| | | It’s going to take a multinational effort for the global AI community—such as it is—to avoid the emergence of a “fragmented AI landscape.” The danger of such an unwanted scenario is having AI developers and end-users navigate jagged safety guardrails riddled with gaps. Stepping up to face down this threat, the nonprofit Partnership on AI is doing what it can to help coordinate the big coordinators. That’s apparent in a report released this month. In 40 or so pages, the San Francisco-based group compares, contrasts and otherwise analyzes eight policy frameworks. Two of these hail from the U.S.—President Joe Biden’s 2023 executive order “Safe, Secure and Trustworthy Development and Use of Artificial Intelligence” and the AI Risk Management Framework from the National Institute of Standards and Technology (NIST). The report lays out nine recommendations for formulating “a more coherent, effective approach to managing the risks and harnessing the potential of foundation models” so as to ensure accountability and transparency while fostering innovation in the global AI ecosystem. Here are all nine. 1. When identifying foundation models in need of additional governance measures, national governments and the EU should prioritize cooperation. ‘Agreeing on a common definition and thresholds for the models covered by policy frameworks should flow through to greater alignment between the frameworks.’
2. The G7 presidency should continue developing the Hiroshima Code of Conduct into a more detailed framework. ‘This work should seek input from foundation model providers, civil society, academia and other stakeholder groups equally.’
3. When creating and approving initial Codes of Practice for the EU AI Act, all involved parties should prioritize compatibility with other major AI governance frameworks. ‘The involvement of non-EU model providers, experts and civil society organizations will help advance this objective.’
4. To support the development of standardized documentation artifacts, standards development organizations (SDOs) should ensure that their processes are informed by socio-technical expertise and diverse perspectives as well as required resources. ‘To that end, SDOs, industry, governments and other bodies should invest in capacity building for civil society and academic stakeholders to engage in standards-making processes, including to ensure participation from the Global South.’
5. The development of standardized documentation artifacts for foundation models should be a priority in AI standardization efforts. ‘This would promote internationally comparable documentation requirements for foundation models, [encouraging] interoperability and establishing a baseline for best practice internationally.’
6. International collaboration and research initiatives should prioritize efforts that will support the development of standards for foundation model documentation artifacts. ‘Documentation is a key feature of foundation model policy requirements, and common requirements for artifacts will directly improve interoperability. It will also make comparisons between models from different countries easier, promoting accountability and innovation.’
7. National governments should continue to prioritize both international dialogue and collaboration on the science of AI safety. ‘This work will inform a common understanding of what should be included in documentation artifacts to promote accountability and address foundation model risks.’
8. National governments should support the creation/development of AI Safety Institutes (or equivalent bodies), and ensure they have the resources, functions, and powers necessary to fulfill their core tasks. ‘Efforts should be made to align the functions of these bodies with those common among existing AISIs. This will promote efforts to develop trusted mechanisms to evaluate advanced foundation models—and may, at a later stage, lead to the potential to work towards institutional interoperability.’
9. The fledgling International Network of AI Safety Institutes—and bodies with equivalent or overlapping functions such as the EU AI Office—should be supported and efforts should be made to expand its membership. ‘Consideration should be given to how this network could support broader AI Safety research initiatives.’
Go deeper: |
| | | |
| |
| | | | Buzzworthy developments of the past few days. - CVS, Humana and UnitedHealthcare cover close to 60% of individuals enrolled in Medicare Advantage. Their respective leadership teams must feel pretty good about that. But federal lawmakers have lately been looking hard at all three. And now an October report from a no-nonsense Senate subcommittee documents how the companies have lately been denying claims for post-acute care—and sometimes using AI to do it. In announcing the report’s release, Democrat Richard Blumenthal of Connecticut issues a reminder: “In December of 2022, a UnitedHealthcare working group met to explore how to use AI and machine learning to predict which denials of post-acute care cases were likely to be appealed.” To this Blumenthal adds that, despite warning flags raised over abuses and excesses, “insurers have continued to deny care to vulnerable seniors—simply to make more money.” The group, which Blumenthal chairs, is the Senate Permanent Subcommittee on Investigations. Full report here.
