I believe it’s fair to predict that AI will have an impact on all facets of our society. Not by performing new tasks, but mainly by executing tasks that are already managed by humans, with high competence.
The substitution of humans with AIs may not be particularly riveting. However, when an AI undertakes a task previously done by a human, the nature of the task changes.
There are potential modifications across four key dimensions: Velocity, magnitude, range, and complexity. The issue with AIs engaging in stock trading isn’t that they surpass humans—it’s that they are quicker. Computers excel in chess and Go due to their application of more complex strategies than humans. Concerns arise with AI-managed social media accounts due to their operating capacity on a superhuman scale.
The transformation becomes intriguing when incremental changes lead to qualitative shifts. High-speed trading stands fundamentally distinct from conventional human trading. AIs have devised entirely new strategies in the realm of Go. Millions of AI-run social media accounts could fundamentally alter the landscape of propaganda.
It is these evolving dynamics and the impact of AI on democracy that I aim to explore.
To commence, I will outline some of the essential capabilities of AI. Firstly, it excels as a summarizer. Secondly, AI is adept at elucidating concepts and instructing with endless patience. Thirdly, and relatedly, AI holds persuasive abilities. Propaganda emerges as a byproduct of this. Fourthly, AI fundamentally operates as a predictive technology. Forecasts on whether taking a left or right turn will lead to a quicker route. Forecasts on the malignancy of a tumor could enhance medical diagnoses. Forecasts on the likelihood of the next word can aid in composing an email. Fifthly, AI can evaluate. Assessment necessitates external context and criteria. While AI may not excel in evaluation, it is steadily improving. Sixth, AI can make decisions. A decision comprises a prediction coupled with an evaluation. AI is already leveraged for various decision-making processes.
The translation of these competencies into tangible AI systems hinges significantly on the specifics. The extent to which AI will replicate or supplant human cognitive functions remains uncertain. Moreover, the timeline for such developments remains ambiguous. Confining environments pose fewer challenges. AIs already outperform humans in chess and Go. Unbounded environments pose more significant hurdles. Considerable obstacles persist in the realization of fully AI-driven automobiles. Technologist Jaron Lanier provides an interesting insight that AI thrives best when “human activities have been performed numerous times, albeit not in the same manner.”
In this discourse, I will adopt a largely upbeat stance on the technology. Detailed discussions concerning the operational mechanics of AI systems are not my primary focus. Much of what I discuss pertains to the future. It resides in the realm of science fiction, albeit not unrealistic science fiction.
My optimism will wane and realism will prevail when discussing the social repercussions of this technology. Once again, my focus lies beyond the direct substitution of humans with AI. I am more intrigued by the second-order effects resulting from these substitutions, particularly how the fundamental systems will evolve in response to shifts in velocity, magnitude, range, and complexity. My objective is to envision feasible scenarios to prepare for their eventuality.
Throughout this exploration, consider a few pivotal questions: Will this transformation disperse or consolidate power? Will it engender increased or diminished individual engagement with democracy? What prerequisites need to be fulfilled for individuals to place their trust in AI within this context? What potential risks might arise if malicious entities manipulate AI within this context? Lastly, what measures can we, as security technologists, undertake to assist?
My examination of democracy embraces a broad perspective, extending beyond representations and elections. Democracy is depicted as a mechanism for equitably distributing decisions across a populace. It functions as a conduit for translating personal preferences into group decisions, encompassing administrative verdicts as well.
With this objective in mind, I will delineate five distinct realms where AI will influence democracy: Political landscape, legislative processes, administrative proceedings, the judicial system, and ultimately, the citizenry.
I: AI-enabled politicians
It has been highlighted that AIs excel in persuasion. Politicians will capitalize on this trait. AI-driven propaganda is a widely discussed topic. Politicians are likely to leverage this avenue as well. Let’s explore the potential positive outcomes of this scenario.
Previously, candidates utilized books and speeches to establish connections with voters. In the future, candidates may also employ personalized chatbots to directly interact with voters on diverse topics. AI can also facilitate fundraising endeavors. The potency of personalized appeals is self-explanatory. AI can conduct surveys as well. There are intriguing studies on large language models assuming varied personas and answering queries from those perspectives. Unlike humans, AIs are available ceaselessly, capable of addressing thousands of queries without fatigue or boredom, thus warranting greater reliability. While AI won’t supplant surveys entirely, it can enhance them. AI can support human campaign managers by streamlining campaign operations, formulating talking points, conducting media outreach, and facilitating get-out-the-vote campaigns. These activities are already human-driven, hence, there’s no groundbreaking revelation.
