AGDQ 2020 INTERVIEW: TASBot and Speedrunning (Part 2)

There are two types of speedrun categories: speedruns done by humans (which comprised almost all of the speedruns at AGDQ 2020 that oprainfall attended last month) and tool-assisted speedruns. At AGDQ 2020, I met up with Allan Cecil, who is the keeper and spokesperson for TASBot. After AGDQ was done (raising over three million dollars in the process), we wound up talking a short period after the event about all things TASBot and tool-assisted related.

In Part Two of this two-part interview, Allan Cecil (also known as ‘dwangoAC’) and I talk about the debut of the Super Mario Maker 2 tool-assisted speedrun at AGDQ 2020, the issues that are inherent to developing a TAS for a Nintendo Switch title (including for The Legend of Zelda: Breath of the Wild), how tool-assisted speedruns are art, and more. If you missed Part One of our interview where we talk about what a tool-assisted speedrun is and about the history of TAS and TASBot at Games Done Quick events, then I suggest you check that out here first.

You can find out more about Allan Cecil and TASBot at their official website, by tweeting at them on Twitter, following on Reddit, chatting on the official Discord, subscribing on Twitch and on YouTube, and at their Patreon page.

You should also check out the official TASVideos website.

You can check out Awesome Games Done Quick at their official website, on Twitter, on Facebook, on Instagram, and on Twitch.

Finally, you can find out more about the Prevent Cancer Foundation at their official website, on Facebook, on Twitter, on linkedin, on Pinterest, on Instagram, and on YouTube. And you should definitely go donate now.

This interview has been edited for clarity and content.

Operation Rainfall: At AGDQ 2020, you showed off TAS runs of Pokémon Blue, Super Mario Bros. 2 and 3, and you showed off the first TAS for the Nintendo Switch Super Mario Maker 2 once the event raised three-million dollars. Where did the idea of using Super Mario Maker 2 for your first Switch tool-assisted speedrun come from?

Allan Cecil: Practicality drove our selection of Super Mario Maker 2. Unlike Super Mario Maker, on the Wii U, Super Mario Maker 2 has completely deterministic behavior. When you start a game of Super Mario Maker 2, each level that you start begins with the same seed of randomness. Meaning, player input that is identical will always result in identical outcomes in the level. This is a really important part of making a tool-assisted speedrun. For a TAS to work, playback has to be deterministic.

Because [if there is] randomness that is not controlled by us, the game will desynchronize. Because Super Mario Maker 2 is very deterministic, we were able to use other techniques that we were not able to discuss due to risks around damage to the community. We were able to use other techniques to send the right input on each frame and advance through the level flawlessly.

“We’re very conscious that what we’ve built [a Super Mario Maker 2 TAS] could cause a lot of problems if either Nintendo does not enforce the leaderboards or discovers that people are cheating and takes the leaderboards down entirely. So the risk to the community is not to our community, it’s to the real-time community, which we don’t want to break.”

OR: In the excellently written ArsTechnica article about Super Mario Maker 2 at AGDQ 2020, there is a reference to a ‘Switch timing problem’ that [isn’t] really explained beyond [how] TASBot does not not know when to send the button presses. What causes this ‘Switch timing problem’, exactly?

AC: I can explain that part in detail.

A normal NES console from the 1980’s or an SNES from the 1990’s using a method of asking the controller for input that is in lockstep with each frame of video rendered in the game. In other words, in every single frame, the controller is polled for input. Because of that, we can consistently stay in sync with the game. Every new frame asks us for new button presses. On the Nintendo Switch, there is no correlation to when the game asks the controller for button presses compared to what frame it’s on.

The problem with the Nintendo Switch is that it polls the controller at 125 hertz. So 125 times a second, give or take. It’s not an exact number. It’s a little bit higher than 125. It doesn’t divide into 60 frames a second. So we have some challenges to align the input to know where we’re at.

The secret sauce of how we did that is what we’re uncomfortable releasing due to the risks to the community.

