After AGDQ 2015, micro500 began working on a DS Bot to be used ag AGDQ 2016. Brain Age seemed like a great game to showcase TASing and to provide significant entertainment. There is even a published TAS with a star. In discussions about developing a movie specifically for AGDQ, some great concepts (spoilers warning) were developed.
The published, starred, Brain Age movie was entertaining because while entering a variety of detailed drawings the author was able to fool the games character recognition algorithm. While working on the AGDQ concept though, xy2_ and micro500 discovered that it was possible to write an invisible answer separately from an aesthetic drawing. This is demonstrated in the following video (at least the invisible answer part):
Link to video
These two tricks (which I'll allow them to describe in more detail) combined totally break apart the game. This is very impressive and will be exceedingly helpful for the AGDQ demonstration. The problem is that this really takes some of the magic away from tricking the game with complicated drawings (a sentiment that Masterjun and ais523 shared in IRC).
So, here's the pickle: We now have the ability to answer the Brain Age question correctly, off screen and then draw whatever we want visibly. These are two independent actions and that makes the drawing 100% superfluous and a 1:1 speed to entertainment tradeoff. Judging the entertainment of such a run quickly becomes more of an art studio than a technical accomplishment.
Here's my take. I still think using these tricks for AGDQ makes sense (but I think we should hold back on the technical nuance that the answers are completely separate from the drawings). For a TAS submitted for publication on this site, however, I think we've entered an area that is too much "Color a Dinosaur" or "Mario Paint" where we draw simply because we can. To me, such a TAS does not belong on the site, or at least it does not clearly fit in the current scheme. I'm curious what others think.
I think we shouldn't be showcasing something if people would feel cheated when they learn how it's done.
In the early days of TASing, people felt cheated when they found out that there wasn't a human doing the input, and it lead to a lot of bad blood between the speedrunning and TASing communities (which has mostly since evaporated). We've since fixed the problem by managing expectations; a TAS often isn't interesting for the same reason as a speedrun is, but it has other compensating benefits.
In this case, again, I'd feel cheated if I watched a run that looked like it was about hiding numbers inside images, and was actually about drawing numbers invisibly/offscreen and (independently) drawing images to make it look like something more technically demanding was going on. I'm not quite sure that managing expectations will help fix the problem in this case.
The fancy and interesting part is tricking the pattern recognition algorithm into "recognizing" the correct answer from a drawing that seemingly has no correlation to it.
If you remove that, then you have nothing. It would make no sense. The entire point of the TAS would be gone. You could just as well show a slideshow of pictures instead of the game.
As ais523 above says, if you were to outright mislead the viewers into thinking that it's the visible drawing that's fooling the program (but behind the scenes it's an off-screen trick, and the drawing has zero effect on the program), then that would really be akin to cheating. Even if you were to outright state how it works, it would lose all of its marvel.
I don't think using the off-screen trick is suitable either for any marathon showcase nor any publication at tasvideos.org.
To describe the tricks in more details, it is recommended to have a basic understanding on how recognisition of numbers (aka recon) works in Brain Age. If you haven't, check out the first and second Brain Age post by me in the DS forums.
The first trick is called edge breaking. When you draw something in a single input (no matter how long it is), edges are placed at the start and end of your line. Edges are also used to manipulate models. Manipulating anything with 6-8 edges is almonst impossible. This would mean that we would have to do most drawings in a single stylus input without stoppi g input.
However, by going on the edges of the drawable field for 1 frame then going back in wherever we want, we can manage to stop the line we are currently drawing and start it somewhere else without generating an edge. This means all the drawings in the run will have 2 edges (thank god) or 4 if for a double digit answer. This is seen as okay with everyone since it is recon manipulation.
The other thing with edge breaking is that when you are out of bounds with edge breaking, you can go back in bounds for 1 frame to make one input invisible. Note that while out of bounds, you still generate keypoints (they are just pushed back to the border of the drawable field.)
The second glitch is called godspots. By placing inputs on a very specific position (x:246) we can make the game not show up the pixel since it is too high and is blocked by the top of the screen, but will still count as a keypoint/edge. Using this, we can make completely invisible models. Note that drawings still generate their own keypoints/edges, so we have to carefully make completely in visible edges or keypoints (and sometimes redraw over alderaly deawn portions of the drawing field if we need to be back in bounds for more than one frame), so manipulating with godspots is extremely tough.
