ChatGPT has only been available for nine months, but we're still learning about how it will impact our lives, careers, and self-government.
Many public discussions lack imagination when it comes down to the threat AI poses to our democracy. We have dealt with doctored videos and audio for decades, so we are familiar with the dangers of such campaigns. We are on the lookout to find foreign governments who spread misinformation after the traumatizing 2016 US Presidential election. We're also worried that AI-generated opinion will overshadow the political preferences expressed by real people, because we have seen "astroturfing," the use of fake accounts on the internet to create the illusion of support for policies, grow in recent decades.
These threats are salient and seem urgent. We can see the effects of these threats because we know what to look out for.
Truthfully, the future is going to be more exciting. Even the most spectacular potential impact of AI on politics will not be bad. There are some pretty straight lines that we can draw between the current capabilities and outcomes of AI tools, which, by current standards of public understanding, appear truly startling.
In this context, we have proposed six milestones which will usher in a new age of democratic politics powered by AI. All seem achievable, albeit not today with the technology and AI adoption levels we have. But they may be in the near term.
What is a political AI landmark?
Good benchmarks are those that represent significant outcomes with real-world implications. They must be realistic and achievable within a reasonable time frame. They should also be observable, so we can see when they have been accomplished.
The observability test will most likely show that AI's influence on elections is not as real as people think. Elections are complex and the risk of manipulation by robotically promoting a candidate or party is real. As the debate over how Donald Trump was elected president continues, we are unlikely to be in a position to attribute any unexpected electoral result to AI.
Can an AI candidate be elected in the future? In the world of speculative fiction, from The Twilight Zone to Black Mirror, there is growing interest in the possibility of an AI or technologically assisted, otherwise-not-traditionally-eligible candidate winning an election. It is possible that an AI candidate can mimic the media presence and voice of a politician in an age where deepfaked videos are used to misrepresent human candidates' views and actions, and humans can be represented by AI avatars and even robots. Virtual politicians received votes at national elections. For example, in Russia in 2017. This is not plausible. AI will likely be used to support and automate voting and the legal system, but it is still a long way off.
The Next Political Milestones for AI
Let's begin with some achievements that are on the verge of becoming reality. These achievements are within the scope of current AI technologies, and the foundations have already been laid.
Milestone 1: When a legislator or agency accepts a testimony, or comment, generated by an AI and submitted in its name.
We've seen AI draft legislation, but it was under the guidance of humans and presented by human legislators. Following the introduction of some early AI-written bills in Massachusetts and US House of Representatives', many major legislative organizations have held "first AI written bill" or "used ChatGPT for committee remarks" events.
These speeches and bills are often criticized for being a stunt and not serious. These speeches are either short and have little substance or heavily edited by human legislators.
This will mark a significant milestone when AI is used to draft a whole comment or testimony for drafting legislation. AI assistive technologies will benefit a significant portion of future writing. To avoid trivializing this landmark, we must add the second clause "submitted in the name of AI."
This benchmark would be significant if it was submitted under the AI's name, i.e. the AI being accepted by a governing authority as a legitimate voice in the public debate. This one will not take long, regardless of how passionate the public is about AI. The New York Times published a letter titled ChatGPT in response to an opinion article we wrote, and legislators have already turned to AI for high-profile opening remarks during committee hearings.
Milestone # 2 : Adoption of first novel legislative amendments to a Bill written by AI.
Microlegislation is a way for AI-generated policy to be law immediately, even if it does not require a testimony. It involves tweaking existing laws or bills to suit a particular purpose. This is the natural place to start for AI, because it involves small changes that are guided by a clearly defined purpose.
Microlegislation, by design, is usually implemented in a covert manner. Microlegislation may be submitted anonymously in a flurry of amendments, to conceal its intended beneficiary. Microlegislation is often bad for society and can be exploited by generative AI.
Milestone 3: AI generated political messaging beats out campaign consultant recommendations during poll testing.
AI's near-term impact on politics will be largely hidden. As with everyone else, pollsters and political campaigners will use AI to assist them in their work. Campaigners are already using AI-generated images for social media content, and pollsters are simulating results with AI-generated respondents.
Next, AI will develop political messages. The message testing survey is a mainstay in the toolbox of today's campaigners. It involves writing down a few alternative formulations and testing them with audiences to determine which ones will garner more attention and generate a positive response. As an experienced pollster can predict effective messaging strategies based on past campaigns and impressions of the current state of the public discussion, an AI can do the same if it is trained to comb through reams and reams on public discourse, political reporting, and campaign rhetoric.