OR: Explain a little bit more- the thing about doing a TAS for games like Super Mario Bros. 2 for the NES versus Super Mario Maker 2 for the Nintendo Switch is that one of the games has a major online component -such as leaderboards and invitational tournaments- that the other does not. How do you and the TAS community approach developing something that can have major ramifications in the online community if abused by developing TAS for ‘real’ levels instead of just ‘specially designed’ levels as was presented at AGDQ 2020?

And tying it back to my earlier question- how did the community decide what limitations to set on themselves for this type of Switch TAS runs?

AC: In this particular case, this is the first time we have ever withheld how we performed something. All of the hardware we’ve built in the past, including the current-generation TAS TM32 Playback Device, is open hardware with open source software. This is important to me, because I am the President of the North Bay Linux Users Group and I’m an advocate of open source software and open source hardware.

In this case, we did not release the source code for what we did, and we did not release the hardware. The hardware itself, in fact, was inside of a black box on stage at AGDQ while we played Super Mario Maker 2. This was because we knew that if we released this, there would be the possibility that bad actors would use the hardware to play Super Mario Maker 2 Ninji levels that have leaderboards that are being used to track the fastest completion of humans in one-week trials. We did not want leaderboards to become destroyed by bad actors using our devices on the online leaderboards.

This was a controversial decision. I normally would want everything to be released open source, but the risks that bad actors or trolls would use the methods and the device we made to cause damage was too high. So we opted to not release any more details than that. This is the first time we’ve ever been faced with a new-enough console to have active online leaderboards. So it was our first time having to face this decision. And we definitely wanted to approach it with caution.

Awesome Games Done Quick 2020 featured the world-premiere of a Nintendo Switch TAS for Super Mario Maker 2, which was created by ‘KNfLrPn‘.

OR: Are you pleased with the approach that’s been taken so far?

AC: I have mixed feelings.

To quote ‘GrandPooBear’: “This is a bittersweet feeling.” On the one hand, we’ve wanted to see a tool-assisted speedrun on a modern console for a long, long time. But on the other, we’re now playing with something that could damage the real-time speedrun community. And the last thing I want to do, as the keeper of TASBot and the ambassador of TASVideos, is to be the person who causes a rift between the tool-assisted community and the real-time community. We’re very conscious that what we’ve built could cause a lot of problems if either Nintendo does not enforce the leaderboards or discovers that people are cheating and takes the leaderboards down entirely. So the risk to the community is not to our community, it’s to the real-time community, which we don’t want to break.

And I want to say this right- I’ve worked over the last several years to build up relationships with key people inside of the real-time community because we have a very symbiotic relationship. The route that the real time runners develop are used by tool-assisted speedrun content creators and the techniques discovered through TAS testing are often able to be executed in real time by real time runners, which feed back into real time runs. And that symbiotic loop has been extraordinarily beneficial to a number of games.

Celeste is a good example. Techniques found using TAS tools were then incorporated by real time runners like ‘TGH’. And routing developed by ‘TGH’ was used in development of the TAS route. We don’t want to destroy that wonderful symbiotic relationship, so we’ve been very cautious to listen to the concerns of people inside that community and be very cautious with how we proceed.

OR: Something to be cautious about is the Digital Millennium Copyright Act, which states- it’s a long act- but in short, it says that you can’t circumvent software protection to get to copyrighted works.

AC: Correct.

OR: How do you and the TAS community avoid issues with this in relation to the Switch, the Xbox One, PlayStation 4, or future consoles?

AC: When showing content playing back on real consoles, we don’t modify them in any way. The beauty of what TASBot does is that we’re playing on completely original hardware. At no point in the process of playing back a tool-assisted speedrun am I violating the Digital Millennium Copyright Act. In the case of Super Mario Maker 2, we did everything using a completely unmodified console, a completely unmodified copy of the game, and a completely unmodified dock. We only used external ports to communicate with the game.

