Published in The Journal of Robotics, Artificial Intelligence & Law November-December 2018
I am on record in a number of publications as interpreting the First Amendment literally when it comes to freedom of speech for artificial intelligence ("AI").1 The First Amendment does not refer to human beings or natural persons. The First Amendment states that the federal, state, and local governments "shall make no law . . . abridging the freedom of speech."2 A recent bill in the California legislature, Senate Bill No. 1001 (the "Bot Bill"), is testing that, though, as it would make it unlawful for any person to use a bot—defined as "an automated online account on an online platform that is designed to mimic or behave like the account of a person"—to "communicate or interact with another person in California online," subject to certain provisions.3 In order to avoid unlawful use of a bot, the owner of the bot can disclose that the bot is, in fact, a bot.4
The constitutionality of the Bot Bill has been called into question by a number of groups, some of which believe, like me, that the First Amendment protects AI speech and autonomous speech from devices and programs. For example, in a letter dated May 21, 2018, the Electronic Frontier Foundation ("EFF") alleged that the Bot Bill would, among other things, unconstitutionally "unmask" the humans who program and create the bots.5 The EFF appears to question the legitimacy of any law that would restrain internet speech, although it recognizes the need to address "harmful and ill-intentioned bots, such as the Russian bots that interfered with the 2016 U.S. elections or spambots used for fraud or commercial gain."6 While I agree that there are some flaws in versions of the Bot Bill that could make it unconstitutional, I disagree with the notion that any bill that seeks to regulate bot speech on the internet is unconstitutional on its face because of its implications for the freedom of speech of the human programmers behind the bots. There are some specific revisions that would make the Bot Bill a constitutional bill, a better bill, and a bill that we need.7
Anonymous Speech Under the First Amendment
The exact standard used to review government efforts to limit anonymous speech varies depending on the context,8 but there is little doubt that the First Amendment was intended to protect anonymous speech. The Federalist Papers were written anonymously by John Jay, Alexander Hamilton, and James Madison, the author of the First Amendment. The U.S. Supreme Court has recognized the importance of anonymity to free speech: "Anonymity is a shield from the tyranny of the majority. . . . It thus exemplifies the purpose behind the Bill of Rights and of the First Amendment in particular: to protect unpopular individuals from retaliation . . . at the hand of an intolerant society."9 As I have argued, the First Amendment protects all speech, from human and nonhuman speakers alike, so the right to anonymous speech extends to AI and bots.
The Bot Bill potentially runs afoul of the First Amendment's protection of anonymity by requiring that bots identify themselves as bots. If bots are subject to the First Amendment, this mandatory identification is akin to requiring natural person speakers to identify themselves. The most likely standards to be applied to the Bot Bill's prohibition of bot anonymity are intermediate scrutiny or exacting scrutiny. Courts use intermediate scrutiny when reviewing content-neutral laws, i.e., laws that govern the "time, place, and manner" of speech rather than the contents of that speech. Intermediate scrutiny uses a five-part test:
- Is the challenged law or regulation within the constitutional power of government?
- Does the law or regulation further an important or substantial governmental interest?
- Is the governmental interest unrelated to the suppression of free expression?
- Is the restriction narrowly tailored to address the governmental interest?10
- Are there other ample open opportunities of communication?11
If the subject law does not satisfy these five questions, the court will find that the law violates the First Amendment.
When looking at laws that seek to regulate political speech, courts use exacting scrutiny, upholding the law in question only if it is narrowly tailored to serve an overriding state interest.12 If the law is not narrowly tailored to serve an overriding state interest, the court will find that the law violates the First Amendment.
Does the Bot Bill Satisfy the Courts' First Amendment Standards?
When the EFF issued its letter, it was responding to the original version of the Bot Bill, which would have made it unlawful "for any person to use a bot to communicate or interact with natural persons in California online, with the intention of misleading a natural person about its artificial identity."13 There was no language that would have distinguished between commercial bots, political bots, bots that are intended to be humorous, bots that are created by artists as part of an installation or project, etc.14 As a matter of First Amendment review, the fact that the original version of the Bot Bill broadly prohibited anonymity among bots, rather than try to distinguish between bots for particular purposes, likely would have caused the Bot Bill to fail a court challenge. It is doubtful that a court would find such a broad approach to be narrowly tailored as required in both intermediate scrutiny and exacting scrutiny.
