The following essay is reprinted with permission 
The Conversation is an online publication covering the latest research. This has been modified by a Scientific American writer and may differ. original.
Can organizations use artificial intelligence language models such as ChatGPT to persuade voters to take specific actions?
Senator Josh Hawley asked OpenAI CEO Sam Altman this question during the US Senate hearing on artificial intelligence on May 16, 2023. Altman replied that he was certainly concerned that some people could use language models to manipulate, persuade, and engage in one-on-one interactions with voters.
Here’s a scenario Altman might have envisioned/had in mind: Imagine that soon a political technologist would develop a machine called Kroger, a political campaign in a box. Clogger just relentlessly pursues one goal. It’s all about maximizing the chances of your candidate (the campaign to buy Clogger Inc.’s services) winning the election.
Platforms like Facebook, Twitter, and YouTube use AI to help users spend more time on their sites, but Clogger’s AI may have another purpose: to change people’s voting behavior.
How Clogger works
As political scientists and legal scholars who study the intersection of technology and democracy, we believe that the likes of Kroger are using automation to replace the behavioral manipulation and microtargeting techniques that have been used in political campaigns since the early 2000s. We believe we can dramatically increase the scale and potential impact. . Just as advertisers today use your browsing and social media history to target commercial and political ads individually, Kroger will target you and hundreds of millions of other voters individually. will pay attention.
It offers three advances over current state-of-the-art algorithmic motion manipulations. First, the language model generates messages (including text, social media, email, and possibly images and videos) that are personalized to the user. While advertisers place relatively few ads strategically, language models such as ChatGPT generate millions of unique messages for individuals during campaigns and generate millions of ads for others. You can generate messages.
Clogger then uses a technique called reinforcement learning to generate messages that are increasingly likely to change votes. Reinforcement learning is a trial-and-error approach to machine learning in which a computer performs actions and gets feedback on which work is more effective in order to learn how to achieve a goal. Machines that can play Go, chess, and many other video games better than humans use reinforcement learning.
And finally, over the course of a campaign, Clogger’s message may evolve in light of user reactions to previous dispatches and what it has learned about changing other people’s minds. Clogger keeps dynamic “conversations” over time with you and millions of other people. Clogger messages are similar to advertisements that follow users across various websites and social media.
Nature of AI
Three more features (or bugs) are worth noting.
First, the messages Clogger sends may or may not be political. The machine’s sole goal is to maximize vote share, and it is likely to devise strategies to achieve this goal that no human campaigner would have thought of.
One possibility is to send voters information about their non-political passion for sports and entertainment in order to bury the political messages that opposition voters receive. Another possibility is to send offensive messages (such as ads about incontinence) to match the other party’s message. And the other is manipulating voters’ social media groups into thinking their family, neighbors and friends support the candidate.
Second, Kroger ignores the truth. Indeed, there is no way of knowing what is true or false. The purpose of this machine is not to provide accurate information, but to modify votes, so language model “hallucinations” are not an issue.
Finally, since this is a black box type of artificial intelligence, people have no way of knowing what strategy it is using.
Clogocracy
If the Republican presidential campaign deploys Kroger in 2024, the Democrats will likely be forced to do the same with similar machines. Please call me Dogger. If campaign leaders thought these machines were effective, the presidential election would be a Krogger vs. Dogger game, with the winner being the client of the more effective machine.
The content that won that day would not have come from a candidate or party, but from an AI that had no political ideology of its own and was solely focused on winning. Machines, not humans, would have won the election in this very important sense. Elections would no longer be democratic, even if all the normal activities of a democracy – speeches, advertisements, messages, voting, counting of votes – took place.
A president elected by AI may choose one of two methods. He or she can use the electoral role to pursue Republican or Democratic policy. But since the party’s thoughts may have had little to do with why people voted that way (Kroger and Dogger don’t care about policy views), the president’s actions aren’t necessarily the voters’ will. does not necessarily reflect Voters would not be free to choose political leaders and policies, and would be manipulated by AI.
Another route is to pursue the messages, actions, and policies that machines predict will maximize a president’s chances of re-election. On this road, the president would have no specific program or policy other than to maintain power. The president’s actions, guided by Kroger, are most likely to manipulate voters rather than serve their true interests or the president’s own ideology.
Avoiding clogocracy
If candidates, campaigns, and consultants all allowed the use of such political AI, it would be possible to avoid election manipulation by AI. We believe that is unlikely. If politically effective black boxes were developed, competitive pressure would make their use almost irresistible. In fact, political consultants may consider using these tools according to their professional responsibilities to help candidates win. And when one candidate uses such effective means, the opposition can hardly be expected to resist by unilaterally disarming them.
Better privacy protection would help. Kloggers rely on access to vast amounts of personal data to target, persuade or manipulate individuals, craft customized messages, and track and retarget them during campaigns. Any denial of that information by corporations or policy makers will make the machine less effective.
Another solution lies in the Electoral Commission. They may ban or heavily regulate these machines. There is intense debate as to whether such “replicant” speech, even if political, can be regulated. Many leading scholars argue that it cannot, due to the US’ extreme free speech tradition.
But there is no reason to automatically extend First Amendment protections to the products of these machines. It may be natural for the nation to choose to give rights to machines, but it should be a decision based on today’s challenges, and the 1789 observations of James Madison intended to apply to AI. It is not a misplaced assumption that something is.
European Union regulators are moving in this direction. Policy makers amended the European Parliament’s draft artificial intelligence law to designate “AI systems that influence voters in election campaigns” as “high-risk” and subject to regulatory oversight.
One of the smaller but constitutionally safer measures, already partially adopted by European internet regulators and California, is to ban bots from impersonating humans. For example, regulations may require a campaign message to include a disclaimer if the content included in the campaign message was generated by a machine rather than a human.
This is similar to the ad disclaimer requirement “Paid by Sam Jones of a Congressional Committee”, but amended to reflect the origin of AI and states that “This AI-generated ad was paid by a Congressional Committee was paid by Sam Jones of A stronger version says, “This AI-generated message is being sent to you by the Sam Jones Congressional Committee. This is predicted by Kroger to increase your chances of voting for Sam Jones by 0.0002% by doing so.” At the very least, we think voters have a right to know that a bot is talking to them, and why.
The potential of Krogger-like systems shows that the path to human collective incapacitation may not require superhuman general artificial intelligence. It may just require dedicated campaigners and consultants with powerful new tools that can effectively push many buttons for millions of people.
This is an opinion and analysis article and the views expressed by the author are not necessarily those of the author. Scientific American.
This article originally appeared in The Conversation. Please read the original article.
