Written by a Human (For Now)

Just a quick follow-up to my last post, discussing the possible consequences for constitutional law of the increasing role of algorithms in the (re-)creation of information. While that post, and Josh Blackman’s essay on which it was based,focused on search engines, a post on The Guardian’s website this weekend discusses the “writing” of actual news stories by algorithms.

As the post explains,

Forbes.com already uses an artificial intelligence platform provided by the technology company Narrative Science to generate automated news from live data sets and content harvested from previous articles. What makes it possible is that business news content tends to be formulaic and data-heavy, listing places, stocks and company names. The LA Times, meanwhile, uses robots to report on earthquakes: the organisation relies on an algorithm that pulls in data on magnitude, place and time from a US Geological Survey site.

Sports news might be next on the list, and who knows what after that.

And so the question of constitutional protection for algorithm-generated expression may well be even more important than just the issue of the regulation of search engines (important though it is!) might suggest. So are, of course, other questions about the regulation or potential liability of “robot”-speakers. Should, for instance, the law of defamation take the same shape for algorithm-generated defamatory materials as it does for human-produced ones? The mental state of a defendant (knowledge or malice) can matter in a defamation action, but does an algorithm have a mental state?

I have no answers for now, of course. There is a lot of fascinating work to be done here for lawyers. And maybe for robots too. Who better than them to tell us about their mental states, right? But this blog, at least, will continue to be written by a human being. For now!

UPDATE: At the Volokh Conspiracy, Stuart Benjamin also has a post about articles generated by algorithms (such as the LA Times’ “Quakebot0t”), asserting that they are ― in the United States ― constitutionally protected speech.

FURTHER UPDATE: The BBC also has a brief story on “robo-journalism”.

Charter, Meet Google

Josh Blackman has just published a fascinating new essay, “What Happens if Data Is Speech?” in the University of Pennsylvania Journal of Constitutional Law Online, asking some important questions about how courts should treat ― and how we should think about ― attempts to regulate the (re)creation and arrangement of information by “algorithms parsing data” (25). For example, Google’s algorithms suggest search queries on the basis of our and other users’ past searches, and then sort the available links in once we hit ‘enter’. Can Google be ordered to remove a potential query from the suggestions it displays, or a link from search results? Can it be asked to change the way in which it ranks these results? These and other questions will only become more pressing as these technologies become ever more important in our lives, and as the temptation to regulate them one way or another increases.

One issue that is a constant theme in the literature on this topic that prof. Blackman reviews is what, if any, is the significance of the fact that “with data, it is often difficult to find somebody with the traits of a typical speaker” (27). It thus becomes tempting to conclude that algorithms working with data can be regulated without regard for freedom of speech, since no person’s freedom is affected by such regulation. If at least some uses of data are, nevertheless, protected as free speech, there arises another issue which prof. Blackman highlights ― the potential for conflict between any such protection, and the protection of privacy rights, which takes of form of prohibitions on speaking against someone (in some way).

The focal point of these concerns, for now anyway, are search engines, and particularly Google. Prof. Blackman points out, as Google becomes our gateway to more and more of the information we need, it acquires a great deal of power over what information we ever get to access. Not showing up high in Google’s search results becomes, in effect, a sentence of obscurity and irrelevance. And while it will claim that it only seeks to make its output more relevant for users, the definition of “relevance” gives Google the ability to pursue an agenda of its own, whether it is punishing those who, in its own view, are trying to game its ranking system, as prof. Blackman describes, or currying favour with regulators or commercial partners, or even implementing some kind of moral vision for what the internet should be like (I describe these possibilities here and here). All that, combined with what seems to some as the implausibility of algorithms as bearers of the right to freedom of speech, can make it tempting for legislators to regulate search engines. “But,” prof. Blackman asks, “what poses a greater threat to free speech ― the lack of regulations or the regulations themselves?” (31) Another way of looking at this problem is to ask whether the creators and users of websites should be protected by the state from, in effect, regulation by Google or Google should be protected from regulation by the state (32).

The final parts of prof. Blackman’s essay address the question of what happens next, when ― probably in the near future ― algorithms become not only tools for accessing information but, increasingly, extensions of individual action and creativity. If the line between user and algorithm is blurred, regulating the latter means restricting the freedom of the former.

Prof. Blackman’s essay is a great illustration of the fact that the application of legal rules and principles to technologies which did not exist when they were developed can often be difficult, not least because these new technologies sometimes force us to confront the theoretical questions which we were previously able to ignore or at least to fudge in the practical development of legal doctrine. (I discussed one example of this problem, in the area of election law, here.) For instance, we have so far been able to dodge the question whether freedom of expression really serves the interests of the speaker or the listener, because for just about any expressive content there is at least one speaker and at least one listener. But when algorithms (re-)create information, this correspondence might no longer hold.