However, since the EFF issued its letter, the California legislature has amended the Bot Bill. As currently drafted, it would be unlawful "for any person to use a bot to communicate or interact with another person in California online, with the intent to mislead the other person about its artificial identity for the purpose of knowingly deceiving the person about the content of the communication in order to incentivize a purchase or sale of goods or services in a commercial transaction or to influence a vote in an election."15 The new language moves the Bot Bill away from being a content-neutral law, as it now identifies the type of content it applies to; that appears to eliminate intermediate scrutiny in favor of exacting scrutiny. But the new language also makes the legislation more narrowly tailored.16
Does informing consumers (in the commercial context) and voters (in the political context) that the online party they are interacting with is a bot qualify as an overriding government interest? All signs point to yes. The Supreme Court has ruled that requiring that candidates or political action committees selfidentify themselves in political ads does not on its face violate the First Amendment, although the Court has noted that as-applied challenges are available if a group can show a "reasonable probability" that disclosing its contributors' names would "subject them to threats, harassment, or reprisals from either Government officials or private parties."17 That type of disclosure can be justified by a government interest in providing the electorate with information about election-related spending sources.18
Disclosing bot status shares a similar government interest. As the 2016 presidential election and the subsequent federal investigations indicate, there is an overriding government interest in ensuring reliable and accurate elections by preventing foreign nations and bad actors from using bots to illicitly influence votes in an election. Requiring that bots that attempt to influence votes in an election identify themselves as bots is a narrowly tailored approach to addressing that government interest. As the Supreme Court has noted when upholding disclaimer and disclosure requirements in election advertising, "they 'impose no ceiling on campaign related activities . . .' or 'prevent anyone from speaking.'"19
The Bot Bill's Effect on Bot Anonymity Versus Human Anonymity
The EFF's objection regarding anonymity was not primarily concerned with the anonymity of bots; it was concerned with the human programmers' anonymity. With regard to commercial and political speech, this appears to be an irrelevant concern. Anything a human programmer might want a bot to say politically or commercially on an online platform, the human programmer can tweet or post him- or herself through an anonymous account or username. The Bot Bill does not attempt to limit those communications. While I agree that the EFF's concern that the Bot Bill's original language would limit how artists and other similar speakers use bots to experiment with the bot form itself, the current version of the bill carves those speakers out from its restrictions.
Improving the Bot Bill
The amended Bot Bill, however, removed key language that held online platforms—Twitter, Facebook, etc.—accountable for the bots operating on their systems. The proposed Section 17942 stated:
(a) An online platform shall enable users to identify and report bots that the user suspects of violating [the requirement that bots identify themselves].
(b) (1) After receiving notice of a bot pursuant to subdivision (a), an online platform shall expeditiously investigate and determine whether or not to disclose that the bot is not a natural person or remove the bot.
(2) An online platform's investigation and response shall be expeditious if it occurs within 72 hours of receipt of the notice.
(c) Upon request of the Attorney General, an online platform shall provide reports detailing notices received pursuant to subdivision (b) and actions taken in response.
It was a mistake to remove these provisions from the Bot Bill. Online platforms are the parties in the best position to remove identified bots and minimize their influence on human users, which is the ultimate goal of the legislation. Without requiring online platforms to have a stake in the successful enforcement of the Bot Bill, the California government will be left to its own devices and resources, which are already thinly spread. Removing all responsibility from online platforms significantly reduces the bill's chances to be successful.20
I can understand potential objections from online platforms. A 72-hour turnaround time to remove bots is tight, and the requirements of this section would likely result in the accounts of some real people being removed. However, the solution is not to remove the requirements; the solution is to revise the requirements. Seventy-two hours is not enough time? We can make it 96, 120, etc. Will too many people be removed from online platforms? We can change "remove" to "suspend," permitting any natural person whose account is suspended to contact the online platform and correct the issue.21
Informing people when they are talking to AI, autonomous systems, or bots has been identified by a number of people as an important element of adapting to a world with widespread AI.22
The Bot Bill represents a positive development in the regulation of AI for that reason. But to give the bill its best chance to result in the widespread self-identification of bots as bots, the legislation should put part of the cost of enforcing that requirement on the online platforms where bots have proliferated.
