SMAC: STEPN Model for Anti Cheating

Introducing STEPN’S Anti-Cheating System — SMAC

The word ‘cheating’ in any context carries negative connotations. And why should it not? To cheat, by definition, is to deprive a person of something valuable through the use of deceit or fraud. In traditional video games, cheaters deprive normal users of in-game resources, assets or positive gaming experiences. It is rampant, and frustrating for those who are playing fair.

For Play2Earn games where money is involved, solving cheating and closing the door on bad actors is of utmost importance. Given how cheating potentially undermines carefully crafted game ecosystems, developers work hard to take anti-cheating software to the next level. A significant amount of resources are dedicated to combating cheating.

Further, it is natural that when an app gets popular, many are tempted to abuse or exploit the system, causing developers to have their hands full with detecting and banning those who cheat.

STEPN is no exception — as it explodes in popularity, cheating is becoming a prominent problem, so much so that our developers have devoted nearly half of our resources to developing anti-cheating mechanisms.

Fake users and cheating?

A fake user is an account which fakes the user identity and motion data in an attempt to gain unjust profit from the STEPN app.

And simply put, those who cheat in online games are those who manipulate the mechanics of the game to gain an unfair advantage, subverting the rules for their own benefit. This involves engaging in actions that aren’t a part of standard gameplay, making it easier for the cheater to achieve what they want — without dedicating the hours of effort that other players have.

There are numerous telltale signs that a player may be cheating. For example, a player may be reaping a massive amount of rewards out of nowhere or displaying patterns that reveal the use of illegal technologies to boost their gameplay.

For STEPN, there are unfortunately people who are engaging in all sorts of dishonest tactics in order to earn more than their fair share of earnings. As a STEPN user, you might think — alright, I see that it’s annoying and unfair when players cheat to earn more than they should be earning, but…is it really that big of an issue?

The long-drawn battle against cheaters

The answer to the above question is a resounding yes. This isn’t the first time that Web3 earning games have faced this problem.

Axie Infinity too has attracted millions of gamers since launching in 2018. Many in Southeast Asia are able to earn a real living wage and even buy new homes through their earnings. But at the same time, the game has also attracted a significant population of bad actors that try to game the system.

One example of cheating is win-trading. Win-trading is a cheating mechanism that has become commonplace in games with ranked leaderboards. It occurs when a player will purposefully lose a match so that the winning player can raise their rank and reach a higher position on the leaderboards.

This has even occurred outside of play-to-earn games in online battle arena games such as League of Legends, and has a toxic effect on the game as a whole. Stefan, a gamer and online writer, wrote that “Win trading is rightfully the most toxic act in League of Legends. It destroys the solo queue experience for many players and helps rank up people who don’t really deserve it.”

With respect to Axie Infinity, win-trading is also becoming a massive issue for the game. One gamer noted that the practice would “ruin the competitive integrity of the game. The leader board will be illusory and would be unfair to the people who grind their MMR in the LEGITIMATE way.” Recognizing this, Sky Mavis (the developer behind Axie Infinity) has taken steps to investigate and combat this form of cheating.

But even beyond win-trading, there are many other forms of cheating that the developers must fight against. This year alone, Sky Mavis also uncovered and banned over 30,000 Axies from its game over accusations of energy abuse, where gamers manipulated the system to maximize their energy while gifting Axies to third parties. And in 2020, it banned SLP farms to prevent any one person from playing on five or six different accounts simultaneously.

Zooming Out

In the past decade, big competitive online games have scaled up their anti-cheating operations significantly — experts claim that the overall marketplace for cheats has grown to a whopping $100 million.

To name just a few examples, leading video game publisher Activision has banned over 500,000 cheating “Warzone” accounts on Call of Duty while Bungie (behind Destiny, Halo, Myth, and Oni) has filed lawsuits against sites that sell cheat software. Meanwhile, Ubisoft acquired anti-cheat company GameBlocks last year for an undisclosed price, and Epic Games acquired Easy Anti-Cheat in 2018.