There are many other questions to think about. Is there some kind of baseline right to have Google take notice of you? Is the access to online information of such public importance that its providers, even private ones, effectively take on a public function, and maybe incur constitutional obligations in the process? How should we deal with the differences of philosophies and constitutional frameworks between countries?

This last question leads me to my final observation. So far as I can tell ― I have tried some searching, though one can always search more ― nothing at all has been written on these issues in Canada. Yet the contours of the protection of freedom of expression under the Canadian Charter of Rights and Freedoms are in some ways quite different from those under the First Amendment. When Canadian courts come to confront these issues ― when the Charter finally meets Google ― they might find some academic guidance helpful (says one conceited wannabe academic!). As things stand now, they will not find any.

Not Private Parties

The development and use of massive voter databases and sophisticated “micro-targeting” techniques by political parties are raising concerns about the privacy rights of the people targeted by these efforts. When I wrote about the use of these techniques by the Obama campaign in the last presidential election in the United States, I suggested that “the future is creepy.” I am not the only one to think about this stuff, of course. The CBC had a story about the use of databases by Canadian parties, not only to identify and turn out their own supporters, but also to discourage opponents’ supporters from voting, though it did not specifically discuss privacy issues. Radio-Canada’s internet blogger, Vincent Grou, raises them in a recent post, and points to a report on the topic prepared last year for the Office of the Privacy Commissioner of Canada by Colin J. Bennett and Robin M. Bayley.

The main takeaway from this report is that, although parties now collect and use large amounts of information about voters, members, donors, activists, and candidates, they are not subject to the existing privacy legislation in Canada, which typically only applies to governments and to commercial enterprises, except for British Columbia’s Personal Information Protection Act, S.B.C. 2003 c. 63. (Readers with good memories may remember me saying the contrary in the post linked to above. I was wrong. Mea culpa.) This is troubling, because parties do not respect the rules that other organizations are subject to, and put Canadians’ personal information at risk. At the same time, parties are very important in the Canadian democratic process, in particular in organizing and promoting political participation. Thus, “in general terms, the debate centers on the balance between the two values of  personal privacy and political participation” (3).

There report also notes a number of background factors which this debate must take into account. One is the continuing decline of the importance of parties for mobilizing the electorate, and loss of public trust in parties. In this context, the report argues, one party’s sins of commission (such as an abuse of information on voters) or omission (such as a failure to protect information) might prove very harmful for the democratic system as a whole. A second is that the technologies which parties use directly (their databases and data-mining software) or indirectly (for example social networks on which they can gather information) are developing very quickly, and thus the parties’ behaviour is difficult to regulate in ways that will make sense not only at present, but also in the future. Yet another point to keep in mind is that in many ways, the new data-mining and micro-targeting methods are developments of existing practices rather than completely new phenomena. Parties have long made efforts to identify their likely voters ― but the new technologies allow them to take these programmes to a much higher level.

Another interesting part of the report is its survey of practices in other democratic countries. They vary widely. In continental Europe, parties are subject to privacy regulations, and micro-targeting is apparently not developed. By contrast, in the United States, privacy rules are much less extensive and do not apply to political parties (which indeed are shielded from much regulation by the constitutional protection of freedom of speech).

Although prof. Bennett and Mr. Bailey do not say so in the report’s conclusion, they seem to favour at least some form of regulation of the political parties’ behaviour with respect to privacy. They acknowledge the difficulty of striking the right balance between what they consider to be the parties’ special role in our democracy and the citizens’ privacy concerns. But they are unimpressed with the parties’ purported attempts at self-regulation in this respect ― their privacy policies, when they exist at all, are hard to find, vague, or incomplete. And they worry not only about the privacy interests of citizens but also about the risks for the political system as a whole if one or more parties fail, or are regarded as having failed, to protect the privacy of the millions of people about whom they collect information.

This is all very interesting and worth thinking about. But I would like to suggest a few other factors to add to our reflection.

One is the possibility, which I raised in my previous post, that constitutional law has something to say on the subject of the permissibility of regulating the gathering and use of data by political parties. As I noted then, the Supreme Court is now considering a case in which a union argues that its constitutionally protected activities should be exempt from the application of provincial privacy legislation. The activities of political parties too enjoy some constitutional protection, the Supreme Court having recognized their special role in the electoral process. Depending on the outcome of that case, political parties might be able to challenge any extension of the privacy legislation to cover their activities.

In this connection, another point to keep in mind is the difference between rules that limit the parties ability to gather information about voters and those that would attempt to make them better custodians of the information they collect, for example by requiring better security measures or training for the party workers and volunteers who handle personal information. The former sort of regulations, I should think, would be more problematic than the latter.

