Open-vocabulary models like CLIP achieve high accuracy across many image classification tasks. However, there are still settings where their zero-shot performance is far from optimal. We study model patching, where the goal is to improve accuracy on specific tasks without degrading accuracy on tasks where performance is already adequate. Towards this goal, we introduce PAINT, a patching method that uses interpolations between the weights of a model before fine-tuning and the weights after fine-tuning on a task to be patched. On nine tasks where zero-shot CLIP performs poorly, PAINT increases accuracy by 15 to 60 percentage points while preserving accuracy on ImageNet within one percentage point of the zero-shot model. PAINT also allows a single model to be patched on multiple tasks and improves with model scale. Furthermore, we identify cases of broad transfer, where patching on one task increases accuracy on other tasks even when the tasks have disjoint classes. Finally, we investigate applications beyond common benchmarks such as counting or reducing the impact of typographic attacks on CLIP. Our findings demonstrate that it is possible to expand the set of tasks on which open-vocabulary models achieve high accuracy without re-training them from scratch.

Questions regarding implicitness, ambiguity and underspecification are crucial for understanding the task validity and ethical concerns of multimodal image+text systems, yet have received little attention to date. This position paper maps out a conceptual framework to address this gap, focusing on systems which generate images depicting scenes from scene descriptions. In doing so, we account for how texts and images convey meaning differently. We outline a set of core challenges concerning textual and visual ambiguity, as well as risks that may be amplified by ambiguous and underspecified elements. We propose and discuss strategies for addressing these challenges, including generating visually ambiguous images, and generating a set of diverse images.

IN BRIEF America's Pacific Northwest National Laboratory is looking into how AI technologies can be used to create a "Digital Police Officer" or "D-PO" in the future. Freedom-of-information requests filed by the Electronic Frontier Foundation show the US Department of Energy-funded lab envisions cops may one day be able to partner up with a virtual crime-fighting assistant. D-PO would be capable of, for instance, tapping into facial recognition systems to alert a police officer on patrol to a suspect nearby, and can even offer advice on how best to apprehend the suspect. The EFF warned against the plod teaming up with software like D-PO, citing concerns over inaccurate facial recognition matches and biased predictive policing policies. "The good news is that in the emails we obtained, one of the authors acknowledges in internal emails that elements like a D-PO taking over driving is a'long way off' and monitoring live drone feeds is'not a near-term capability,' the digital privacy-focused non-profit said.

Role requiring'No experience data provided' months of experience in Chantilly Pay if you succeed in getting hired and start work at a high-paying job first. Get Paid to Read Emails, Play Games, Search the Web, $5 Signup Bonus. AM LLC is seeking a SQL, DBA to support the Veterans Administration (VA) customer's enterprise Digital GI Bill instance. Candidates must reside within daily commuting distance to Washington DC or Chantilly, VA. The role will start as remote but has the possibility of changing to working onsite.

White House press secretary Karine Jean-Pierre told reporters during an audio-only gaggle Friday that the U.S. has no indication that Russia plans to use nuclear weapons, after President Biden warned of "Armageddon." The war in Ukraine has revealed how the digital age is leveling the playing field between great powers and smaller countries. Ukraine has skillfully deployed precision munitions, drone technology and sophisticated encrypted software to gain the upper hand against Russia's invading conventional military, but Russian President Vladimir Putin's most recent remarks, and his move to illegally annex portions of Ukraine, make it clear that digital warfare will also unleash a second nuclear age. Western technology, including encrypted command and control, the High Mobility Artillery Rocket System (HIMARS), drone and counter-drone systems, combined with Ukrainian savvy and resolve have arrested Russian advances and recently rolled back Russian gains. Chips and software have proven more potent than tanks and soldiers.

That was the response from AI policy and ethics wonks to news last week that the Office of Science and Technology Policy, the White House's science and technology advisory agency, had unveiled an AI Bill of Rights. The document is Biden's vision of how the US government, technology companies, and citizens should work together to hold the AI sector accountable. The US has so far been one of the only Western nations without clear guidance on how to protect its citizens against AI harms. Tech companies say they want to mitigate these sorts of harms, but it's really hard to hold them to account. The AI Bill of Rights outlines five protections Americans should have in the AI age, including data privacy, the right to be protected from unsafe systems, and assurances that algorithms shouldn't be discriminatory and that there will always be a human alternative.

Caitriona Fitzgerald is EPIC's deputy director and Ben Winters is EPIC counsel. The White House Office of Science and Technology Policy last week released a "Blueprint" for an "AI Bill of Rights." While the principles set out in the blueprint do not have the force of law, there are several actions the White House can take to put them into practice within the federal government while simultaneously pushing for new legal protections. The Biden Administration should lead by example. The major principles set out in the AI Bill of Rights are that AI systems must be safe, be effective, be free of discrimination, respect data privacy, make their use known, and have an extensive structure of human oversight.

A task force focused on soldier lethality is adding new initiatives to its portfolio, including a lighter Javelin missile, identifying how artificial intelligence can help squads, and looking into robot dogs as infantry battle buddies. The Close Combat Lethality Task Force, established in 2018 under then-Secretary of Defense Jim Mattis, pushed for the Next Generation Squad Weapon, a 6.8mm intermediate-caliber replacement for the M4 for close combat troops, which was selected this year and begins fielding to troops in 2023. It also lobbied for additional funding and prioritization for the Integrated Visual Augmentation System, a $22 billion program for a mixed-reality goggle aimed to put fighter pilot situational awareness tools in the view of squad-level soldiers. On the human side, the task force supported efforts to revitalize infantry and close-combat training, increase unit cohesion by keeping infantry troops in the field longer, and reduce training tasks not related to combat. But the task force has mostly remained in the shadows and sought a home since Mattis left office in 2019.

As chief data officer of payroll and benefits management company ADP, Jack Berkowitz has three primary responsibilities. One is to oversee the organization's data overall, ensuring that functions like data governance, security, and analytics, are running well. Another is to build ADP's data products, such as people analytics and benchmark tools. But the responsibility that's of most interest to Me, Myself, and AI hosts Sam Ransbotham and Shervin Khodabandeh is Jack's oversight of the organization's use of artificial intelligence. In this episode of the podcast, Jack describes how focusing on the outcomes the organization wants to achieve leads to better processes and results. He also dives into the topic of AI ethics and outlines how other organizations might consider assembling an AI ethics board. Jack Berkowitz is chief data officer at ADP, where he leads the company's data security and governance, data platforms, and analytics/machine learning operations. His role also involves partnering with stakeholders to develop new data initiatives to improve clients' experience and ADP's competitive position. Berkowitz joined ADP in 2018 as the senior vice president of product development for the DataCloud people analytics and compensation benchmarking solution.

Airports and industries should be required to publicly disclose their use of facial recognition, while the National Security and Intelligence Committee of Parliamentarians should review any military or intelligence use of the technology, they said. Tamir Israel, a lawyer with the Samuelson-Glushko Canadian Internet Policy and Public Interest Clinic, testified at the committee that travelers might not be aware they're subject to the technology, such as at the customs screening mechanism at the Pearson Airport in Toronto. The government should disclose its own acquisitions of the technology, and "create a public AI registry in which all algorithmic tools used by any entity operating in Canada are listed," MPs said. Privacy lawyer Carole Piovesan told the committee that while discussions on FRT "tend to focus on security and surveillance," the technology is also used by other sectors, including health care, retail and e-commerce, and telecom and IT. The technology is also more accurate in identifying white individuals, and less accurate in identifying people of colour.