The transformation primarily pertains to scale. AIs can engage with voters, conduct surveys, and fundraise on a scale that surpasses human capabilities—irrespective of the election size. They can also contribute to lobbying strategies. AIs possess the potential to devise more intricate campaign and political strategies than humans. Anticipate a competitive scenario as politicians begin harnessing these tools. The political landscape may witness an era of heightened sophistication in strategies, and it remains uncertain whether these tools will exhibit bias towards specific political ideologies.
Furthermore, future politicians will predominantly rely on AI assistance. This doesn’t imply that AI will supplant human politicians. Absent a substantial cultural revolution and significant legal amendments, such a transition seems unlikely. However, as AI starts exhibiting more human-like traits, our human politicians may appear more AI-driven. This evolution is likely to be embraced, as it aligns with the progression we have been traversing for a considerable period. Present-day major politicians merely serve as the outward facade of a multifaceted socio-technical framework. When the president delivers a speech, it’s well-understood that they didn’t author it. Similarly, when a legislator dispatches a campaign email, it is common knowledge that they did not draft it themselves—even if their signature adorns it. Greeting cards received from these figures were signed with an autopen, a practice ingrained in contemporary politics. In the future, the public will acknowledge that the majority of communications from their leaders would be AI-crafted. It will be widely accepted that AI tools aid in political and policy decision-making, campaign strategizing, and other operational facets. While these developments may not be inherently negative, they will undoubtedly redefine the essence of politics and political figures—akin to the transformations witnessed due to television and the internet.
II: AI-supported legislators
AIs demonstrate adeptness in summarization. This capability can be harnessed in understanding constituent feedback: summarizing correspondence, comments, and extracting insights from constituent inputs. Summarization can extend to public gatherings. Given the overwhelming scale of this domain, AI implementation can significantly impact the process. Beyond summarization, AI can identify compelling arguments or detect patterns in mass letter-writing campaigns. They can contribute to political negotiations as well.
AI can also contribute to legislative drafting. In November 2023, Porto Alegre, Brazil …Porto Alegre cemented its status as a trendsetter by being the initial city to implement a regulation crafted entirely by artificial intelligence, as reported byAP News. The law concerned the monitoring of water meters. A council member proposed the idea of utilizing ChatGPT, which successfully generated a full legislative proposal. Subsequently, this proposal was discreetly presented to the legislative body without disclosing its AI origins. Remarkably, the humans in the legislature approved the bill without making any alterations.
Legislation essentially consists of text collectively agreed upon by a government and adopted into law. Just like any other profession, policymakers may turn to AI for assistance in formulating and refining text. AI possesses the capability to interpret and grasp the true meaning behind human-written laws. Numerous laws involve recursive frameworks that make references to various sections and phrases from other laws. AI technologies are adept at deciphering and making sense of such intricacies.
This underscores the potential of AI in identifying and potentially exploiting legal loopholes. This concept is explored in depth in my recent publication, A Hacker’s Mind. Uncovering loopholes bears resemblance to discovering vulnerabilities in software. Additionally, there exists a notion known as “micro-legislation,” which pertains to the smallest actionable unit within a law that holds significance for individuals, ranging from a single word to a punctuation mark. AIs have the aptitude to embed micro-legislation into larger legislative texts. On a positive note, AI can facilitate the identification of unintended consequences resulting from policy alterations by simulating interactions between the modification and existing laws, alongside human behavior.
Furthermore, AI can produce more intricate legal drafts compared to those drafted by humans. Presently, laws predominantly adopt a general approach with specifics being refined by relevant government agencies. AI proposes a paradigm shift that allows lawmakers to introduce and vote on all aspects of legislative details. This shift alters the power dynamic between the legislative and executive branches of governance. While such a shift might not pose a significant issue when the executive and legislative arms are under the control of the same party, it becomes a focal point when these branches are held by disparate political factions. Concerns arise that AI might equip powerful factions with additional mechanisms to advance their interests.