That means that we did not circumvent any encryption and thus did not invoke the Digital Millennium Copyright Act. In fact, we were in full compliance with Nintendo’s Terms of Service. We played the game entirely in airplane mode offline, and we did not violate their rules in any way.

OR: Has Nintendo reached out to y’all since the AGDQ 2020 demonstration?

AC: No. We’re actually thankful for that. Because if they had reached out in a way that they have reached out to other communities -if they had sued us- that would have been very uncomfortable.

We did not anticipate that they would do so. For one reason in particular: the optics of suing a charity mascot, for Nintendo, wouldn’t look very good for [them].

Pokemon Blue was a donation incentive TAS game, written by ‘gifvex‘, at Awesome Games Done Quick 2020.

You can find out more about how GDQ sets donation incentive levels in Part One of my interview with Matt Merkle at AGDQ 2020.

OR: So what’s on deck for SGDQ 2020? Are there any plans to have a TASBot run for that event?

AC: We have not yet started the process of submissions for that event. We are looking at what can be done.

Off-script but on the record, one of the challenges that I face is that in 2019, I took TASBot to seven different events. And some of them were very far away. I went to European Speedrunner Assembly, I went to Minneapolis for SGDQ [and] the East Coast. The challenge is that the travel costs are rather high. And ultimately I became limited by the amount of donations that I could get to offset travel and equipment costs. In fact, I went the opposite direction- I had been at the same day job for ten years, and I was laid off late last year. And that impacted me, because I did not have revenue for three months. And I fell quite behind.

And I have TASBot physical here with me. I take him to events as his keeper. But that costs money to fly me and TASBot around. He doesn’t ship well. I’m not saying he’s fragile, but he’s not something you just chuck in a Fed Ex box and ship across the country. And upon a digression, we are trying to make a robust replacement body for TASBot that is not based on a ROB base. In part because of legal reasons, we don’t want to use anything trademarked by Nintendo in our event content if we manage it- we don’t want our mascot that we’re trying to have [a] copyright have any legal concerns with Nintendo.

So we’re building a new TASBot body that is more robust for travel. But even then, it will generally be me taking TASBot from one event to another. So we’ve kind of hit a limit from a financial standpoint of ‘how do we get to all of these events- how do I get with TASBot to all of these events?’ So last year, I started a Patreon and that support has helped defray the travel costs [that] has allowed me to continue to go to events.

But as for SGDQ [2020], there are no guarantees that I can afford to go yet. I have to see what the support in my community is like to defray my travel costs, equipment costs for any games we may want to submit, and things like that. But the goal is to go to as many events as possible.

“So I see tool-assisted speedruns as an art and a very unusual medium that will become more accepted as art as generations pass and we look back on what was created. And I believe we will see this more and more as true art.

[…]

[We’re] making content that evokes emotions, that has its own beauty, that is a marvel in and of its own right.”

OR: On the Hair of the Dogcast, you mentioned that you were planning on submitting Dust Force for AGDQ 2020. Obviously, it wasn’t shown. Was it a case of the game being submitted and denied, did the community hold it back for another reason, or what happened?

AC: We decided to defer that TAS because the TAS authors wanted to make additional improvements to the game. I believe that run will be shown at a future Games Done Quick event.

OR: You also mentioned in that podcast that your most-wanted game was The Legend of Zelda: Breath of the Wild.

AC: Yes.

OR: Where is the TAS community currently at in completing The Legend of Zelda: Breath of the Wild as a tool-assisted speedrun?

AC: Zero percent. The issue remains that there are currently no emulators with TAS tools available for the Nintendo Switch with enough stability to move forward on a project of that magnitude. Beyond just making the tool-assisted speedrun, there is the additional challenge of console verification.

Whereas Super Mario Maker 2 had deterministic results, Breath of the Wild is not like that. We may never be able to do a true Breath of the Wild tool-assisted speedrun on original console without making additional modifications that would be more invasive than what we’re doing today. However, we might be able to an augmented tool-assisted run, where it goes back to the roots of what tool-assisted speedruns meant in the original version of it- where a human is playing the game but is assisted by macros and other techniques to complete the game faster than they would be able to complete with their own human skill.