The glitch everyone argues on is out of bounds drawing. By making our first frame draw out of bounds, then going back in bounds, our entire drawing will not be recon and will not have keypoints or edges. This breaks the entire game and makes the drawing and invisible models separate, what most people are against.
Moderator, Senior Ambassador, Experienced player
(898)
Joined: 9/14/2008
Posts: 1007
Warp wrote:
I don't think using the off-screen trick is suitable either for any marathon showcase nor any publication at tasvideos.org.
I could not possibly disagree more, on all sides, and I'm quite passionate about this.
First, for GDQ events we do some really technically challenging things - for this upcoming Brain Age pitch, it will involve live capture of the screen, live OCR processing to determine the question being asked, adapting an arbitrary image consisting of nothing but line drawings *on the fly* to come up with the correct answer, and then sending it back to the DS by hijacking the touchscreen. On top of all of that, we're even able to do the equivalent of miscalibrating the screen (something a human could do easily) to allow us to tap out-of-bounds, and we abuse that ability to have complete and total control over what we draw. I find this *very* impressive, and I'm amazed at what xy2_ has discovered in the last week.
I think everyone in this thread is biased by the existing Brian Age run made when this trick had not yet been discovered and thus now that we have the ability to draw anything we want it's seen as cheap. That's fine, I'm OK with that insinuation, but saying that we shouldn't use this amazing discovery at GDQ is ridiculous to me. We do occasionally resort to some small amount of stage magic at GDQ events (the camera didn't work like it should have, although it *could* have :) but I don't think this is stage magic, merely a very clever exploit.
Downplaying its existence entirely won't help us, but I agree that explaining it all right up front is also a bad idea; we want people to see things unfold, see the answers becoming subtly more wrong, and then get to a point of being outright completely wrong, and to that end I'm fine with using the methods of the existing Brain Age TAS for at least a few panels, but to not take advantage of this massive discovery seems like being intentional Luddites to me.
Regarding publication on the site, we've long wondered how to showcase things we do at GDQ's on the site. There is no place to put these works yet which is something Nach has made a proposal to to resolve by adding a technical showcase tier. I feel strongly that the existing Brain Age TAS should *not* be obsoleted by anything we do at a GDQ, but I also feel strongly that runs that could each be considered a work of art in their own regard should be able to stand equally on the site.
This is the best possible type of game to have this discussion about - now that these discoveries have been made, Brain Age is literally just a game where you can make art by drawing similar to what Alden did with Mario Paint and even similar to the way that Otocky in music maker mode or something like Electroplankton provides ways to make music if you wanted to make a TAS of it. Each result is a work of art that should not obsolete the previous work of art because it's no longer about speed, and I think it's time to have that discussion again. Thoughts?
Haven't read the thread, but I think we should not blindly abuse invisible drawings to maximum for gdqs, but we need to put it somewhere as a yet another bug in the game, answering just a single question with emptiness. Actual drawings also should be at least tried to include a real answer in some way.
And yes, for TASing this game, it's become too much of a sand box now, so it completely removes all the challenge that was there formerly.
Warning: When making decisions, I try to collect as much data as possible before actually deciding. I try to abstract away and see the principles behind real world events and people's opinions. I try to generalize them and turn into something clear and reusable. I hate depending on unpredictable and having to make lottery guesses. Any problem can be solved by systems thinking and acting.
I believe the basis for the published run is the idea that the OCR in the game is so loose/broken that you can draw a picture and the game will accept it as the correct answer. You would probably think somehow the game is finding the answer in what you drew. Watching the run with that expectation leaves me entertained, because I enjoy seeing the game being broken in that way. For some of the answers that is exactly what happens, but for others I see that he added extra taps (both in and out of the drawing area), and also traced over some of his old work to trick the OCR.
This is where the line gets fuzzy for me since my expectations were a little off. I assumed he was only drawing the picture, but there is more going on. If I had known that up front I would have still been impressed, but a little less so. I still find it neat that the game somehow finds the answer in everything you drew, even when you factor in the extra taps.
The out of bounds drawing glitch allows us to draw anything we want, and the game does not use any of that input for the OCR. Your dots and lines do show up on screen, but they are completely ignored by the OCR. No answer is detected by the game, not even "?". This lets us draw anything we want without having to do the technical feat of tweaking our drawing to trick the buggy OCR. We then could draw right on top of our image with the answer and the game would accept it. With the invisible models glitch we can draw the answer with nothing being shown on screen, but in this case it is fed to the OCR. Combining these two glitches means we can draw anything we want, lift the pen, and then just draw the right answer on top (invisibly).