Another concern when thinking about possible regulation is the congruence of the proposed rules with the expectations and practices of the people being burdened, and those purportedly being protected, by the regulation ― here, respectively, the parties and the citizens. On the side of the parties, it is worth noting that Canadian political parties are actively trying to learn from their American counterparts. They are thus taking on board the practices and expectations of what might be the world’s least regulated environment from a privacy standpoint. The more they do so, the more difficult and intrusive forcing them to change their ways is going to be. As for the citizens, the report notes that Canadians say they are very concerned about their privacy. But deeds do not necessarily match words, and it is not clear that regulations should protect people in accordance with their stated wishes when they themselves seem not to act on them.

A final point I will make here is that we might want to question one of the report’s important assumptions ― the special role of political parties. I have written a good deal about why should be skeptical of campaign spending rules that favour political parties (here, here, and here). As the report notes, parties are losing the importance they once had ― but that is not necessarily a bad thing. It is not so obvious that they ought to benefit from special rules with respect to privacy, though I am not saying that they ought not to ― it’s a difficult question.

As it stands, Canadian political parties are not very respectful of our privacy. Whether to try to make them more so, and how, are difficult questions. They are bound to become more pressing as the parties’ gathering and use of data on voters continues and increases, and we would do well to start thinking about them now.

The Future is Even Creepier

There is an interesting story in today’s New York Times that brings together a couple of my recent topics, the tracking of internet users by the websites they visit and the use of the data thus generated in advertising, about which I wrote here, and the use of target-specific outreach and advertising by President Obama’s re-election campaign, about which I wrote here. There are even, for good measure, overtones of human dignity there.

The story is about the way the data gathered when we use the internet, whether just browsing or searching for something in particular, are then used to throw those annoying targeted ads at us wherever we go. The data is collected by computers of course; it is computer algorithms, too, that analyze it and use it to assign us to some fine-grained category (depending on our inferred interests and means); and it is still computers that sell the right to show us a display ad to companies that might be interested in the specific category of consumer each of us is deemed to belong to.

This is roughly similar, if I understand correctly, to what the Obama campaign did in studying the data it had collected about voters and using it to target each person specifically according to his or her likely interests and concerns, except that the field of application here is commerce rather than politics. And just as some people have doubts about the morality of that tactic in the political realm, there are those who are convinced that its application in the commercial one is immoral. The Times quotes a consumer-rights advocate as saying that “[o]nline consumers are being bought and sold like chattel [sic]. … It’s dehumanizing.” As with what the Obama campaign did, I’m not sure about that. I’m not convinced by the description of the process as selling people―it involves selling information about me, and the right to show me a message on which I remain free to act or not, not my personhood. I don’t feel dehumanized by those ads―just creeped out, which, I think, is a very human reaction, by the way (I doubt that cattle are creeped out by being sold).

Perhaps there is an echo here of the debate, in human dignity scholarship, over whether dignity and its violations are an objective matter, meaning that one’s dignity can be violated even though one doesn’t feel that it is ,or a subjective one, meaning that one’s perception is determinative. (A classic example of this problem is the controversy over dwarf-tossing: the dwarf consents to being thrown around for sport and makes money out of it―but can the state prohibit the activity regardless, on the ground that it is a violation of his dignity even if he doesn’t think it is?)

I should note one possible difference between what is happening in the commercial advertising context and what the Obama campaign did. The companies that track internet users claim that those whom they track are not identified in any recognizable fashion. When they sell the right to show me ads to advertisers, they might describe me as something like “the guy who reads legal blogs and news websites a lot and has been looking at cell phones recently.” The Obama campaign, of course, was identifying people by name, address, etc., in order to reach out to them. So maybe the internet-ad people are less creepy than the politicians. But maybe not. The Times’ article suggests people are very skeptical about the actual anonymity of internet users tracked by advertisers, so the difference might be illusory.

As I said above and in my previous posts, even if this is not immoral and/or illegal, it is creepy. Perhaps “do not track” features of internet browsers will save us from the onslaught of creepiness. But not only are advertisers trying to fight them but, as they are pointing out, their use might undermine the bargain at the foundation of the internet―in exchange for putting up with ads, we get to enjoy all sorts of great content (such as this blog, right?) for free. Perhaps we are now finding out that the bargain was a Faustian one. But it’s likely too late to get out of it.

