convictions of most of 102 protesters in 4 major cities charged with assaulting federal authorities
Federal judge rules Trump administration plan for 50 percent staffing cuts at FEMA was unlawful
Contractor says its own repairs, not vandals, caused Reflecting Pool peeling
Flaw in Georgia voting equipment could let someone see how others voted
Heather Cox Richardson on election, post-election threats
Tough summer for U.S. news industry draws to a close with cuts at McClatchy chain
At some universities, parents can hire a local mom to make sure their children are cared for the way they were at home
KFF: A generation of children is suffering as Trump destabilizes immigrant families
You can call the Capitol switchboard, (202) 224-3121, and be connected to the offices of your representative and senators. To email your House member and your two senators, you can connect to their websites at Congress.gov. Most lawmakers seem to only accept emails from their constituents, but these leaders accept emails from Americans nationwide, at:
House Democratic Leader Hakeem Jeffries
Senate Democratic Leader Chuck Schumer
Senate Majority Leader John Thune