The U.S. has undertaken a two-pronged effort to close the broadband digital divide. COVID-era legislation such as the Broadband Equity, Access, and Deployment (BEAD) Program is providing tens of billions to wire up unserved areas. And the Affordable Connectivity Program (ACP) provides a subsidy that effectively makes broadband free for as many as 50 million eligible households, based on an analysis of pre-pandemic data. The eligibility criteria are broad, encompassing any household with incomes at or below 200% of the federal poverty guidelines, or participation in any one of a long list of assistance programs, including Pell Grants and Medicaid.

With this sort of focused effort on providing affordable connectivity to everyone, what could go wrong? Some groups have alleged that the broadband industry is still discriminating against poor neighborhoods. Their core claim is that wealthier neighborhoods attract multiple broadband providers, and therefore the government should pay to build multiple networks in poor neighborhoods to create more competition.

For example, a recent study by the California Community Foundation and Digital Equity LA asserts that, within Los Angeles County, the price of high-speed internet is lower in higher-income areas, leaving poorer neighborhoods paying more for access. The claim is that wealthier neighborhoods reap the benefits of competition.

However, the study, while addressing an important issue, suffers from problems  almost too numerous to mention. First, the study is based on only 165 non-randomly chosen residential addresses in LA County, which has more than 3 million households. Second, any disparities in advertised pricing are irrelevant, since with the ACP funding, those most in need of financial assistance can get broadband at effectively zero monthly cost. Third, the preferred solution in the study — to “support independent, community-driven options for internet service” — sounds great until you realize how much money it would cost to lay an additional broadband network in Los Angeles. Indeed, operating a broadband network is an expensive and risky activity: Starry, an independent provider of broadband services in five cities, just filed for bankruptcy.

But the biggest issue with the study is that it is conceptually misguided. With funding in the pipeline for network expansion, and the ACP subsidies widely available, it’s become clear that getting people to sign up for “costless” internet is the major remaining problem to closing the broadband equity gap — the so-called adoption gap.

There’s plenty of evidence that much of the adoption gap is not tied to price. As we noted in our October 2022 blog post, the latest NTIA Internet Use Survey (collected in November 2021, before the ACP became active) showed that only 18% of households without internet access at home cited cost as a reason for being offline. In the NTIA survey, 58% of the offline households “express no interest or need to be online.”

How can the adoption gap be closed? First, building expensive new networks in areas that already have high-speed broadband, like Los Angeles County, is not the way to go. That money could be better spent on extending funding for the ACP.

Instead, we need programs to improve digital literacy and show, step-by-step, how to get online and utilize government and private resources. The Biden Administration has pointed to education efforts as being one of the main drivers of enrollment in ACP, with examples such as a text being sent out to 1.3 million likely eligible households in the state of Michigan resulting in 25,000 additional enrollments, or 1 million SNAP and TANF beneficiaries in Massachusetts being notified about the program resulting in enrollment doubling over just the next 5 days. This kind of effort is not new in LA County where in December 2021, the County launched an outreach campaign for ACP’s predecessor program which generated over 40,000 sign-ups in one month. More recently, numerous LA-based organizations, including the County itself, received modest grants from the FCC’s ACP Outreach Grant Program. Building on these investments and leveraging outreach via multiple organizations will help educate consumers on the benefits of broadband and drive adoption of service.  Continued education efforts are the key to maintaining this momentum.

But by themselves, such programs are not enough since adoption is driven by financial inclusion as well. Without a credit card or a bank account, it’s much harder to use online services such as e-commerce and ride-sharing. Indeed, it’s possible that the people who express no interest in being online are actually responding to these other obstacles. Data show that 4.5% of the U.S. population currently has no access to the financial system, and an additional 14% are “underbanked,” meaning they primarily rely on nonbank transactions. For these households, which include some of those most in need of financial assistance, this acts as an additional barrier to obtaining connectivity.

It’s true that high-speed internet is a sort of prerequisite to participate in today’s society, but if cost is not the barrier for consumers, as is the case with current subsidy programs, examining pricing diverts attention away from the real barriers. In order to fully close the digital divide, the question of internet adoption is what must be addressed by communities and policymakers.

