Keystroke Lotteries

Part II: Informant Lotteries

In Sicily the police worked secretly; an informant’s name is never known. But in America an informant must appear in court. And to inform is to invite swift reprisals. Consequently the already reserved and suspicious Sicilian shrugs his shoulders — “And if I knew, would I tell?” — The Gold Coast and the Slum (1929)

As described so far, keystroke lotteries and its variations would be for willing participants who would receive their own earned tickets and winnings (if any), as in any lottery. But if the keystroke-lottery model were to be successful on a large-enough scale, it might include a secondary market of participants: prisoners, parolees, and those sentenced to home-detention. They would all work in keystroke lotteries without collecting any tickets. Instead, in exchange for their labor they would receive incremental reductions in their sentences, or other credits. (The basis for eligibility to participate might even end up as a class divide among prisoners in general.)

To play those prisoner-generated tickets, anyone (prisoner or not) with information that could lead to the arrest and conviction of individuals responsible for yet-unsolved crimes, would submit that information anonymously via Internet interfaces protected by public/private-key encryption. Use of that protocol (configured especially for this purpose) would enable the police to securely communicate with their informants without requiring those informants’ personal identities. Upon a successful conviction (or some other appropriate stage of the proceedings), a successful informant would be sent, via the same encrypted interface, an agreed-upon number of lottery tickets (formatted as a string of numbers and letters) generated by prisoners. Those tickets would then be played by their new owners in whatever lottery was used to generate them in the first place. Given the digital environment involved, informants could specify the specific games they would accept tickets from, and when they would accept them, thus enabling them to change their odds of winning, exactly as if they were purchasing conventional tickets or earning them as ‘civilian’ participants in a keystroke lottery. As described in the first section, above, such a process would be inexorably educational regarding the realities of such lottery odds.

That education wouldn’t be the only advantage. A Los Angeles Times story several years ago about the role of the police in that city’s south-central district described how gang-members who had killed one young man showed up at his funeral and cheerfully partook of the feast in full view of his friends and family. Everyone knew who the uninvited guests were and what they had done, but nothing in the story indicated that the killers were ever brought to heel by the police or anyone else. Quite the contrary. The conclusion (unstated in the article of course) is that the prevention of local vigilantism is the only effective consequence of a heavy police-presence in dystopically high-crime areas. People living and working in such perversely-controlled environments are necessarily indifferent to anything except their own immediate well-being, a phenomenon well-described by Jane Jacobs in her book “The Death and Life of Great American Cities.” But now imagine the same funeral-feast in the presence of a keystroke-informant network, an amoral mechanism that would make community involvement with law-enforcement irresistible.

And now compare that imagined scenario with the present system. Most crimes get no bounty, and those crimes that do often still go unsolved. And there is no pricing-mechanism available to determine the value of information that could lead to conviction. Instead, that amount must be determined administratively, the main determinant being the municipal budget and the officially-perceived seriousness of the crime in question. Such a procedure is necessarily arbitrary and feedback is inefficient to nonexistent. Police learn that the amount is too low when no information is forthcoming, or the information comes in too slow to prevent further crimes by the perpetrator(s); and they know that the sum is too high when they are flooded with bogus information that also brings no suspects or convictions. And even attempting to earn the bounty by an informant means reporting your identity to the authorities, who can never be trusted to maintain secrecy. That can make possession of critical information useless, effectively reducing the bounty to zero. But an informant-lottery would perform the helpful service of empowering greed to trump fear: information would now be as valuable as it is potentially dangerous, thus making it valuable enough to warrant the easy effort of getting online and anonymously putting the information in play. The authorities might be trusted, on average, to maintain confidentiality of that play long enough to locate a named suspect before word hit the street, although it might not always be important if they didn’t: of the identity of the informant, there could be nothing directly known. That’s why public/private-key encryption is an integral part of this idea.

Valuation-psychology would necessarily change: informants would be competing against each other in cases of crimes publicly witnessed or committed by a group, some of whom would have lesser degrees of involvement. This would make possible a market-based pricing mechanism, possibly along the lines of a Dutch Auction: crimes with no other informants would command the highest price demanded: an open-ended stream of tickets that would stop only upon a successful drawing: Play Until Win crimes — a lottery-based annuity. Where any number of people had the critical information, informants would submit it along with a bid for the lowest number of tickets desired in return, possibly as little as one: Play Once crimes.