If that combination of glitches had been used by the published run it definitely would have crossed a line for me. The OCR is not being abused in any way. You are drawing the real answer, but you just happen to be drawing a picture before it. As others have said, this turns into a "Color a Dinosaur" type run where the drawings serve no purpose. Removing them would not prevent you from achieving your goal.
When you reach this point, "your goal" becomes the important thing to distinguish. Is there a goal to Color a Dinosaur? There is no end game as far as I know in that game. Is there a goal to Brain Age? You can make up a goal like trying to finish the Calculations x100 as fast as possible, or trying to reduce your "Brain Age" to be as low as possible, etc. The current run aimed for a playaround completion of a round of Calculations x100. This kind of goal (playaround completion of a game/minigame) shows off some glitches or fun parts in a game and makes a sacrifice of speed. It can be done in a number of other games, so the appeal of this run is that while wasting time drawing pictures he is entertaining the audience, but the drawings are also being used to complete his goal. When watching the run you can assume the pictures serve some purpose, even if they aren't done optimally like they would be in a run meant for speed. If he had used the above glitches to make the drawing completely ignored, the drawings are truly just a waste of time and serve no purpose. It would be just like showing a few frames in a slideshow in the middle of a sort of speed-oriented TAS.
I'm seeing this set of glitches as a game-breaking glitch like other games have. In my opinion this set of glitches should not be used if someone wants to obsolete the current playaround run. However, I can see them being used as a separate category, distinct from the one I just mentioned, that specifically says it uses this set of glitches. This would clear up any confusion or bad expectations when people watch it as they would know what they are getting into when they hit play. I personally would enjoy seeing another playaround movie in this new proposed category, in the style of my blank input TAS posted in Weatherton's first post. It would be quick, and I personally would find it entertaining, but to me it is clearly in it's own category separate from Ryuto's published playaround. A speed-based movie may also be entertaining. This would involve manipulating the RNG to give you answers that are quick/easy to draw, and also drawing the right thing so the game accepts it as quickly as possible. Both of these ideas of course would not appeal to everyone, but I think some would find them entertaining.
As for the AGDQ aspect of this, I believe we should not abuse the OOB+invisible models glitches. I believe most people watching will just assume the OCR is broken, and that is how we are getting away with the game accepting the answers. If we use the proposed glitches, the drawings are just a misdirect to hide what is actually happening. We could not tell the crowd how we did it, but surely people will ask, and some will find these forum posts. We could lie, but that is a very bad idea. I personally would rather be able to tell the crowd that the game is indeed looking at everything we drew and finding the answers. I find that to be a much more entertaining thing to show, both visually and technically. If we did decide to use the glitches, I would feel obligated to tell the audience what is going on, but I know they would quickly feel less impressed. Many people would see it for what it is: a robot that can draw pretty (but useless) pictures, and then later answer a math question. A bot that can do either one of those, or both, is not impressive to me. The technical feat of hooking it up to a real DS is cool, and so is streaming the picture in from a computer, but I think tricking the OCR is the real technical crux of this, especially when you combine it with all the other technical feats. However, I do find the ability to draw nothing (visually) on screen and still answer the question entertaining, so I am on board with answering one question like that. It would show off the neat glitch we found without deceiving the audience too much. We can tell the audience how that worked without ruining their expectations for the rest of the run. This does mean we need to find a way to make our drawings work for all answers, but that is doable, and it would be an even more impressive thing to tell anyone who asks about the technical details.
However, I do find the ability to draw nothing (visually) on screen and still answer the question entertaining, so I am on board with answering one question like that. It would show off the neat glitch we found without deceiving the audience too much. We can tell the audience how that worked without ruining their expectations for the rest of the run. This does mean we need to find a way to make our drawings work for all answers, but that is doable, and it would be an even more impressive thing to tell anyone who asks about the technical details.
Doing a blank answer doens't actually need to use out of bounds, since you will not be drawing anything: if anything, it should help the audience get the gist of what's happening. The game is very complex in its inner workings and it would be hard to explain everything in a 8 minute segment.
A blank answer needs out of bounds (or something very close to it) in order to enter the "actual" answer.
