AllYouCanTech

Critics Call on Governor to Reject Disputed California AI Safety Legislation After It Secures Legislative Endorsement

## California’s AI Safety Bill: A Crucial Turning Point for AI Oversight

### Introduction

California stands poised to enact a transformative law that may establish a benchmark for artificial intelligence (AI) oversight throughout the United States. Senate Bill 1047 (SB-1047), which has gained substantial backing in both the California State Assembly and Senate, is now pending Governor Gavin Newsom’s approval. Sponsored by State Senator Scott Wiener, this legislation seeks to impose rigorous safety standards for large AI models that could introduce new risks to public safety and security. Nevertheless, the bill has ignited a fervent debate, with advocates and opponents presenting fundamentally opposing perspectives on its possible effects.

### The Essence of SB-1047

At its core, SB-1047 requires that developers of large AI models—those with training costs exceeding $100 million—install a “kill switch.” This feature would facilitate the swift deactivation of an AI system should it start to display behaviors that threaten public safety, particularly when functioning with limited human supervision. By concentrating on larger models, the bill aims to prevent hindering smaller startups, which may lack the capacity to meet such stringent regulations.

Supporters of the bill contend that it is a vital stride toward the responsible advancement of AI technologies. They highlight the swift growth in AI capabilities and the risk of these systems behaving in unpredictable and potentially perilous manners. Geoffrey Hinton and Yoshua Bengio, two prominent figures in the AI domain, have expressed their endorsement of the bill, stressing the importance of external oversight to safeguard public welfare.

### The Dispute: Safety Versus Innovation

Regardless of its good intentions, SB-1047 has encountered notable opposition from various stakeholders. One prominent critic is Fei-Fei Li, a Stanford University computer science professor and distinguished AI authority. In a recent opinion piece, Li asserted that, although the bill is well-intentioned, it risks causing unintended consequences that could hinder innovation not only in California but nationwide. She voiced apprehensions that the bill’s assignment of liability to original developers of altered AI models might deter open-source collaboration, which is essential for academic inquiry and the wider AI community.

Li’s worries are reflected by a coalition of California business executives who have petitioned Governor Newsom to reject the bill. In an open letter, they contended that SB-1047 misappropriately zeroes in on regulating model development as opposed to addressing the misuse of AI technologies. They cautioned that the bill could impose hefty compliance costs and create regulatory confusion, potentially discouraging investment and innovation within the state.

### Governor Newsom’s Predicament

Governor Gavin Newsom now confronts a challenging choice. On one hand, the considerable legislative backing for SB-1047 indicates strong political momentum to pass the bill. Should Newsom opt for a veto, the legislature might be able to override his decision with a two-thirds majority in both chambers—a plausible scenario given the current support for the measure.

Conversely, Newsom has raised concerns regarding excessive regulation of the AI sector. During a UC Berkeley Symposium in May, he pointed out the dangers of overregulation, which could place California in a “dangerous position.” Nonetheless, he also recognized the extraordinary circumstances surrounding AI, where even the technology’s creators are advocating for regulation. “When you have the inventors of this technology, the godmothers and fathers, saying, ‘Help, you need to regulate us,’ that’s a very different environment,” Newsom remarked.

### The Broader Consequences

The result of SB-1047 could have extensive ramifications, not just for California but for the entire AI landscape in the United States and beyond. If the bill is enacted, it might serve as an example for other states and potentially even inform federal regulations. It would also indicate a transition towards more proactive management of AI technologies, concentrating on preemptive actions rather than reactive responses.

Conversely, if the bill is vetoed, it could postpone the establishment of AI safety regulations and leave room for ongoing discussions on how best to balance innovation with public safety. This decision will also likely affect how other states address AI oversight, either motivating them to emulate California’s approach or to take a more cautious stance.

### Conclusion

As the deadline for Governor Newsom’s decision nears, all eyes are focused on California. SB-1047 marks a pivotal moment in the continued discourse surrounding AI regulation, with substantial consequences for the future of AI development. Whether the bill is enacted or vetoed, the dialogue it has ignited will undoubtedly persist, influencing the path of AI governance in the years that lie ahead.