1 John Frank Weaver, "Applying the First Amendment to Artificial Intelligence," TerraLex Connections, July 2018, https://www.terralex.org/publication/pba5b85ea27; John Frank Weaver, "Why Robots Deserve Free Speech Rights," Slate, January 16, 2018, https://slate.com/technology/2018/01/robots-deserve-a-first-amendment-right-to-free-speech.html.
2 U.S. Const. amend I. Although the text of the First Amendment refers only to Congress, the Fourteenth Amendment has been ruled to extend that requirement to the states and their political subdivisions.
3 Cal. SB 1001, §1. The current version, showing revisions to the original version submitted is available at https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml?bill_id=201720180SB1001.
5 Jamie Williams, "EFF Letter Opposing California Bot Disclosure Bill, SB 1001-First Amendment Concern," Electronic Frontier Foundation, May 21, 2018, https://www.eff.org/document/eff-letter-opposing-californiabot-disclosure-bill-sb-1001-first-amendment-concerns. EFF raises other concerns regarding the Bot Bill under the First Amendment, including compelling speech and censoring legitimate speech, that this article does not address, except as otherwise noted.
7 Please note that this article, for the most part, does not address Senate Bill 3127, the Bot Disclosure and Accountability Act of 2018 (the "Federal Bot Bill"), introduced by Democratic Senator Dianne Feinstein, because California's Bot Bill has a greater chance of becoming law. Having said that, there are numerous similarities and areas of overlap between the two bills, and I speak to those as relevant in the notes.
8 Courts apply exacting scrutiny to laws that burden political speech, including laws that prevent anonymity. McIntyre v. Ohio Elections Commission, 514 U.S. 334 (1995). However, courts are willing to accept some regulations prohibiting anonymous commercial speech based on the need for consumer protection. See, e.g., Lorillard Tobacco Co. v. Reilly, 533 U.S. 525 (2001) (noting that commercial speech may be regulated to ensure that it is not false and misleading). EFF appears at least somewhat uncertain of the standard that would apply, contending that courts would apply at least intermediate scrutiny to the Bot Bill. Williams, supra note 5.
9 McIntyre, 514 U.S. at 357.
10 United States v. O'Brien, 391 U.S. 367 (1968).
11 Clark v. Community for Creative Nonviolence, 468 U.S. 288 (1984).
12 McIntyre, 514 U.S. at 335.
13 Cal. SB 1001, §1. (Emphasis added.)
14 Section 4(b) of the Federal Bot Bill has similarly broad language, as it would order the Federal Trade Commission to promulgate regulations that would require "a social media provider to establish and implement policies and procedures to require" any user of its social media service "to publically disclose the use" of any bot.
15 Cal. SB 1001, §1. (Emphasis added).
16 In addition to the broad requirements imposed on social media providers in Section 4(b), Section 5 of the Federal Bot Bill imposes specific prohibitions on political actors. It prohibits candidates and political parties from using bots that are intended to impersonate or replicate human activity online to share public information. It also prohibits political committees, corporations, and labor organizers from using bots intended to impersonate or replicate human activity online to advocate for the election or defeat of a candidate or to send electioneering communications.
17 McConnell v. Federal Election Commission, 540 U.S. 93, 198 (2003).
18 Citizens United v. Federal Election Commission, 558 U.S. 310, 315 (2010) (quoting McConnell, 540 U.S. at 198).
19 Id. (quoting Buckley v. Valeo, 424 U.S. 1, 64 (1976) and McConnell, 540 U.S. at 201).
20 This is a failing of the Federal Bot Bill as well. Although it requires that the Federal Trade Commission draft regulations that would require social media providers to implement policies banning unidentified bots, it does not require that those providers enforce their policies or do anything in particular to stop bots from impersonating real human beings.
21 The Federal Bot Bill seeks to address this concern by requiring that a social media provider create a suspension process and a process through which human users can demonstrate that their activity is in compliance with the terms of its bot policy.
22 See Ethically Aligned Design, v.2, IEEE, 159, available at http://standards.ieee.org/develop/indconn/ec/ead_v2.pdf ("If users are aware that they are interacting with an [autonomous system] in the first place, and know exactly what information is being transferred to it, they will be better suited to tailor their inputs. A government-approved labeling system . . . could be used for this purpose to improve the chances that users are aware when they are interacting with [an autonomous system].").
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.