I'd be OK with drawing an actually blank answer because you can explain it, without lying or anyone feeling cheated, as having drawn the answer out-of-bounds (this is one method to actually do it). Drawing what's effectively a blank answer that doesn't look blank is beyond the line, though.
(Also, saying that there's technical difficulty because the DS TASbot has never been done before is sort-of missing the point; that'd be like making an unoptimized TAS for a never-before-TASed platform, submitting it, and claiming it should be accepted because of the effort that went into writing the emulator.)
I could not possibly disagree more, on all sides, and I'm quite passionate about this.
What would be the point in drawing anything on screen if it has zero effect on the program? Ok, you drew a funny picture which has no consequence on whether the program accepts your "answer" or not. Then what? What's the point?
The fun is not in seeing pictures being drawn. The fun is having those pictures fool the program. It's funny to see the program expecting something like "16", but instead the TAS draws a picture of Samus... and the program accepts it as the correct answer because its OCR was fooled. That's the whole meat of a Brain Age TAS. Remove that, and you have nothing.
If you want to demonstrate in the marathon how the program can be fooled by an off-screen picture, then do it once or twice (and emphasize it by leaving the visible portion empty.) But not all of the answers. Fool the program with funny pictures, not the viewers.
Moderator, Senior Ambassador, Experienced player
(898)
Joined: 9/14/2008
Posts: 1007
I personally feel we have an echo chamber here and that the only people who care about whether we use these two combined glitches at the same time is us. My point remains that we have destroyed the game so thoroughly that we can do whatever we want - why would we *ever* want to hide this? Why would we want to artificially handcuff ourselves as if we were sorry for what we had done to the game?
Having said that, if we decide to go the extra mile to make the run work without these glitches, I assume the consequence is the artwork will be less impressive to look at but will be more impressive technically. To me, I'd error on the side of better art - after all, we've literally reduced this game to the point that it's a series of drawing pad sheets and we're effectively exhibiting art that serves a purpose in and of itself (albeit using a quite absurd method). The art itself *IS* the point because at that point it's the artwork that gets the attention, not the game!
This is probably where I branch out from others here at TASVideos. I look at the site as being part destroying games, part art, or using games to create art, and I think we ignore or devalue the art aspects far too often. Warp, it is clear to me that you don't find any point in a run that is purely focused on art, and that's fine, there are lots of runs on the site that meet your needs. I'd like to have a way to distinguish runs that *are* primarily about art from ones that aren't to make your life easier.
Warp, quoting micro500 about the existing Brain Age run, "I see that he added extra taps (both in and out of the drawing area)". The currently published run thus also uses some aspects of these tricks. Do you feel it should be sent to gruefood for this? For my part, I think [1734] DS Brain Age "playaround" by Ryuto in 06:33.66 should never have obsoleted [1569] DS Brain Age "playaround" by Ryuto in 04:08.10 - both are playarounds, both are primarily about art, and both should be left to stand on their own merits of being interesting to view from an artistic (not speed) perspective. Again, I realize I am likely in the minority here as I have repeatedly shown myself to be an inclusionist but it's the only solution that seems sane to me.
I assume the consequence is the artwork will be less impressive to look at but will be more impressive technically.
Not true: almonst any drawing, unless you're on a 63/42/81 and it has less than three keypoints or keypoints close togheter in a bundle spaced out from the next bundle of keypoints (which will never happen, unless in the . . . screens) can be made into anything from my experience.
I personally feel we have an echo chamber here and that the only people who care about whether we use these two combined glitches at the same time is us. My point remains that we have destroyed the game so thoroughly that we can do whatever we want - why would we *ever* want to hide this? Why would we want to artificially handcuff ourselves as if we were sorry for what we had done to the game?
I was going to weigh in with a similar remark: I quite agree that both glitches should be used, abused and combined to make the run.
In my opinion, whether the OOB glitch i used or not, the net result is the same: a TAS that abuses the OCR system by making it ignore almost everything being drawn on-screen. That it was actually recognizing something drawn offscreen, instead of mistaking part of the drawing for the answer, is pretty minor, given that you only need only a few pixels to give any answer anyway.