AI’s potential extends to crafting laws that surpass human comprehension. Legislation generally falls into two categories: specific laws, like speed limits, and laws necessitating discretion, such as those addressing reckless driving behaviors. Consider a scenario where AI is trained on a vast repository of street camera footage to identify reckless driving patterns, surpassing human acumen in recognizing behaviors that often lead to accidents. With real-time access to ubiquitous camera feeds, AI could spot such behavior across various locales. Although the AI could specify criteria based on its findings, effectively becoming a black-box neural network, lawmakers could adopt legislation grounded on the AI’s assessments. Consequently, the legal framework would transcend human understanding, thereby introducing laws that are incomprehensible to humans. This scenario could manifest in diverse domains where legal definition hinges on judgment calls, such as market manipulation, medical malpractice, and false advertising, paving the way for AI-assisted legislative resolutions.
III: Revolutionizing Administrative Processes with AI
Generative AI already showcases proficiency in handling a multitude of administrative documentation tasks and is poised for further enhancement. Several key areas stand out where AI intervention is poised to make a substantial impact, including streamlining benefits administration to determine eligibilities, a task typically carried out by humans leading to backlogs due to resource constraints. Additionally, AI can revolutionize contract audits and negotiations by scrutinizing all government contracts with precision. Its application extends to aiding in complex regulatory compliance for government contractors navigating intricate rule sets.
Moreover, AI’s influence can be pivotal in diversifying negotiation scenarios across various domains by serving as a strategic consultant with the potential to foster intricate negotiation strategies. While AI’s role mirrors that of a human advisor, it could steer negotiations towards greater complexity by simultaneously considering thousands of variables. Imagine employing AI in international trade talks, where the AI suggests intricate strategies surpassing human comprehension. Would decision-makers unequivocally align with the AI’s recommendations, especially after a track record evaluation on accuracy?
Lastly, AI’s evolutionary trajectory poses the question of whether it could design superior institutional frameworks compared to current standards and whether these innovative designs would be implemented successfully.
IV: Transforming the Legal Landscape with AI
The integration of AI in legal proceedings encompasses a broad spectrum, encompassing legal advisory and judicial processes alongside ancillary events.
AIs can fulfill roles as legal practitioners. Initial endeavors involving AIs drafting legal briefs faced setbacks as highlighted in Reuters, however, subsequent refinements to AI systems have enhanced accuracy levels. Current chatbots have the capacity to cite sources correctly and minimize errors when writing legal documents, heralding a future where AIs significantly curtail legal service costs. Notably, AIs show promise in handling the mundane tasks traditionally managed by lawyers. Consequently, this transformation raises pertinent considerations.
Chief among the implications is the financial accessibility of legal guidance and representation, leveling the playing field for individuals lacking the means to afford legal counsel. An AI-powered public defender is poised to outperform an overburdened and less efficient human counterpart. However, assuming that a human-AI collaborative approach surpasses AI in efficacy, affluence could secure this combined service, relegating the less privileged to AI-only representation.
Additionally, the legal arena anticipates a surge in intricate legal arguments resulting from AI’s capability to scour legal archives for precedents in bolstering cases.
Furthermore, AI is primed to redefine the essence of lawsuits. Presently, lawsuits signify a potent social signal due to associated costs. If litigation costs plummet to insignificance, the essence of this signal could fade away, leading to a surge in lawsuits that could potentially overwhelm the judiciary.
An ancillary effect could involve reshaping the legal profession. Legal practice traditionally hinges on mentorship. However, the prevalence of AIs occupying apprenticeship positions poses a challenge for newly qualified attorneys seeking training opportunities. Consequently, the emergence of distinguished legal practitioners might encounter hurdles. The advent of AI-supported lawyers could either foster or diminish human lawyer engagement, an aspect that warrants closer examination.
AI plays a pivotal role in legal enforcement. While automated systems currently assist in law enforcement tasks, AI promises enhanced capabilities, such as automatic detection of tax evasion, fraud identification in government service applications, and real-time monitoring of surveillance footage to issue citations.