Expert Cautions Chatbots May Allow Police to Alter Reports

### The Surge of AI in Law Enforcement: A Double-Edged Blade?

In recent times, the incorporation of artificial intelligence (AI) into diverse fields has been both praised and analyzed. Law enforcement is no different. The emergence of AI-driven tools like Axon’s Draft One, which produces police reports from body camera recordings, has ignited an intense discourse regarding the future of policing and the judicial system. While the technology claims to enhance time efficiency and operational productivity, it simultaneously provokes substantial issues surrounding accuracy, bias, and the risk of misuse.

#### The Potential of AI in Law Enforcement

Draft One, created by Axon—a firm recognized for its tasers and body cameras—marks a notable advancement in policing technology. This tool, powered by OpenAI’s GPT-4 model, can craft comprehensive police reports within minutes following an incident, relying solely on audio captured from body cameras. This feature is particularly enticing to police agencies, where officers typically invest hours in report writing—a duty many perceive as onerous.

In June 2024, the police department in Frederick, Colorado, became the inaugural department globally to adopt Draft One. The department noted that the tool significantly minimized the time officers devoted to paperwork, enabling them to concentrate more on their core responsibilities. Other departments throughout the United States swiftly adopted similar measures, eager to enjoy the advantages of this groundbreaking technology.

Axon has positioned Draft One as a transformative instrument that can “accelerate justice” by negating the necessity for manual data entry. The firm asserts that the AI-generated reports are as precise, if not more so, than those composed by humans. In a double-blind evaluation, Axon discovered that Draft One’s reports were equivalent to or superior to those created by humans concerning completeness, neutrality, objectivity, terminology, and coherence.

#### The Dangers and Issues

Notwithstanding the evident advantages, the rollout of AI-generated police reports has triggered concerns among legal professionals, civil rights proponents, and digital rights organizations. The primary apprehension involves the capacity for AI to incorporate errors, biases, or even intentional inaccuracies into police reports—documents that are essential to the justice framework.

One of the major dangers is the chance of AI “hallucinations,” where the system fabricates information that is inaccurate or unrelated to the incident. While Axon has asserted that it has mitigated such errors by “dialing down the creativity” of Draft One, the risk lingers. AI systems, including those based on GPT-4, are recognized for occasionally producing erroneous or misleading data, particularly when navigating intricate or nuanced contexts.

Furthermore, reliance on AI-generated reports could exacerbate pre-existing biases in policing. Body camera footage, which documents events from the officer’s viewpoint, already possesses the potential to skew the narrative in favor of law enforcement. If AI systems are trained on biased datasets or are affected by how officers articulate their observations, they could reinforce these biases, culminating in unjust results.

Legal scholars such as Andrew Ferguson have cautioned that AI-generated reports could “digitally contaminate” the evidence-based evolution of criminal proceedings. The apprehension is that AI might subtly modify the narrative in manners that skew police perspectives or mislead judicial entities. This could have expansive implications, especially in instances where the police report constitutes the principal piece of evidence.

#### The Risk of Misapplication

Another issue is the potential for the misuse of AI-generated reports. As the technology gains traction, there exists a risk that officers may employ it to manipulate the portrayal of an incident. For instance, an officer could consciously phrase their comments to sway the AI’s interpretation, leading to a report that favors the officer’s rendition of events, even if it lacks total accuracy.

This concern is heightened by the reality that AI-generated reports might be utilized in more serious scenarios, despite initial advisories to restrict their application to minor occurrences. In some jurisdictions, officers have already begun applying Draft One to a wider array of cases, raising alarms regarding the accuracy and dependability of the reports in more intricate situations.

Civil rights advocates are apprehensive that the extensive integration of AI-generated reports could lead to heightened police scrutiny, particularly in marginalized populations. If the technology simplifies the reporting process for officers, they may become more prone to pursue charges in circumstances where they might have earlier decided to dismiss the matter.

#### The Call for Transparency and Oversight

In light of the potential dangers, experts are advocating for enhanced transparency and oversight concerning the usage of AI-generated police reports. Ferguson has suggested that any implementation of this technology in legal contexts should be accompanied by thorough documentation regarding how AI models were developed, what data they utilized, and how the models were assessed. Such transparency is vital for ensuring that the reports are both accurate and trustworthy.