It strikes me as a bit overly dishonest to do that trick in conjunction with drawing a picture, but I'm having a hard time rationalizing that, even to myself. I kind of feel like it's the difference between doing actual sleight of hand (even if you're shown/know how it's done), and pretending like you're doing sleight of hand but it's actually camera tricks. Both are tricks, but I'm pretty sure most people would feel betrayed if they found out something was the latter instead of the former. It's violating the audience's 'trust' in their assumptions about what's happening.
At the same time though, I had assumed (before glancing over and not fully understanding the other thread) that the OCR for Brain Age was based on the last few strokes drawn, and what was going on in those was that after the image was fully drawn, the correct answer was giving within it in some way, which would be trivial for anything with a solid portion of any appreciable size.
I doubt I put it well enough, but my gut agrees with the sentiment others have expressed that it's fine enough to show a few times that it'll accept a 'blank' input, but otherwise, it shouldn't be used and perhaps something else, like highlighting the input that the game is actually 'recognizing' to get the answer, would be preferable if a new one is ever made to try to be artistically superior.
I’m not gonna read all these walls of text but I support the tricking the game and audience or whatever in order to make a more spectacular showing when it’s done live. Doesn’t matter if you’re “cheating” or whatever.
Edit: Oops this was about publishing such a TAS on the site. Dunno. I don’t think it’s that tragic a thing but I haven’t thought much about it.
To me, I'd error on the side of better art - after all, we've literally reduced this game to the point that it's a series of drawing pad sheets and we're effectively exhibiting art that serves a purpose in and of itself (albeit using a quite absurd method). The art itself *IS* the point because at that point it's the artwork that gets the attention, not the game!
I disagree. While the art is an important part of this, making a robot to draw a predefined image is not very impressive. Making a robot to draw the right answers in a clear font would not be that impressive either. But making a robot that draws a funny picture that is also processed and accepted by the game as a real number is impressive. It is the fact that the silly drawing was instead seen as a number is the cool part about this, not just silly drawing alone. If the silly drawing has no effect on the answer at all, why bother drawing anything at all? If this is only about the art, why don't we just show a Mario Paint run instead?
This glitch is rather game breaking and also repetitive. It ends up feeling like the Final Fantasy stairs glitch. The idea is cool, but you get the idea after seeing the same thing several times in a row. Our glitch definitely warrants a quick demonstration, but it gets old really fast if you repeat it for every answer. This is why on this site we have two separate categories when a game-breaking glitch is found. An example is Pokemon Gold/Silver. We have a run using memory corruption and one without. Some people want to see more of the game played out, so the author artificially limits him/herself to not using a simple awesome glitch in favor of showing off how far they can push the game without it. This usually ends up showing more of the game, as well as more of the author's skill in the game. Sure, the author could do a simple glitch and save themselves a lot of time, but they wanted more of a challenge, and (some of) the audience enjoys that.
I feel like it is the same case with Brain Age. Making a run that shows off this glitch for every answer is interesting in it's own respect (and might be worthy of it's own category, but that is a different debate), but showing off a run that tricks the OCR for every question is even more interesting. The challenge there is to make our drawings fool the OCR. If we are planning to wow the crowd with a run that is both entertaining and technically impressive, why wouldn't we take the harder route?
However, I propose a compromise. We know how much work it is getting ready for a *GDQ, and I know how much work it will (might?) be to manipulate the images, so in the interests of time I think we should convert as many drawings as we can to not use the glitch, but if it ends up being that we can't convert them all so be it. We will need to make it clear to the audience what we did, but we are also showing off the broken OCR
I'd error on the side of better art - after all, we've literally reduced this game to the point that it's a series of drawing pad sheets and we're effectively exhibiting art that serves a purpose in and of itself (albeit using a quite absurd method). The art itself *IS* the point because at that point it's the artwork that gets the attention, not the game!
Then I say we take out ALL the stops and just TAS drawings and animations into Flipnote Studio.
I agree that the original TAS was interesting because it hid the answers inside the visible drawings. Using invisible stuff that counts and visible stuff that does not is, well, cheating.
Here are good examples of what you can do with the new glitches that would be entertaining.
Write one wrong answer, cross it out, write another wrong answer, cross It out, write a right answer normally.
Somethinng like (OOB) hmm i wonder... oh, it's.. (normal) the answer.
write a wrong answer very CLEARLY and visibly, and the right one invisibly. this gives you the obviously impossible result of the game accepting an answer that's obviously wrong as correct.
Any number of other things. Combining visible ignored and invisible not ignored together when the visible ignored is a picture is definitely out of line.