Objection!

*Informant lotteries wouldn’t work in cases where testimony is required.
True, but there would be fewer trials where testimony is required, and fewer trials at all because there would be fewer crimes in the first place.

*If there is no impediment to submitting lies and hunches, then a flood of them will result.
Reply: There have never been impediments. For the price of a postcard, phone-call, or email access anybody can submit information anonymously (although doing so wouldn’t win anything). The question is whether the chance of winning an informant-lottery would corrupt it to irrelevance. It’s more probable that, just as in Part I’s discussion of collusion, useful information would predominate. Even if it didn’t, tips that named too many different people, or even differed at all, would provide a useful measure of the probability of the particular crime ever being solved. It would be even more useful if the extent of the divergence were public knowledge. What wouldn’t be public knowledge is who the informants were.

*The slim chance of winning a lottery wouldn’t attract informants.
Reply: It certainly would attract informants, but their incentive, compared to conventional bounties, would be turned inside out. Now, community indifference, which currently forestalls action, would become a driving force. Informants would submit their information for no other reason than they might as well. It’s the flip-side of the coin of moral apathy: ask not why you should inform — ask why not? Would you neglect to pick up an unused Lotto ticket from the sidewalk?

*Does the keystroke get entered immediately, or only upon submission of the entire work?
Reply: I assume that you mean the entire work of a specific player who types a few seconds or minutes and then quits the project, rather than the entire work of all the typists put together upon expiration of the job’s deadline. I think each keystroke would be entered immediately, but that the ‘player’ would have to Save or Send his keystrokes before closing the browser to register them. By Saving your keystrokes, you might leave open the possibility of coming back later and continuing, or even retyping it after researching a point. You also might want to see what the other typists are doing with that particular string of text, assuming that capability were allowed (and this might be a critical point). Note that the players are not racing against each other to win the prize. At least as I envision it, each job worked on would have a specific deadline, and no money would be awarded before its expiration. I do think this point is essentially a technical one and not fundamental to implementation of the idea, although I should add that I’m not technically educated in the computer field.

*Does the work periodically get submitted; if not, how was it submitted when the user in the hypothetical story closed their browser?
Reply: It would be a bit like online Chat. You type your message, and the person you are chatting with is informed that you are typing a response, but can’t see it until you Send it. The computer would be noting your ongoing participation, but would have no data for consensus purposes until you finish typing and then Send it. And of course, you could have the option of configuring your browser to automatically Send, Save, or Delete upon closing the window. Again, I think primarily a technical question.

*Suppose I added some whitespace, or accidental characters in my transcription, leading to all of my letters being out of place by a few positions — how does the verification system adjust for that?
Reply: It’s important to remember that there are no “correct” keystrokes per se, just consensus-validated ones. Thus, strictly speaking, your whitespaces and accidental characters would be disqualified unless everyone else made the same ones. However, that only partially addresses your point. The greater problem of being out of sync with other typists because of a single added word space or character brings up the greater objection of how to link a given typist’s keystrokes with a specific string of text on copy. After all, lots of copy (such as legal text) is full of repetitious boilerplate. How would the computer know which section of copy you typed from? And is this important?

To answer this question, bear in mind that the whole idea of a keystroke lottery is based on massive levels of participation providing a lavish redundancy of keystrokes. This is a valid premise, because this system is, after all, a system of gambling, and people are attracted to easy ways to win money while gambling, and in a keystroke lottery a single consensus-validated keystroke makes you eligible to win the pot. That’s pretty easy work. Even if the pot were quite small, say $5 (or even less), you wouldn’t have to do much to have a chance to win it, and such low winnings would probably offer a guaranteed winner. If there was a lack of enough participation, then it would be possible to win with a consensus of one—your own keystroke(s). But the playing community isn’t going to allow that. Every last character on copy is going to be typed many times by many different players because every character they don’t copy is one less chance of winning for them and one more chance of winning for someone else.