However, the proliferation of AI-led law enforcement introduces challenges such as false positives and potential biases. Contesting AI-generated accusations might prove cumbersome if courts exhibit a predisposition towards accepting AI determinations as infallible. Presently, courts encounter difficulties when contesting automated evidence like Breathalyzer results, underscoring complexities in challenging software-based findings. In addition, potential biases might permeate AI-led enforcement, potentially leading to inequitable outcomes compared to human-led initiatives.
Of paramount significance is the transformative impact of AI on our legal relationship. Individuals often commit driving infractions unconsciously. Widespread deployment of automated enforcement mechanisms could necessitate a behavioral shift in driving practices, reflecting a scenario where routine enforcement could lead to substantial alterations in societal norms. Resistance towards endorsing such a future is widespread, evident in reluctance to fund institutions like the IRS, despite their notable success rate in catching tax evaders and enforcing compliance. Valid apprehensions persist regarding the prospect of an automated enforcement ecosystem and its societal repercussions.
be applied fairly.
The application of regulations can be aided by AI. Rather than facing a scarcity of rules and regulations, the issue lies in the lack of time, resources, and determination to enforce them effectively. This results in many companies disregarding regulations without repercussions. AI has the potential to rectify this by separating the enforcement of rules from the required resources, making enforcement more scalable and efficient. For instance, by deploying cameras in all slaughterhouses nationwide to monitor animal welfare compliance, or utilizing AI-powered warehouse cameras to detect labor violations. Such initiatives could significantly shift the power dynamics between government entities and corporations, a change that may encounter strong resistance from corporate entities.
AI systems can offer expert insights in legal proceedings. Consider an AI system trained on a vast dataset of traffic accidents, including video evidence, car telemetry, and past court rulings. This AI could reconstruct accidents, assign fault, and present this information to the court. In cases where human experts are scarce, AI can prove to be more competent due to its extensive experience.
Furthermore, AI can undertake judicial responsibilities, evaluating evidence and making decisions, albeit not in conventional court settings in the immediate future. This automation could address the scarcity of adjudicators in various governmental sectors, potentially enabling prompt justice for all. Initial levels of adjudication could be handled by AI, with humans overseeing subsequent appeals. The integration of AI into contract disputes could revolutionize the arbitration process, making it more cost-effective. Consequently, this might lead to a surge in dispute resolution cases.
Envision a scenario where resolving disputes is both economical and swift. If two business partners face a disagreement, they could obtain a verdict within minutes and seek resolution multiple times a day. Will this automation diminish our capacity to resolve conflicts independently, or will it facilitate stronger partnerships built on trust?
V: AI-supported citizens
AI offers substantial assistance in deciphering complex political matters. This ranges from utilizing partisan and nonpartisan chatbots to providing political analysis and commentary at various levels, including local elections that typically receive insufficient media coverage. Ongoing research explores AI’s roles as a moderator, facilitator, and consensus builder. While human moderators are currently more effective, the scarcity of human moderators necessitates the advancement of AI technologies, which are poised to enhance over time by providing scalable moderation services to every decision-making group, online community, or local government assembly.
AI can function as a vigilant overseer of governmental procedures. In regions where local government proceedings occur behind closed doors due to the absence of media coverage, AI could illuminate these processes by generating summaries and highlighting any shifts in stances.
AIs can assist individuals in navigating bureaucratic systems, aiding in form submissions, service applications, and challenging bureaucratic decisions. This support proves particularly beneficial for marginalized individuals who struggle to navigate these intricate structures. Once again, this is a realm where there is a shortage of proficient human resources. While the proposition sounds promising, not everyone may favor this assistance, as administrative hurdles could be intentionally crafted.
Finally, AI has the potential to render traditional politicians redundant. Although this notion appears futuristic, initial studies demonstrate AI’s capability to deduce individual political preferences. An AI personal assistant, well-versed in and perpetually attuned to your political inclinations, could offer guidance on matters to support and recommend candidates to endorse. It might even be entrusted to vote on your behalf or serve as your personal representative.
This concept leads to intriguing possibilities. In our representative democracy, elected officials are tasked with translating our collective preferences into policies. However, this process has inherent flaws as representatives are essential due to public disinterest in politics and the limitations of accommodating all voices in debates. By selecting a single individual to enact laws on behalf of numerous constituents, democracy encounters what has been termed a “lossy bottleneck”. AI could revolutionize this system, enabling personal AIs to engage in policy dialogues collectively, reaching a consensus on legislative matters.