I'm just wondering why are we're referring it as "cheating". It may be deceiving, but cheating? In context to (offline) video games, isn't that usually referring to Action Replay and other devices? The word "cheating" makes it sound like you're editting values rather than using the bot in write stuff for a casual viewer.
I believe they are talking about it as making the viewer feel cheated. A casual observer would assume that the crazy drawings are being interpreted as the answer, when in fact the drawings are completely arbitrary and don't actually affect anything. This could possibly lead them to feel cheated as they could question the whole point of the drawings if they were meaningless in terms of answering the questions.
I'm just wondering why are we're referring it as "cheating". It may be deceiving, but cheating? In context to (offline) video games, isn't that usually referring to Action Replay and other devices? The word "cheating" makes it sound like you're editting values rather than using the bot in write stuff for a casual viewer.
Doing one thing, but masking it with other input that appears to be what's actually going on but is a complete fabrication leads to bad situations. eg
http://tasvideos.org/forum/viewtopic.php?t=12652
It's easy to imagine some kind of graphical corruption that would then let somebody clandestinely use cheat codes or passwords that they wouldn't normally be allowed to. That's why people don't have any issue with the seemingly blank input. It's an accurate reflection of the trick that's being used without a bunch of completely unrelated things obfuscating the actual situation.
Joined: 4/8/2005
Posts: 1573
Location: Gone for a year, just for varietyyyyyyyyy!!
So, a game is broken in a silly way. It is hardly the first time. Just explain to the audience that the game is broken. Then make the most entertaining thing you can. Do not underestimate the audience's capacity to understand what's happening. This whole thing will be awesome. It can literally be the cure for cancer.
Joined: 8/14/2014
Posts: 188
Location: North Kilttown
If we're talking about publications on this site, I'd personally like to see both the current run as a playaround (based off of one glitch) and one with this new trick being utilized for speed purposes (basically an Any% run). I don't see the issue with that.
As far as 'GDQ appearances, what's wrong with showing the game being broken in multiple ways - some doodle answers and some new glitch answers? Why hide it at all? The point is to be entertaining while, in some ways, saying "hey, look at how broken this is. Isn't that neat?" which is entertaining in itself.
My point remains that we have destroyed the game so thoroughly that we can do whatever we want - why would we *ever* want to hide this?
It's not a question of "hiding" anything. It's about what is impressive and what is not.
Drawing a picture and fooling the OCR to think it's a correct answer is impressive and interesting. Drawing a picture that has no consequence on anything is not. It's pointless.
More obnoxiously, making it look like to the casual viewer that it's the visible picture that the program is misinterpreting as the correct answer, when in reality it's not, is devious, deceptive and disappointing. The casual viewer will feel cheated when explained what's really happening, and that the visible picture had nothing to do with it.
Just consider the Ikaruga TAS in the last marathon. It was a real TAS run on an emulator... but what was shown was just a video file played with a video player on a PC. This caused disappointment from many viewers and quite some negative comments.
How much more disappointed would they feel in this case, where the drawings have no consequence?
And why exactly would you want to do it the deceptive way? Why draw a random inconsequential picture on screen, and have the actual answer be off-screen? Why? What's the point? Why even use Brain Age for this? If you want to draw pictures that have no consequence on anything, use Mario Paint or something.
Edit: In fact, there could very well end up being a different kind of deception (and literally hiding information from the viewers): Even if you explain clearly that the actual answer is being drawn off-screen and the visible picture has nothing to do with it, a viewer who knows nothing about Brain Age or TASing it could easily get the impression that it's actually not possible to fool the OCR with a seemingly unrelated picture. In other words, the viewer may get a false impression of the situation. Unless you are willing to explain "we could fool the OCR with the visible picture, but instead we are just giving the right answer off-screen, and the visible picture has nothing to do with it."
I really can't understand what exactly is the problem you are having. I already suggested using the visible picture to fool the OCR for the majority of answers, and then give one or two answers with a completely blank picture, explaining clearly to the viewers what's happening. (In other words, something along the lines of: "All these pictures fooled the OCR to think it's the correct answer. However, we can go even further and give the answer off-screen, making it look like the program is accepting a blank picture as the answer.") If it's just one or two such answers at the end of the demonstration, I don't see a problem with it. As long as the previous visible-picture-answers were genuine OCR-fooling.