As for the computer, it starts out with a blank slate—it is not comparing your keystrokes with the copy you are reading from. It is only comparing keystrokes from many different players. So any player could start typing at any point in the document, from the first character to any random one in the middle or at the end. Eventually, given sufficient participation, the computer(s) will find matches for that character, and (amazingly) be able to logically assemble everyone’s keystrokes together in a way that matches the original copy as perceived by the (ever changing) collective of typists. It will be able to do that because, although many people will type just small segments (perhaps especially in low-payout jobs, which may require that they be aggregated with others for payout purposes), as many people will type much longer ones, perhaps even the entire job. Those longer strings, sufficiently repeated and overlapped, will provide the chain the computer needs to assemble the smaller segments in order.

That means that you can’t just start typing letters at random, knowing that your letter ‘a’ must be validated, thus earning you tickets. Certainly you could type a single highly popular letter or word or phrase and then log off, and it would inevitably be validated. However, if you (or a bot) typed “a a a a a a a a a a a a a a a”; or: “the the the the the the”, such a string is simply not going to be validated by anyone else unless it matches the copy (which, in the example above, would be possible, although rare). As for a botnet trying this approach, that could be easily spotted and blocked.

The same logic applies to any other random string of letters. The system is going to consider your entire output (for that session) as the consecutive representation of what you are reading straight from copy. Of course, you could log on to a job, type a letter ‘a’ then log off, then log onto another job, type another letter ‘a’, etc. But why bother? Why not just log onto one job and start typing? This addresses a point I made in the article about collusion. You could try, but in the end you would better off spending your time just doing what everyone else is doing.

So to answer your original question, your added whitespace or accidental characters would be ineligible but the rest of your text, if consensus-validated, would be rescued.

I think this is the key to the success of my imagined lottery: you log on, just start typing anywhere, and type for as long or as little as you like. The computer-network, in combination with overlapping and linking strings provided by others, does the rest.

*If the submission is literally as you press each key, what happens when you press e.g. Backspace? Does that ‘undo’ the submission of that key, or is another entry accepted?

Reply: I think this question is addressed in my second answer: the backspace key wouldn’t register. But there might programmed into the system a unique keystroke-combination to type if you did want it to register it. There may conceivably be times when a backspace could be appropriate in typing copy.

*Wouldn’t it be possible to script/similarly automate something that simply deletes and retypes the same key over and over, giving vastly better odds—which could also be distributed across a bot-net for even better odds, while remaining economical?

Reply: In the system as I’ve described it, there would be no benefit to retyping the same key, nor in repeating it, nor by churning out random keystrokes or having a bot do this. Besides the fact that your keystrokes wouldn’t be validated by anyone else, you’d have to program the bot to type the letters at a human rate, not instantaneously, since the system would be programmed to look for bot-like behavior, and it could look very closely at the intervals between strokes and compare them to others churning out the same letters. Such botnet collusion would be obvious. The log-in process would be configured with Captcha-like requirements to block automated log-ins; and rigorous real-name registration would likely also be required. This would be perfectly acceptable to most people–the likelihood of having to pay taxes on huge payouts doesn’t hinder people from playing state-sponsored lotteries. (But as for Informant Lotteries and real names, read on.)

*I suppose your earlier points about blocking the IP’s of suspicious clients is mitigation that could be applied, but I wonder what the relative cost of keeping such a system secure would be? Certainly it would dwarf the meager prize amount ($100 in the example story) if it needed to be resilient against DDoS attacks, for example. Someone also has to host this service to the potentially millions of clients, who will all want it to be reliable and highly available, lest they were to be unfairly robbed of their ‘lottery ticket(s)’. The server and storage costs would not be insignificant.

Reply: When you look at this idea not as a system of work but as a system of gambling, the problem of how to pay for it disappears. After all, what is the cost of installing and maintaining Lotto machines in thousands of liquor stores, gas stations, and shopping malls? What is the cost of building casinos in the middle of a desert? As people look to keystroke lotteries as a source of gambling, as they began to catch on to it, they would very likely quit those other forms that are most similar but less convenient–and which require them to actually purchase a ticket to play. Although smaller payouts would probably require a guaranteed winner, larger payouts would not. Potentially there could be routine Powerball-sized, world-wide-funded payouts. Thus, a piece of the uncollected auction-money (from games with no guaranteed winner) could be used to support the system. After all, the funders of the lottery–the information-owners–wouldn’t expect to have all that work done at no cost to themselves and (as noted above) winners of big payouts don’t expect to keep it all. The system may even require that some auction-funders surrender their entire offer upon completion of the work, especially for those backing smaller-payouts. (Here an economical question arises: for larger payouts, would the amount surrendered equal the amount that would be paid out under a conventional hourly or freelance arrangement?) The percentage subtracted from the pot as a fee would no doubt vary, based on any number of metrics. Advertising would certainly play a role in funding as well.

