The Impact of AI on Internet Freedom


The latest report from global watchdog Freedom House reveals a concerning trend: for the 13th consecutive year, internet freedom has declined. This is where the increasing power of artificial intelligence (AI) tools comes into play, presenting both positive and negative implications for internet freedom. The outcome largely depends on the manner in which governments and individuals utilize AI.


While the degradation of internet freedom primarily stems from existing technologies, AI has the potential to exacerbate the situation. Freedom House's investigation encompassed 70 countries, and an alarming 29 witnessed worsening conditions for civil liberties compared to the previous year. Shockingly, in a record 55 nations, individuals faced legal repercussions for their online expressions. Approximately 4.9 billion people have internet access, with 78 percent residing in countries where individuals have been arrested or imprisoned due to their online activities.


Regarding censorship, AI does not introduce new forms of repression; instead, it simplifies existing methods. Freedom House found that at least 22 countries required social media companies to utilize automated systems for content moderation to comply with severe laws. For instance, India incorporates automated systems into its information technology framework to identify and remove categories of content deemed to undermine public order, decency, or morality. These behind-the-scenes tools also serve to conceal the state's involvement in suppressing dissent, enabling regimes to evade the risks associated with imposing restrictions and potentially provoking public outrage.


The perils associated with information warfare, a preferred tactic of insecure autocrats, become even more pronounced with the advent of increasingly sophisticated AI. While propaganda campaigns predate the development of semiconductors, AI's ingenuity amplifies the creation of persuasive falsehoods. States are already capitalizing on these emerging capabilities: Freedom House uncovered instances in at least 16 countries where AI tools were employed to generate deceptive images, text, or audio. Venezuela's government has disseminated synthetic videos featuring fabricated news anchors from an international English-language channel, while an Israel-based group, Team Jorge, has utilized text generated from keywords to propagate misinformation through an army of false accounts. The objective is to cast doubt on troubling allegations made against the director of Mexico's criminal investigations unit.


One particularly noteworthy development in machine learning this year exemplifies how these technologies can either empower or disempower individuals, depending on who holds the reins. ChatGPT and other large-scale generative language models, under favorable circumstances, enable users in repressive states to circumvent restrictions on the free flow of information. The training data for these mainstream systems is derived not from the strictly controlled speech environments of authoritarian regimes, but rather from the global internet.


However, the censorious governments are well aware of this. Vietnam, for instance, cautioned its citizens against using ChatGPT, claiming that it defames the Communist Party. Similarly, Chinese regulators have prohibited the inclusion of the model in services offered by the country's major technology conglomerates or via any app store. China is also manipulating its domestically-created models, mandating adherence to "truth, accuracy, objectivity, and diversity" as defined by the Chinese Communist Party. They are also tightening control over the output generated by these models. If one were to ask Baidu's Ernie Bot about the events at Tiananmen Square, it would refuse to engage in the conversation.


In conclusion, AI's potential impact on internet freedom is a dual-edged sword. It can either facilitate or impede freedom depending on its usage. The concerning trends witnessed in the latest Freedom on the Net report should serve as a call to action for ensuring that AI is harnessed in a manner that promotes and safeguards internet freedom.


The application of the aforementioned principle extends to various other types of artificial intelligence. Consider, for instance, the automated systems designed for content moderation. Rather than limiting expression, these tools have the potential to be used in ways that strengthen it. They can be employed to counter disinformation campaigns and promote the dissemination of accurate information regarding human rights abuses. In order to ensure more favorable outcomes, it is imperative for countries like the United States to develop domestic frameworks that prioritize civil liberties when creating AI tools. Such frameworks should include transparency requirements regarding the data on which chatbots are trained, as well as safeguards against discrimination and excessive surveillance. As a promising step, the White House is anticipated to present its blueprint for an AI bill of rights, but it would be even more advantageous if Congress were to enact legislation in this regard.


The United States, along with its like-minded allies, should advocate for this approach on a global scale. This involves integrating AI and the internet into democracy assistance initiatives, as well as utilizing collective influence in international forums. China and Russia have successfully influenced the United Nations' convention on cybercrime in a manner that could be deemed excessive. It is imperative that the United States and its allies prevent the occurrence of similar overreach in the domain of AI.


During the early days of the internet, leaders in democracies generally assumed that it would enhance freedom worldwide. However, they failed to take sufficient action to protect and promote this notion. The silver lining amidst the challenges associated with AI is that now, with the emergence of this new technology, there exists an opportunity to rectify past mistakes and ensure that the right course of action is undertaken.