Who has the “right to repair” a product after it’s been purchased? The intuitive answer would be the person or business that purchased it, but in the case of products with sophisticated technology and software — an increasingly common characteristic for just about any product — this becomes more complicated. In these cases, repairs likely require specialized equipment or software programs from the manufacturers, which in turn runs a higher risk of exposing trade secrets or reducing the security built into the product by requiring backdoor entry for repairs or diagnostics.

Legislators have grappled with trying to find this balance for some time. For example, take the case of cars, which now have complex onboard computers. Massachusetts passed a comprehensive right to repair bill in 2013. It required that manufacturers allow independent mechanics access to the same diagnostic and repair information provided to franchised dealerships. Similar issues have arisen around repairs for other complex technologies such as medical and agricultural equipment, for which 11 states have introduced legislation creating a general right to repair.

But consumer tech, such as smartphones, raises the issue of security to a new level. Opening any device to repair by third parties increases its vulnerability to unauthorized, unwanted software such as malware or spyware. It also gives third-parties access to sensitive user data which, without proper vetting of the third party, may put consumers at risk. Things like phones, tablets, and laptops store an enormous amount of personal data, and by ensuring third parties can access devices, backdoor security risks are inevitable.

Device makers have legitimate concerns about brand integrity and proprietary information. Consumers may blame manufacturers if unauthorized repairs damage products in ways that may not be immediately obvious, either because of software issues or incorrect assembly. In these cases, there is also the risk that any third party looking to get into the repair business will have access to technology and potential trade secrets, which would otherwise be the property of the manufacturer.

In an attempt to balance the benefits of third-party repair with potential business and consumer harms, New York has passed the Digital Fair Repair Act, which will go into effect this summer. A sort of compromise legislation, the New York approach will require manufacturers to provide manuals and parts to both consumers and independent repair providers, but with a few significant limitations.

Notably, the bill does not require any sort of backdoor access to bypass locks on a device. This provision ensures that third parties, and even consumers themselves, will be able to make repairs, but does not give up control entirely in a way that compromises the basic level of data security offered by the device. The bill also shields the original manufacturer from legal liability for damages to a product by a third-party during repair, while allowing the manufacturer to supply assemblies of parts rather than individual pieces.

Nonetheless, state efforts to address repairs for consumer tech goods point to the larger question of the need for right to repair legislation at the federal level. As is true in most cases which impact national industry, a national framework would be better than a patchwork of state legislation. However, as we observe the impacts of the legislation in New York, careful consideration must be given to the ways in which consumers may be harmed by regulatory action. It also must be considered how these fit into existing competition law with regard to aftermarkets, and the impacts to businesses outside of the consumer tech space.

Though the momentum currently lies in the states, Congress will ultimately have to act to create a national standard for right to repair. In the meantime, states should heed New York’s example and ensure any bill has the safety provisions necessary to ensure consumer protections, in contrast with bills currently moving in states like Washington and Minnesota, which both take a looser approach to repair where third parties are allowed more full access to devices for repair. Until Congress is ready to address this issue for both consumer tech products and other high-tech devices, states should take a cautious approach to legislation that will impact the national market.

Artificial intelligence (AI) is the new subject of large-scale regulation by governments around the world. While AI has many benefits, such as increased productivity and cost savings, it also presents some risks and challenges. For example, AI systems can sometimes be biased or discriminatory, leading to unfair outcomes. They can also raise concerns about privacy and data security, as these systems often rely on large amounts of personal data.

As a result, governments around the world are starting to introduce regulations to ensure that AI is developed and used in a safe, responsible, and ethical manner. These regulations cover a range of issues, from data privacy and security to algorithmic transparency and accountability.

This piece will unpack the novel AI regulation in the U.S., EU, Canada, and China and how each country approaches the technology as they seek to balance economic, social, and public priorities with innovation.

European Union: Artificial Intelligence Act (AIA)
The European Union introduced the Artificial Intelligence Act (AIA) on April 21, 2021. The current text proposes a risk-based approach to guide the use of AI in both the private and public sectors. The approach defines three risk categories: unacceptable risk applications, high-risk applications, and applications not explicitly banned. The regulation prohibits the use of AI in critical services that could threaten livelihoods or encourage destructive behavior but allows the technology to be used in other sensitive sectors, such as health, with maximum safety and efficacy checks by regulators. The legislation is still under review in the European Parliament.