AI and Democracy: More Future Achievements
Let's explore some of the truly revolutionary possibilities that are now within reach. These concepts, which seemed impossible a year earlier, are now more conceivable using current or near future technologies.
Milestone #4: AI builds a party that attracts human candidates and wins elections.
It is possible that an AI could be able create a political party. A political platform could be created by an AI to appeal to a certain section of the population. It could then register as a party, either independently or with the help of a human (a hired consultant, lawyer or political consultant). It could gather signatures in order to gain a position on the ballot and recruit human candidates to run under its banner.
The Danish Synthetic Party campaign in 2022 has taken a big step towards this goal. A Danish artist collective created an AI chatbot that interacted with members of their Discord community, examining political ideologies in conversation and on the analysis of historic party platforms. This was done with older generations of general-purpose AI and not the current systems such as ChatGPT. The party did not receive enough signatures for a place on the ballot and, therefore, was denied parliamentary representation.
Future AI-led initiatives may be successful. A generative AI that has skills on par or above today's top technologies could create a set or policy positions to gain support from a certain demographic or even a consensus platform capable of gaining broad support. In a multiparty system akin to that of Europe, we could imagine a party with an AI at its heart and a powerful news hook. This would attract attention and vote.
Milestone #5 AI generates profit autonomously and contributes to political campaigns
Next, let's look at the most important capability in modern politics: Fundraising. A realpolitik definition for a politician might be "an entity that can direct contributions to a fund." AI could potentially do this.
An AI, like a human being, could generate contributions in many different ways to a campaign. It could invest a small amount of money from the human who controls the AI to generate a profit. It could create a revenue-generating business. Auto-hustling is a growing area of interest and experimentation: AI agents which grow businesses or generate profit autonomously. ChatGPT generated businesses have not yet taken the world by a storm. However, they are in the same spirit of the algorithmic agents that power modern high-speed trading.
Like most political entrepreneurs AI can generate political messages to persuade humans to invest their money in a campaign or cause. AI would need humans to be in the loop and to register their activities with the government.
Milestone #6 AI achieves coordinated policy outcomes across multiple jurisdictions.
We come to our most important impact: the achievement of outcomes in public policies. AI may not have any preferences or desires, but it can be programmed to achieve a certain goal. For example, lowering taxes, or removing a regulation on the market.
AI uses many of the tools that humans do to accomplish these goals. It can advocate by formulating messages and promoting ideas via digital channels such as social media posts and video. It can lobby to influence key policymakers and even write legislation. It may spend, see milestone #5.
This milestone is based on the "multiple jurisdictions". One law can be attributed to a variety of factors, including a charismatic leader, a political campaign, or a change in circumstance. It is easier to demonstrate the influence of a single actor, like an AI, if they are successful in multiple places. AI's digital scalability gives it an advantage when it comes to achieving coordinated outcomes.
How will we know the future?
Most of these milestones are difficult to observe. Will we recognize them when we see them? It's possible that the first campaign consultant to lose to AI in terms of ideas will not want to share this information. The campaign will not either. In terms of fundraising, we find it difficult to identify the humans who make the contributions that are known as "dark-money" and which control much of the modern political finance. Will we be able to tell if the future dominant force for fundraising for political action groups is an AI or not?
Some of these milestones will be observed indirectly. We may see politicians shifting their campaign dollars to AI-based consultancies at some point.
We need not be afraid of these technological milestones, even though they are often unsettling. It is not neutral to support a new political platform, even if it wins the public's approval. This could lead to positive or negative policy outcomes. A successful policy program can be good or bad for one group of constituencies.
The six milestones we have outlined are some of the most meaningful and viable future interactions between AI, democracy and technology. However, they are not the only ones to be considered. Our AI-driven future will be more than just fake campaign ads and letter-writing campaigns. We need to be more creative and vigilant about the future.
————————————————————————————————————————————————————————————
By: Bruce Schneier, Nathan E. Sanders
Title: Six ways that AI could change politics
Sourced From: www.technologyreview.com/2023/07/28/1076756/six-ways-that-ai-could-change-politics/
Published Date: Fri, 28 Jul 2023 08:30:00 +0000
Leave a Reply