Additionally, there is an essential need for independent evaluation and auditing of AI-generated reports. Civil rights organizations like the Electronic Frontier Foundation (EFF) have urged police agencies to grant access to these reports for further examination. This would enable impartial experts to evaluate the precision and equitability of the AI-generated reports.

EU Probes If Telegram Downplayed User Figures to Avoid Regulation

### Telegram under EU Examination: An In-Depth Analysis of the Ongoing Probe

Telegram, the well-known messaging platform renowned for its encrypted communication features, is presently undergoing scrutiny by the European Union (EU) for potential violations of the Digital Services Act (DSA). The investigation focuses on whether Telegram has accurately reported its user figures within the EU, a vital aspect that influences the level of regulatory scrutiny the platform must meet.

#### The Heart of the Inquiry

The EU’s apprehensions arise from doubts that Telegram might have downplayed its user count in the area to avoid surpassing the 45 million user benchmark. Platforms that exceed this threshold are categorized as “very large online platforms” (VLOPs) under the DSA, which subjects them to stricter regulations. These regulations encompass improved content moderation, independent audits, and necessary data sharing with the European Commission.

As of February 2024, Telegram claimed to have 41 million users in the EU. However, it did not furnish an updated count as mandated by the DSA, instead indicating that it had “significantly fewer than 45 million average monthly active users in the EU.” This lack of clarity has led the EU to commence a technical investigation to ascertain the true number of Telegram users in the area.

#### Consequences of Being Designated a VLOP

Should Telegram be found to exceed 45 million users in the EU, it would be categorized as a VLOP, resulting in a range of additional responsibilities. These responsibilities include:

– **Content Moderation:** Telegram would be required to establish more effective systems for monitoring and removing illicit content, including misinformation and propaganda.
– **Data Sharing:** The platform would need to share more information with the European Commission, ensuring adherence to EU regulations.
– **Independent Audits:** Routine assessments by external organizations would become obligatory to confirm that Telegram complies with the DSA’s stipulations.
– **Advertising Restrictions:** The platform would be barred from targeting advertisements based on sensitive user information such as religion, gender, or sexual orientation.

These measures form part of the EU’s wider initiative to limit the dominance of large online platforms and ensure they function in a way that safeguards user rights and public safety.

#### Telegram’s Global Popularity and Challenges

In recent years, Telegram has experienced a significant rise in popularity, boasting a global user base approaching 1 billion. The app is especially favored for its encrypted messaging capabilities, offering enhanced privacy compared to other services. Nevertheless, this same aspect has rendered Telegram a hotspot for illegal activities, prompting investigations in various countries, including France.

Concurrently, French authorities are investigating Telegram for supposed criminal activities facilitated through the platform. This situation has resulted in the arrest of its founder, Pavel Durov, a billionaire originally from Russia who has since obtained French-Emirati citizenship. Durov, who left Russia in 2014 after refusing to comply with Moscow’s demands for access to Ukrainian user data, has denied any illicit behavior, asserting that he has “nothing to hide.”

#### The EU’s Future Actions

The EU’s Joint Research Centre, its internal data and science body, is currently conducting a technical probe to ascertain the validity of Telegram’s user statistics. Thomas Regnier, a spokesperson for digital affairs at the European Commission, stated that the EU possesses its own systems and analyses to confirm user data. If Telegram is found to have provided misleading information, the EU could independently classify it as a VLOP based on its findings.

The DSA regulations for VLOPs came into effect a year ago, affecting some of the world’s most significant online platforms, including Instagram, Google, and TikTok. These enterprises have needed to significantly enhance their compliance initiatives, employing thousands to meet the DSA’s rigorous demands. Some platforms have even pursued legal action against the EU, contending that the regulations are excessively demanding.

#### Final Thoughts

The ongoing inquiry into Telegram’s adherence to the DSA marks an important milestone in the EU’s strategy to regulate large online platforms. If it is determined that Telegram has more than 45 million users in the EU, it will encounter a range of new responsibilities aimed at promoting greater transparency, accountability, and user protection. As the investigation progresses, it will be essential to observe how Telegram reacts and whether it can fulfill the EU’s rigorous regulatory standards.

This situation also underscores the broader challenges facing global tech firms as they navigate increasingly intricate regulatory environments. With the EU at the forefront of digital regulation, other regions may follow suit soon, making compliance a paramount concern for platforms such as Telegram operating on a worldwide scale.