The AI Act is a type of legislation that regulates all automated technology rather than specific areas of concern. It defines AI systems to include a wide range of automated decision-makers, such as algorithms, machine learning tools, and logic tools, even though some of these technologies are not considered AI.

Canada: The Artificial Intelligence and Data Act (AIDA)
In June of 2022, Canadian Parliament introduced a draft regulatory framework for Artificial Intelligence using a modified risk-based approach. The bill has three pillars, but this piece will just examine the section dealing with AI, the Artificial Intelligence and Data Act (AIDA). The goal of Canada’s AI rules are to standardize private companies’ design and development of AI across the provinces and territories.

The modified risk-based approach is different from the EU’s approach as it does not ban the use of automated decision-making tools, even in critical areas. Instead, under the AIDA regulation, developers must create a mitigation plan to reduce risks and increase transparency when using AI in high-risk systems. The plan should ensure that the tools do not violate anti-discrimination laws. These mitigation plans or impact assessments aim to decrease risk and increase transparency in the use of AI in social, business, and political systems.

United States: AI Bill of Rights and State Initiatives
The United States has yet to pass federal legislation governing AI applications. Instead, the Biden Administration and the National Institute of Standards and Technology (NIST) have published broad AI guidance for the safe use of AI. In addition, state and city governments are pursuing their own regulations and task forces for AI use. In a break from the EU model, regulation thus far targets specific use cases rather than seeking to regulate AI technology as a whole.

At the federal level, the Biden Administration recently released the AI Bill of Rights, which addresses concerns about AI misuse and provides recommendations for safely using AI tools in both the public and private sectors. This AI strategy would not be legally binding. Instead, the Bill of Rights calls for key safety strategies such as greater data privacy, protections against algorithmic discrimination, and guidance on how to prioritize safe and effective AI tools. While the blueprint is not legally binding, it serves as a guide for lawmakers at all levels of government who are considering AI regulation.

In addition, NIST, which is an agency in the Department of Commerce that develops technology standards, published standards for managing AI bias. NIST also tracks how the public sector integrates AI tools across the federal government.

In 2022, 15 states and localities proposed or passed legislation concerning AI. Some bills focus on regulating AI tools in the private sector, while others set standards for public-sector AI use. New York City introduced one of the first AI laws in the U.S., effective from January 2023, which aims to prevent AI bias in the employment process. Colorado and Vermont created task forces to study AI applications, such as facial recognition, at the state level.

China: Algorithm Transparency and Promoting AI Industry Development
China has set a goal for the private AI industry to make $154 billion annually by 2030. China has yet to pass rules on AI technology at large. Recently, however, the country introduced a law that regulates how private companies use online algorithms for consumer marketing. The law requires companies to inform users of AI for marketing purposes and bans the use of customer financial data to advertise the same product at different prices. However, not surprisingly, the law does not apply to the Chinese government’s use of AI.

Along with China’s federal regulation, in September of 2022, Shanghai became the first province to pass a law focused on private-sector AI development. The law titled Shanghai Regulations on Promoting the Development of the AI Industry provides a framework for companies in the region to develop their AI products in line with non-Chinese AI regulations.

Next Steps for Global Regulation:
Artificial intelligence is a promising tool that is stimulating global growth and driving the future of innovation. Despite the positive impacts of AI, there is no question that some regulation is needed to combat the misuse of AI and to protect consumers.

The different approaches summed in this piece offer methodologies for how policymakers around the world are approaching specific harms from AI, as well as AI as a whole. The EU’s approach regulates the use of any automated decision-making tools and outlines the sectors where they can and cannot be used. The U.S. offers voluntary recommendations and standards at the federal level, with states and cities pursuing their own targeted studies and rules based on specific harms. The modified risk-based approach in Canada regulates all AI tools but stops short of banning the technology in certain spheres by allowing companies to define their own risk-mitigation strategies. And the Chinese approach seeks to increase transparency for consumers and become a global power in AI standards.

To prepare, companies will need to further develop global stances on AI ethics and compliance for their products in order to meet transforming regulations. In addition, legislators should focus on legitimate harms to consumers and keep apprised of how stricter regulatory regimes affect AI innovation.

