Advancing Government Services With Responsible Generative AI

(ix)    work with the Security, Suitability, and Credentialing Performance Accountability Council to assess mechanisms to streamline and accelerate personnel-vetting requirements, as appropriate, to support AI and fields related to other critical and emerging technologies. (g)  Within 30 days of the date of this order, to increase agency investment in AI, the Technology Modernization Board shall consider, as it deems appropriate and consistent with applicable law, prioritizing funding for AI projects for the Technology Modernization Fund for a period of at least 1 year. Agencies are encouraged to submit to the Technology Modernization Fund project funding proposals that include AI — and particularly generative AI — in service of mission delivery. (ii)   Within 240 days of the date of this order, the Director of NSF shall engage with agencies to identify ongoing work and potential opportunities to incorporate PETs into their operations. (a)  Within 365 days of the date of this order, to prevent unlawful discrimination from AI used for hiring, the Secretary of Labor shall publish guidance for Federal contractors regarding nondiscrimination in hiring involving AI and other technology-based hiring systems.

As a founding member, the United States has played a critical role in guiding GPAI and ensuring it complements the work of the OECD. Many of the current regulations are still being drafted and are therefore often vague in their exact reporting and documentation requirements. However, it is expected that the EU AI law will include several documentation requirements for AI systems that disclose the exact process that went into their creation. This will likely include the origin and lineage of data, details of model training, experiments conducted, and the creation of prompts.

Fortune Cyber 60

The public sector deals with large amounts of data, so increasing efficiency is key., AI and automation can help increase processing speed, minimize costs, and provide services to the public faster. While safety is critical, some argue that government regulation of AI could also serve as a “wolf in sheep’s clothing” — a means to consolidate control over AI gains in the hands of a few. As Yann LeCun recently called out, leaders of major AI companies including Altman, Hassabis, and Amodei may be motivated by regulatory capture more than broad safety concerns. Andrew Ng has made similar arguments that there is a financial incentive to spread fear.

With these guardrails, we are working to help enterprises seamlessly manage security and compliance capabilities as they scale their AI workloads. Artificial intelligence companies and governments should allocate at least one third of their AI research and development funding to ensuring the safety and ethical use of the systems, top AI researchers said in a paper on Tuesday. Based on current trends, creating frontier AI models will likely soon cost upward of hundreds of millions of dollars in computational power and also require other scarce resources like relevant talent. The regulatory approach we describe would therefore likely target only the handful of well-resourced companies developing these models, while posing few or no burdens on other developers. Nonetheless, by increasing the burdens to those developing the most advanced systems, the market for such systems may become more concentrated.

Part V: Implementation and Enforcement

Non-proliferation of certain frontier AI models is therefore essential for safety; but it is difficult to achieve. As AI models become more useful in strategically important contexts, and the costs of producing the most advanced models increase, AI companies will face strong incentives to deploy their models widely—even without adequate safeguards—to recoup their significant upfront investment. But even if companies agree not to distribute their models, bad actors may launch increasingly sophisticated attempts to steal them. Much recent progress in AI has stemmed from harnessing huge amounts of computational power to train a handful of systems. One analysis finds that the computing power (compute) employed to develop noteworthy AI systems has increased by 4.2 times every year.

The objective of regulation should be to compel users and AI vendors to take care not to increase the risks of the mentioned negative things happening. And new AI applications will necessitate frequent revisions of regulations as they arise. The US took broad and sweeping action to regulate AI, covering many areas of civil society.

This ensures that users can maintain for government cloud operations. Government agencies – or the public sector in general, with strict data regulations – must use systems you can trust to keep your data safe in your environment. Microsoft recommends setting up a sandbox with Azure Open AI and what is called GovChatGPT, where organizations can have an isolated environment to start testing and learning what AI can do. High-risk activities — like AI use in educational fields or training, law enforcement, assistance in legal actions, the management of critical infrastructure and other similar activities — would be allowed, but heavily regulated. There is even an entire section in the AI Act that applies to generative AI, allowing the technology but requiring users to disclose whenever content is AI-generated.