As for security, I don’t think this would be more difficult than providing security for online banking. However, the question of security (and cost) might have a direct bearing on whether the entire system would be proprietary or peer-to-peer. I’m not qualified to address this point in detail, but I think it likely that there would initially be both kinds of systems. I suspect that informant lotteries might start peer-to-peer and stay that way, a reflection of a governmental disinclination to get involved (at least initially) in solutions of that kind, in the manner of the Bitcoin system (a point I address below.)

*Consider the further technical challenges presented by translation efforts. Individual translations are likely to differ in words, rather than mere keys, meaning it would be very difficult to award a lottery entry based on a keystroke. In particular, what will the impact be on the consensus-validation logic? Even if many entries are submitted, they are likely to all differ in several places due to word-choice and other variables that are highly expected when translating (consider that even amongst the automated translation tools available, the phrasing chosen varies non-trivially). This means that each submission will be impossible to compare on a keystroke basis, since it is highly unlikely that even two submissions will contain the same set of words, in the same order.

*Although I’m monolingual myself, like many people I’m aware of the problems, both literal and artistic, inherent in translations. I have to disagree that there would not be sufficient keystroke-consensus regarding certain kinds of translations. Your objection would probably apply to modernist poetry, but surely not for the bulk of business letters and technical papers, and perhaps even computer-program notations. As for the more problematical field of general literature, remember that the typists are going to be thinking: What is everyone else going to translate this to be? So there would doubtless be a flattening effect unpleasant to certain minds. Yet it might get the point across to many more minds, and so be useful for the price. And remember that all of the variations would be available to the auction-funder (the information-owner), who might use the variations to polish the translation as he saw fit.

As for technical matter (or even for literature), as I pointed out in the article, all the differences could be made available to readers upon completion, with all kinds of measuring metrics (perhaps expressed in colors, typesizes, etc.) available for comparison.

Your question is probably only resolvable by experimentation. Backwards translations via keystroke lotteries might be a useful testing tool. But ultimately, the only real test that matters is the price such translations would require, as established by auction. Multilingual typists would decide amongst themselves, spontaneously, whether a particular document is worth spending their time on. And don’t forget my point in the article about the growth of a Wikipedia-like culture and mindset. That could result in a new kind of translator, thinking with the crowd, instead of alone, as well as a new kind of reader of translations. I suspect both would arise, and make consensus-translations practical, at least for certain kinds of subject matter.

*Regarding the second idea [Informant Lotteries], I’m not sure how you would automate the analysis of the testimonies. They would have no base ‘source’ text to go from, and as such probably the only commonality between them will be keywords; notably, proper nouns (names of suspects, victims, localities etc.). Hence, it becomes difficult to weed out a ‘good’ entry as opposed to a ‘bad’ entry (there will probably be a lot of submissions with mentions of names that have appeared in media already- these will be difficult to weed out as ‘noise’, since they will all have a lot of keywords in common with each other).

Reply: I should have clarified this point in the article: informant lotteries would be a derivative function of keystroke lotteries but the two would have nothing technical in common–there would be no keystroke consensus-building at all in an informant lottery. Instead, tickets generated by prisoners and home detainees (or even volunteers) would be used as reward-currency, instead of cash, as I described it.

There is a problem though with weeding out bogus tips by criminals themselves or people just trying to game the system. (I actually addressed this point in an earlier version but inadvertently deleted it.) The solution is that informants would have to have skin in the game to participate. They would do this by working in keystroke lotteries themselves and accumulating a sufficient number of ticket-credits, just like a convict would. Criminal organizations could recruit plenty of typists, and bad information can result in conviction, but of course in general the better the incriminating information, the greater the likelihood of conviction. Another twist is that such attempts at collusion could themselves be made the subject of an informant lottery. Indeed that would probably be inevitable.