In Britain, Parliament is debating the oft-revised U.K. Online Safety Bill, which seeks to regulate harmful and illegal content on the internet for children and adults. Without further modifications, however, the bill could undermine that laudable goal.

Fashioned by Ofcom, the U.K.’s digital communications regulator, the bill creates a formal “duty of care” for online platforms to remove harmful content on their websites with additional responsibilities for websites that also serve young people. If passed, adult websites would need to establish age verification and commit to removing illegal content such as content depicting hate crimes, sexual abuse, and terrorism. Websites with youth users would need to prevent young people from seeing illegal content as well as potentially harmful, but not illegal, content such as that promoting eating disorders, self-harm, or suicide. Penalties for failure to comply with these requirements include fines and potential criminal liability for platforms and tech executives if they fail to comply with Ofcom’s information requests.

As I’ve written previously, managing harmful content online, and creating specific youth-protection schemes, raises a host of thorny questions. In a recent blog looking at a U.S. youth online safety bill, I wrote that content moderation is technically challenging, because content moderation systems aren’t perfectly accurate at flagging inappropriate content. The systems can overcorrect and potentially censor acceptable information. In another piece about children’s privacy, I noted that definitively identifying internet users’ age online is both technically challenging and undermines everyone’s privacy.

The latest version of the U.K. Online Safety Bill fails to strike the right balance on content moderation versus censorship, or on privacy.

Content moderation creates a host of challenges, like defining what content is harmful, censorship of content that is not illegal, and encoding all the rules into an algorithm. Even with clear definitions, content moderation is technically challenging.

Content moderation algorithms are not perfectly accurate at flagging harmful content. This means that if an algorithm is 98% effective, 2% will slip through the cracks. On popular websites, 2% could be millions of posts. It remains unclear if firms will be punished if some illegal content passes through the filters. Free speech advocates worry that firms will overcompensate content removal because of the harsh penalties and the technical compliance challenge.

In addition to challenges of censorship, the bill also threatens privacy. First, by requiring users to disclose more personal data to access websites. Second by requiring a backdoor on encrypted messaging platforms for Ofcom to scan private messages for illegal conversations or content.

There is no easy way currently to verify age online without the user voluntarily giving up personal information. While the main trend in privacy legislation is toward decreasing the amount of personal information consumers have to provide to apps and websites, the U.K. bill points in the opposite direction, requiring users to make accounts and verify age through technologies like facial recognition checks or uploading a government-issued ID card to a website.

If personal information disclosure for every website wasn’t enough, the bill also contains provisions to allow Ofcom to search private messages on encrypted platforms. The bill as currently written gives new, large scale, citizen surveillance abilities to Ofcom regardless of wrongdoing. Seventy organizations, cyber security experts, and elected officials signed a letter to warn against allowing Ofcom to search private messages. Their main message is that creating a backdoor for the government creates a backdoor for anyone and undermines the rights of British citizens to privacy.

The U.K. government isn’t the only entity searching for a better solution to make the online space safer for users. This bill, however, stifles free speech online while undermining too many peoples’ privacy.

With the Supreme Court set to take on Gonzalez v. Google this term — a case with momentous implications for the legal viability of internet services as we know them — the fate of Section 230 of the Communications Decency Act is in question.

Section 230 is the statute that grants online platforms protection from liability for the content posted by their users, a fundamental protection that has been integral to the internet ecosystem’s explosive growth. By allowing internet platforms to take down third-party content that they deem harmful to their users in “good faith,” while also ensuring that they are not treated as the publishers of such content, Section 230 is the legal mechanism that has facilitated innovative business models which give a platform to user-generated content, shaping a robust digital economy enjoyed by both consumers and entrepreneurs today.

From a consumer standpoint, these business models provide a plethora of free resources, entertainment, and educational materials. In the case of entrepreneurs, the online creator economy is estimated to be worth more than $100 billion worldwide, with more than 425,000 full-time equivalent jobs reportedly supported by the YouTube platform alone.

In Gonzalez v. Google, however, the central question goes beyond Section 230’s protections for third-party content, asking instead whether the targeted algorithms employed by these platforms enjoy the same protections.