- Some insurers are self-motivated to try for faster, less hassle-prone prior authorizations with an assist from AI. Count Blue Shield of California among them. The company is partnering with Salesforce to take a shot at approving care in near real time. “I think that’s where the world probably should have been a long time ago,” Salesforce EVP and general manager Jeff Amann tells Yahoo Finance. To this Blue Shield of California CEO Paul Markovich adds: “At no point are we going to be denying care without a human being in the loop and having the availability for the clinician to speak to a qualified [insurance] counterpart.” The plan is to have the AI push through cases that match pre-approval policies and “should be easy to greenlight,” the news outlet reports.
- Jeff Shuren is all for greater international engagement over healthcare AI. At last week’s Medtech conference in Toronto, the former director of the FDA’s Center for Devices and Radiological Health also voiced his keenness for a “voluntary alternative pathway” to right-size reviews of AI-enabled medical devices. The U.S. regulatory system presently handling these products dates to 1976, Shuren pointed out. “[I]t was literally designed for my grandmother’s technology,” he told Regulatory Focus. “It’s been modified a little bit over the years, but it is, in many cases, not fit for purpose for a lot of modern technologies. You’ve just got to fix it.”
- Device makers and regulators alike simultaneously underestimate and overestimate AI. A B2B service vendor’s quality chief hammers this point home. Kellen Giroux of Network Partners foresees manufacturers better balancing the two extremes by investing more heavily in postmarket surveillance than they do now. “The benefits of utilizing AI and machine learning to analyze field data, safety information and customer experiences will greatly benefit how we investigate patient events and beyond,” he predicts. In turn, this approach will inform device updates and algorithm fine-tunings. Giroux made the comments for MD+DI to preview a talk he’ll be giving at MEDevice Silicon Valley in November.
- Ransomware attacks even hurt hospitals that aren’t hit. How so? By forcing them to manage exploding caseloads when victimized neighbors transfer critical patients. Microsoft’s Security Insider notes dangerous spikes at indirectly affected facilities in stroke cases, cardiac arrests, ambulance arrivals, waiting times and more. “The disruption to healthcare operations caused by a ransomware attack can severely impact the ability to effectively treat patients—not only at affected hospitals but also at those in nearby areas,” the post reads. Get the rest.
- Nervous? Who’s nervous? I’m not nervous. So thinks many a surgery patient, it seems safe to assume, before getting wheeled into the OR. Now comes a “frontline digital teammate”—aka a humanlike chatbot—to answer all sorts of questions, offer preparatory tips and otherwise slow down the patient’s racing mind. The helper is the work of Nvidia in partnership with Deloitte. The two developed it with overworked hospital administrators in mind too. “Avatar-based conversational AI agents offer an incredible opportunity to reduce the productivity paradox that our healthcare system faces with digitization,” Deloitte Canada partner Niraj Dalmia says in an Nvidia blog post.
- AI is often a mirror—and we don’t always like what we see when we look in the mirror. So says computer scientist Sanmi Koyejo, PhD, principal investigator at the Stanford Trustworthy AI Research (STAIR). “But sometimes it’s actually useful when a mirror reflects to us some of the gaps in decision-making that individually seem rational, but collectively suggest bias of various kinds.” Koyejo made the remarks at the Penn LDI AI in Health Care Conference. Find out what he meant by all that mirror business here.
- Zoom is veering away from the videoconferencing thing to focus on AI for work. Toward that end it’s partnering with Suki, the AI-powered medical notetaking company, to offer healthcare providers telehealth support. Zoom chief product officer Smita Hashim tells Fast Company the relevant product in the space can reduce documentation overhead by as much as 70%.
- Recent research in the news:
- Notable FDA Approvals:
- Funding news of note:
- From AIin.Healthcare’s news partners:
|
| | | |
|
| |
|
|