One interesting twist is that informants would be generating tickets that they could win now in order to win a chance for a winning ticket later. Informants with no ax to grind in a particular case would need incentive to make the effort. Probably the incentive would be that, upon a conviction based on an informant’s information, that informant’s ticket-stream would be based on how many tickets that person had generated, in conjunction with the seriousness of the crime and the number of competing informants, if any. The stream might match the generated tickets one-for-one for some crimes, and be a big multiple for other crimes. For example, if the information led to the capture of a suspect at large, then ten generated tickets might earn a stream of 100 tickets upon conviction. If the information led to a second conviction of someone already in prison, it might earn less, depending on the crime. Thus, although informants would be sacrificing tickets generated now for a chance to win later, the whole idea is that they have sufficient faith in the value of their information to bother doing so. And they would have a general social interest in punishing crime. Those motivations would be sorely lacking in typists submitting fake information (and thus bad information), either for themselves, a friend, or by coercion.

*It would also still be impossible to completely guarantee the anonymity of informants, even with private + public key encryption technology—and the fewer informants, the higher the risk. A few reasons are listed below:

*Informants would need a completely secure terminal to work on. If the hypothetical guilty party suspected one or more informants, they could potentially plant keyloggers / other malicious solutions to capture the information. It is then trivial to correlate personally identifying information with a submitted testimony.

*The lottery ticket must be cashed by the informant to claim the prize — it is also a unique code, which is known by the issuing agency. If they were compelled by law (or by some other means), they could find out the identity of whoever cashes that entry code- this could easily be traced by determining the identity of the owner of the bank account the money is deposited into.

*The layman informant will be unlikely to take sufficient steps to conceal their identity when submitting information (probably connecting from a personal device, with no proxy)– their IP would be trivial to attain, and hence their location and identity could be established.

Reply: It’s true that the fewer the informants, the higher the risk for those informants. However, there is also a greater chance of winning, given accurate information. As I point out in the article, greed can trump fear. So does outrage. But remember also that if informant lotteries caught on, it would stress any group or pair of criminals contemplating a crime. It would tend to spread distrust among them, and so inhibit them in their activities. Also the fact that a crime was in play wouldn’t necessarily be made public immediately. Thus, the informant could have time to take protective measures—measures no doubt planned in advance, given that that person would also know of the lack of other potential informants to hide among.

I’m not sure informants would need a completely secure terminal to work from. People who bank online don’t. Although a criminal party could install a keylogger, as the knowledge of exactly how to get a keylogger and install it spreads among thugs and their enablers, so will knowledge of those keyloggers spread to their intended victims. The solution is a simple one: don’t use your home computer–go somewhere else, or have someone else outside your immediate area do it for you. I think this point also addresses your third asterisk’s comment: even non-technical people can (and will) quickly learn the basics of computer security, if they haven’t already. Who doesn’t know that computers come with security risks? Certainly an informant lottery log-on page could state the risks and make recommendations, like logging in through a proxy-server to avoid IP tracing. Those deeply concerned might boot up with a thumb drive using Linux, a technique recommended to those seeking strong Bitcoin-wallet security. (Admittedly that would ask a lot of the public, at least at first.) And of course they could use their own or a trusted friend’s smartphone.

As for the second asterisk: the ticket would indeed have a unique code of some kind, but I’m not sure that it would necessarily be known by the issuing agency. There might at first not even be an issuing agency. Instead, informant lotteries might start out as an underground peer-to-peer system based on–of all things–trust, with (rare) lottery payouts made in Bitcoins. Indeed, informant lotteries might turn out to be Bitcoin’s killer app. Of course, all this describes an ungoverned system, with all the benefits and perils. There would be a lot of trial and error. Later, as the system (and Bitcoin) grew in popularity, governments would probably accept the system and possibly even improve it. At that point, an agency would issue the tickets, and couldn’t there be a way to create a PGP-protected ticket-generator to prevent tracing?

But even granting your point about the lack of perfect security (even Bitcoins aren’t perfectly anonymous), don’t lose sight of the fact that an informant lottery would surely be better than the present method and its hopeless shortcomings. Perfect the lotteries would not be, but potentially far better than the current alternative.

Next: Postscript