That is still a possibility in the future. And even the possibility of a full tool-assisted speedrun playback on Breath of the Wild is possible. It’s just very hard. Many of the things we do are incredibly hard, so much more difficult than what most people understand. It’s a miracle that tool-assisted speedrun console verification is even possible.

The first-ever disc-based console verification was for Super Monkey Ball, as shown here at AGDQ 2018 and created by ‘CyclopsDragon‘ and ‘byrz‘.

OR: I want to talk with you about your untimely demise.

AC: Yes, let’s talk about that. *dryly*

OR: You mentioned to me at AGDQ 2020 and then later on the live stream about how to protect TASBot if you got “hit by a bus”. What steps have you taken to take care of the assets after your demise or you decide to stop running?

What sparked this?

AC: I don’t intend to abandon the community or stop doing anything in the TASBot community. I don’t intend to leave the TASBot community as a leader, in part because doing something where I’m living beyond myself brings me an immense amount of joy. Being able to do something larger than myself for charity has been amazing.

Last year alone, we raised 280,000 dollars and I wouldn’t want to give up organizing something with that kind of impact. But even though I don’t plan on leaving the TASBot community, I recognize two things: I need to be prepared in the event that I am no longer able to maintain that role and I need to make sure that I am not the ‘blocking’ person preventing the community from moving forward.

Right now, I am the primary gatekeeper that we show at each of these events. But as we start to look at other events, as we start to organize content for a broader variety of events, we have to allow others to be able to step up to be leaders [and] to have the ability to organize events. We have to scale by taking me out of the critical path. And a lot of the concerns right now are ‘what would happen legally with TASBots assets if I were to pass away?’ But the second part of that is ‘how do we make something like this a legacy that lives on beyond all of us? How do we make something that is more than just what I can accomplish in my lifetime?’ I want to see tool-assisted speedrun content being used for charity for as long as possible as people have interest in that field. And that means that I have to, to a certain extent, give up some amount of control. And that’s scary, but it’s the right thing to do. I don’t intend to abandon the community anytime soon.

And in fact, some members of the community have attempted to draft me as a benevolent dictator for life. Despite the ‘BDFL’ status, I recognize that the community needs the ability to grow beyond what I can do on my own. I’m saying this somewhat badly, but I think you get the idea. I formed TASBot as an L3C company to hold all of TASBot’s assets. That also gives us a legal shield of sorts, so if we did irritate some company in our charitable efforts, it would be TASBot-the-company that would get sued and not us as individuals.

OR: A couple more questions: Throughout this interview so far, you have referred to tool-assisted runs as ‘art’ more than once. Could you please elaborate on that?

AC: Any new medium that has content that is unique, that is built up from stuff that came from before but is almost unrecognizable as art, eventually emerges as being recognized as art in the future. You could say that in the beginning of rock ‘n’ roll, what people were doing with an electric guitar was not considered by a previous generation as art in any capacity. And now, it’s very widely accepted that someone playing an electric guitar is still performing a work of art. We even have awards for these things.

So I see tool-assisted speedruns as an art and a very unusual medium that will become more accepted as art as generations pass and we look back on what was created. And I believe we will see this more and more as true art. And I say that because, like every other medium where art happens, we’re sampling things that have happened before, we’re making transformative works, we’re building on what happened prior, and we’re producing new and imaginative creations. [We’re] making content that evokes emotions, that has its own beauty, that is a marvel in and of its own right. I very much consider what we show as art.

OR: One of the biggest donation driving-runs at Games Done Quick events is when they run Super Metroid, and one of the donation incentives comes into play at the very end of the game. So I have to ask you: Save or kill the animals?

AC: TASBot has, at every event, killed the animals. And at one event, TASBot played Super Metroid and in fact the audience voted to save the animals, and [so] we did. But TASBot was sad.