In the immediate future, AIs may contribute to an increase in ballot initiatives, potentially surpassing the current count of five or six to hundreds, provided that AIs can reliably guide voters. The repercussions of this shift remain uncertain, as there is a concern that individuals might become politically disengaged if AI takes control. Nonetheless, it could lead to the enactment of more legislation aligned with the majority’s interests.
What lies ahead with AI?
This outlines potential transformations where incremental changes could usher in fundamental shifts. Enhanced AI capabilities in legislation might lead to intricate laws, altering the equilibrium of power between executive and legislative branches. The widespread use of AI in legal processes could make litigation accessible to all, causing a surge in legal cases. The rapid adjudication facilitated by AI could expedite the resolution of contract disputes, transforming the landscape of settlements. AI’s extensive enforcement reach could render certain laws impervious to evasion, prompting the affluent and influential to reassess their approach.
This transformation appears inevitable, with the exact timeline remaining ambiguous as technology progresses in these trajectories.
All these applications necessitate some form of security measures. Can we ensure confidentiality, integrity, and availability where required? AIs function as computers and inherit the security vulnerabilities of conventional systems, in addition to new risks arising from the training, deployment, and utilization of AI. As with any security aspect, the effectiveness hinges on specific details.
Primarily, the incentives play a pivotal role. Users may seek both security and accuracy in AI utilization, while others might aim to exploit the system, exemplified by prompt injection attacks. Often, the AI’s owners differ from its users, a dynamic observed in search engines and social media platforms where surveillance and advertising form the prevalent business model. Consequently, the users’ objectives might conflict with societal expectations.
Secondly, the risks involved carry significant weight. The consequences of inaccuracies vary depending on the application. While a chatbot suggesting an absurd policy by a candidate can be rectified easily, an error in AI assistance for immigration paperwork could result in deportation. Evaluating the rate of AI errors against human errors is imperative to assess the systems’ reliability and integrity.
errors—and also understand that AI errors are perceived differently compared to human errors. There are also various kinds of errors: incorrect positives versus incorrect negatives. Additionally, AI systems can make distinct types of errors than humans do—and that is crucial. In each situation, the systems must have the capability to rectify errors, particularly within the framework of democracy.
Several of the uses are in competitive environments. If two nations are utilizing AI to aid in trade discussions, they will both attempt to breach each other’s AIs. This will encompass assaults on the AI models as well as conventional attacks on the computers and networks that are operating the AIs. They will try to undermine, spy on, or interrupt the other’s AI.
Certain AI applications will require operation in protected settings. Extensive language models function ideally when they have access to everything, for the sake of training. This contrasts with traditional classification rules concerning compartmentalization.
Moreover, authority is significant. AI is a technology that inherently amplifies the power of the individuals who employ it, but not uniformly across users or applications. Can we devise systems that lessen power disparities rather than exacerbate them? Reflect on the privacy versus surveillance discourse in the realm of AI.
Similarly, fairness is crucial. Human willpower is crucial.
Lastly, credibility is essential. Whether or not to rely on an AI is less about the AI itself and more about the application. Several of these AI applications are personal. Several of these applications are communal. Whether factors such as “impartiality” are crucial depends on this. Furthermore, there exist numerous conflicting interpretations of fairness that hinge on the specifics of the system and the application. The same holds for transparency. The necessity for it is contingent on the application and the motivations. Democratic applications are inclined to necessitate more transparency than corporate ones and likely AI models that are not possessed and managed by global tech monopolies.
All of these safety concerns are broader than AI or democracy. Analogous to all of our safety background, implementing it to these new systems will necessitate some new deliberation.
AI will end up being one of the most critical innovations for humanity. That is probably correct. What we are unsure of is whether this is the juncture where we are innovating it or if the systems of today are just more overblown technologies. Nonetheless, these are safety dialogues that we will ultimately need to engage in.
AI essentially serves as a power-amplifying technology. We must guarantee that it allocates power and does not further centralize it.
AI is asserting itself in democracies. Whether the alterations are an overall beneficial or detrimental change depends on us. Let us contribute to swaying matters towards the positive.
This piece is modified from a keynote address delivered at the RSA Conference in San Francisco on May 7, 2024. It originally emerged in Cyberscoop.
About Author