• A Governing magazine report found that 53% of local government officials cannot complete their work on time due to low operational efficiencies like manual paperwork, data collection, and reporting.
• Specifically, these assets include the datasets used to train the models, the algorithms themselves, system and model details such as which tools are used and the structure of the models, storage and compute resources holding these assets, and the deployed AI systems themselves.
• (iii)  consider providing, as appropriate and consistent with applicable law, guidance, technical assistance, and training to State, local, Tribal, and territorial investigators and prosecutors on best practices for investigating and prosecuting civil rights violations and discrimination related to automated systems, including AI.
• One major step is the enactment of strict laws and regulations governing the collection, storage, and use of individuals’ personal data.
• As enterprises look to address these requirements and achieve growth while adopting innovative AI and hybrid cloud technologies, IBM will continue to meet clients wherever they are in their journeys by helping them make workload placement decisions based on resiliency, performance, security, compliance and cost.

However, starting small, focusing on citizen needs, and communicating benefits and limitations clearly can help agencies overcome barriers. The public sector can navigate obstacles to harness AI responsibly with proper care and partnerships. Within the public sector, conversational AI has the potential to augment and even fully automate aspects of citizen services by providing 24/7 support for everyday administrative tasks. More advanced copilots can also assist public employees in their day-to-day work, reducing process ambiguity and friction points. Microsoft’s AI rollout builds upon the June launch of its Azure OpenAI Service for the government to allow federal agencies to use powerful language models to run within the company’s cloud service for U.S. government agencies, Azure Government. For example, AWS Premier Partner Leidos recently held an Experience-Based Accelerator (EBA) with AWS to build a generative AI training module based on large language models using AWS services.

How are SAIF and Responsible AI related?

Backlogs also result when models are not centrally discoverable, making reuse and reproducibility challenging. In addition to government-to-government cooperation, partnerships with international organizations such as the United Nations or Interpol could play a very important role. These organizations offer platforms for conversations and coordination between countries on issues related to data privacy and security. Google has an imperative to build AI responsibly, and to empower others to do the same.

Toan brings his leadership experience and industry knowledge of data management, security, and analytics to the United States Federal government, where he helps the largest agencies understand data to better inform their decisions in the most critical missions. The AI Bill of Rights serves as a foundational document, acknowledging the importance of upholding human rights in an AI-driven world. It offers a framework for ensuring that AI technologies are developed and used responsibly, ethically and in ways that benefit humanity rather than harm it. Since the launch of OpenAI’s generative AI models, top academics and prominent CEOs such as Elon Musk have warned about the risks on AI, including calling for a six-month pause in developing powerful AI systems. The U.S. government still lacks many of the authorities needed to act on any concerning information it may receive.

Better manage risk and compliance

The AI Safety Summit in the UK and the Executive Order on Safe, Secure and Trustworthy Artificial Intelligence in the US signal an intensification of government intervention in artificial intelligence. Both events demonstrate a growing commitment to address concerns about AI risks that have entered the public consciousness. The term artificial intelligence, or AI, dates back to the 1950s and centered on the idea that the human brain could be mechanized. Over the decades since, scientists have studied how humans think and learn and tried to apply those same methods to machines and data. Today, AI refers to a computerized system that carries out a behavior that typically would require human knowledge.

Connected infrastructure has led to attacks with hundreds of millions of dollars of economic loss. The warning signs of AI attacks may be written in bytes, but we can see them and what they portend. Improve intrusion detection systems to better detect when assets have been compromised and to detect patterns of behavior indicative of an adversary formulating an attack. If key types of data are either missing from or not sufficiently represented in a collected dataset, the resulting AI system will not be able to function properly when it encounters situations not represented in its dataset.

Automation of prompts against data sets.

Regardless of the reason for doing so, placing AI models on edge devices makes protecting them more difficult. Because these edge devices have a physical component (e.g., as is the case with vehicles, weapons, and drones), they may fall into an adversary’s hands. Care must be taken that if these systems are captured or controlled, they cannot be examined or disassembled in order to aid in crafting an attack. In other contexts, such as with consumer products, adversaries will physically own the device along with the model (e.g., an adversary can buy a self-driving car in order to acquire the model that is stored on the vehicle’s on-board computer to help in crafting attacks against other self-driving cars).

Read more about Secure and Compliant AI for Governments here.