The Future Is Creepy

I had the chance today to be at a talk by two of the members of the legal “brain-trust” of President Obama’s re-election campaign, NYU’s professors Rick Pildes and Sam Issacharoff. (I have to brag: it was one of those moments that make NYU the best law school in the world.) Yet although they spoke a bit about the specifically legal work they did for the campaign, the main topic of their remarks was the impressive data-mining operation that the campaign deployed to identify its likely supporters and make sure that support translated into votes. Using publicly or commercially available information, the campaign set up a massive database containing prodigious amounts of information on millions of voters, and used this information to make its pitch to these voters in ways designed (and tested) to be the most effective. According to profs. Pildes and Issacharoff, as well as this account of the data operations by CNN/Time and an envious paean by the Republican columnist Marc Thiessen in the Washington Post, this was fantastically effective. Profs. Pildes and Issacharoff told of the data people on the Obama campaign having predicted the outcome of the early voting in Ohio to within a few dozen votes―out of more than 1,600,000. (Over at the Volokh Conspiracy, however, Ilya Somin had some anecdotal evidence that not everything worked quite that mind-blowingly well.) Assuming that the data campaign really was as effective as insiders claim, its success has some interesting implications, both in the realm of law and in that of political philosophy. (And in that of practical politics too, naturally, but I try to stay away from that here.)

Legally, one obvious place for investigation seems to be privacy law. I cheerfully admit, however, that my knowledge of the subject is, at present, nil, especially as regards the United States. In Canada, there is a patchwork of federal and provincial statutes regulating the collection of personal information of the sort the Obama campaign used (one example that came up several times in the discussion today was magazine subscriptions, which apparently tell quite a lot about one’s politics). Some of these statutes apply to governmental entities; other to the private sector. For the most part, they do not apply to political parties, but they might impede the parties’ collection of information from other sources. They might also be expanded to cover political parties (and other non-profit organizations).

If they are, or to the extent that privacy legislation does apply to parties, there arises a further question, suggested by a case in which the Supreme Court of Canada just granted leave to appeal, as the blog The Court reports. Very briefly, the dispute is about whether preventing a union, pursuant to privacy legislation, from filming people crossing its picket lines is a violation of its freedom of expression. The Alberta Court of Appeal held that it did. If that decision is upheld, it would seem a logical, though probably not an obvious, step to argue that preventing a political party from creating a database it needs to mount an effective campaign is a violation of its Charter rights, notably its right effectively to participate in the political process, which the Supreme Court, in Figueroa v. Canada (Attorney General), 2003 SCC 37, [2003] 1 S.C.R. 912, held was part of s. 3 of the Charter.

Moving from law to political philosophy, the obvious issue to think about is the morality of this enterprise of targeting individual voters on the basis of personal information about them. The (admittedly few) friends with whom I have discussed it think it is ethically troubling. I’m not sure about that, but I do think it is creepy. Yet there is, maybe, a positive spin to be put on it. What the Obama campaign did was, in effect, a shift from trying to persuade people to support it to identifying those who already supported it, more or less, and making sure that they voted. This is disturbing if we hold to the high-minded notion that election campaigns are about persuading voters. But is this notion so high-minded? Is it so good for democracy if voters are (only) thinking about politics and deciding to vote one way or the other during the 35 (in Canada) or 60 (in the US) days of an election campaign? Arguably not―the process of making up one’s mind should be (more or less) permanent. If an election campaign is the time to reveal preferences rather than to shape them, then the shaping of the voters’ preferences must take place at all times between elections; this means that politicians must be constantly engaged with the electorate, and that might just be a great thing, insofar as it would blunt the not uncommon criticism of representative democracy as being democratic only once every four years. Yet perhaps this is too rosy a view. Constant engagement with the electorate may well degenerate into populism; there is a reason most countries have elections every four or five years, rather than more frequently. And anyway, it’s not clear that (most of) the electorate has the appetite for such engagement. And even if those worries are unfounded, the creepiness factor remains―though I suppose we will get used to it soon enough.

I conclude on an even creepier, indeed perhaps paranoid thought, blending the legal and the political strands of my reflection. The Obama campaign’s database of voters and information on how effectively to communicate with them is immensely valuable. Mr. Obama himself has no use for it anymore, since he cannot run for re-election. But to those who aspire to succeed him, it would be invaluable. And legally, as profs. Pildes and Issachoroff told us, the database belongs to him, or more precisely to his campaign committee, over which he―not the Democratic Party―has control. Is it paranoid to suppose that the control of this amazing resource will allow him effectively to designate his successor by handing over to him or her a weapon which no challenger could hope to match? If so, the future might be even creepier than we think.

UPDATE: The New York Times also has a story on this, focusing on how data analysis drove the Obama campaign’s purchases of TV advertisements.

UPDATE #2: The post as originally written was quite obviously wrong in stating that private-sector privacy legislation probably applied to parties. I have re-written that passage. I discuss the issue of privacy and political parties in more detail here.