Most major online platforms use data at various levels to recommend content to users, whether by using specific personal or demographic information to tailor the experience to a user’s specific interests, or by presenting users with popular or relevant content at the top of the feed. These automated decisions curate a feed appealing to the user and beneficial to the content creator, whose work is then highlighted to new audiences.

Colloquially, those referring to social media algorithms are most likely referring to the sophisticated code used by many of these companies to target content to their users. However, from a technological standpoint, the term “algorithm” refers to any type of content sorting — whether it be a simpler iteration that might show content ordered chronologically or alphabetically, or the more complex, individually curated version. There is no default method of content sorting, meaning that every company or developer must choose an algorithm to sort content. The only difference lies in the complexity of the algorithm they chose to employ.

It is difficult to draw legal lines around this complexity. For example, if a platform lists a seemingly harmful piece of content first, thus making it the most obvious choice for users to select, are they liable for that content only if proven to be the result of a curated algorithm, or are they also liable if the reason it is listed first is that the feed is shown chronologically? Either way, there is some risk of exposing users to harmful third-party content. Prohibiting one platform’s algorithms — say Google’s — thus doesn’t provide a general solution.

That means the Supreme Court would either have to define algorithms in such a way that only specific types are implicated for liability, or rule that Section 230 liability protections are lost for all types of content displayed online.

Let’s be clear: A court decision that ended Section 230’s liability protections would make hosting third-party content functionally impossible for websites. On YouTube alone there are over 500 hours of content uploaded every minute, making vetting every video prior to each upload a monumental task. The risk of getting sued will lead most companies to conclude that it’s not worth it for them to offer third-party content. And in the case where only data-driven, targeted algorithms are ruled to be exposed to liability, how likely is it that users would sort through 500 hours of content with no curation in hopes of discovering useful information?

Moreover, the ramifications of making all content providers liable to lawsuits will spread across the entire Internet ecosystem, including online shopping, travel sites, and app stores, all of which rely on user reviews that are curated to reduce fakes and “ballot-stuffing.” In an era of deep fakes and sophisticated artificial intelligence chatbots, it’s all the more essential for online platforms to be able to apply algorithms that users can trust.

There’s no doubt that Section 230 raises difficult issues that need to be carefully considered by policymakers. But subjecting online platforms to lawsuits because their algorithms occasionally highlight content that someone objects to would fundamentally destroy the internet economy, while failing to address the threat posed by truly dangerous online content.

Will the combination of high cost increases and low digital inflation spur reluctant companies to digitize?

One of the pleasant economic surprises in recent months has been the low rate of inflation for digital goods and services, compared to the overall inflationary surge. The latest producer price report, released November 15, shows that a basket of digital goods and services (described below) had a median year-over-year price increase of 1.9%. By comparison, the overall year-over-year price increase for final demand, less food and energy, was 6.7%.

We wrote about this big gap between “New Economy” digital inflation and “old economy” inflation  in a December 2021 blog item. In June 2022, leading economic statistician Marshall Reinsdorf wrote a paper for PPI examining the continued slow rate of price increases for most digital goods and services.

The growing gap between overall inflation and digital inflation means that the relative price of digital goods and services is falling. To put it another way, in the low-inflation era that preceded the pandemic, many companies  enjoyed the benefit of low costs without having to make expensive and potentially risky investments in digitizing their operations.

Now that easy period is over. Companies are looking at technology as a way out of their high-cost trap. Business spending on software, computers, and communication gear hit an all time high in the third quarter of 2022. The layoffs at companies such as Amazon and Facebook notwithstanding, there’s no evidence that companies in the aggregate are cutting back on tech investment. A survey from Gartner predicts a 5% gain in tech spending in 2023.

Nobody likes inflation. But there may be a silver lining if the threat of rising costs forces companies to take digital steps that should have come years ago.

Our price index of digital goods and services includes:

Bundled access services
Cable and other subscription programming
Communications equipment mfg
Computer & peripheral equipment mfg
Data processing and related services
Electronic and mail-order shopping services
Internet access services
Internet publishing and web search portals
Semiconductor and other electronic component mfg
Software publishers
Video programming distribution
Wireless telecommunications carriers
Information technology (IT) technical support and consulting services (partial)