Why is Cheating Serious?

At best, cheating annoys other gamers and causes conflicts and disparities of earnings. But at worst, rampant cheating can actually undermine a game’s entire economy.

Fairness is one issue, but when cheaters affect other players, it becomes a whole different ball game. Those who cheat in STEPN to milk the game of money are hurting the platform’s tokenomics. This comes in the form of affecting the liquidity of GST and the value of GMT, since cheating actions artificially inflate and then deflate token value when cheaters rapidly earn large amounts of tokens and cash out immediately. It also warps STEPN’s sneaker supply, impacting STEPN’s longevity as a whole. Simply put, the actions of a few can throw a carefully designed ecosystem out of tilt, with burgeoning consequences.

In addition, cheating is squarely against our ethos. Players who are in the game simply to hack the system are not aligned with our mission to encourage people to go outside, connect with other walkers/runners, and develop healthier habits and a more active lifestyle.

For the reasons stated above, it is imperative that we undertake strict action against cheating. We have seen how it can compromise the integrity and tokenomics of a game, which more than warrants dedicating significant resources to implementing anti-cheating measures.

Utilising AI to combat cheating

Our Artificial Intelligence (AI) team has now spent months studying patterns and have built a world-class anti-cheating system. It has been based off self-learning algorithms trained on data across the board — including GPS tracking, motion sensor and health data to detect anomalies.

Anomaly Detection is the technique of identifying rare events or observations which can raise suspicions by being statistically different from the rest of the observations. While multiple deep learning techniques exist, STEPN has utilised Autoencoders to piece together the data.

What are Autoencoders?

Autoencoders are basically an unsupervised learning technique that consists of neural networks, which are made up of multiple layers of neurons. Autoencoders discover low-dimensional representations of high-dimensional data and then use this to reconstruct the input.

It consists of 3 main layers:

Encoder — Reduces dataset from high-dimensional to low-dimensional.

Bottleneck — Contains the reduced representation of the dataset.

Decoder — Reconstructs the dataset from low-dimensional back to high-dimensional.

Essentially, autoencoders are fed data sets. The encoding process compresses this input to spit out the low-dimensional representation of the data. The decoding process reconstructs this data to produce the outcome.

The real fun happens in the middle — the bottleneck. To ensure only essential information is extracted, the number of neurons in the bottleneck layer has to be decidedly less than that of the Encoder layers. This forces the bottleneck section to most effectively learn the patterns of the data and ignore the “noise”. Otherwise, if given too much capacity to learn, the bottleneck layer will end up extracting too much non-essential information.

Why Autoencoders?

Autoencoders are an incredibly efficient technique to not only identify anomalies, but also help identify the variables that caused the anomalies. Through learning what is deemed as “normal behaviour”, autoencoders are able to detect when an abnormal input is parsed through it. In its inability to then accurately reconstruct the data to match the original, it is able to highlight whenever an anomaly occurs.

To use an illustration, let’s say your average walking speed ranges from 4 to 6 km/hr. If you were to suddenly start moving at 15km/hour, that will be flagged by the autoencoder as an anomaly.

But how does the system determine a genuine anomaly from one that is fraudulent in nature? Simply put, by fixing a specified threshold value for reconstruction errors, and then cross referencing it across multiple autoencoder data points.

Continuing from the above example, if the data points for fraud identification include data headers like proximity to another STEPN device (< 30cm), duration of proximity to another STEPN device (<5seconds), and movement speed (15km/hr variance), we can cross reference the data accordingly to make the conclusion.

While the actual anti-cheating mechanism is a lot more intricate than this, this should give you a general idea on how autoencoders enable successful weeding out of cheaters.

Presenting STEPN’s SMAC System

After three months of training our machine learning algorithm, we present to you — STEPN’s Model for Anti-Cheating (SMAC) — which we are confident to proclaim will be the best-in-class anti-cheating system.

User’s running data are cross-referenced with our benchmark standard to check global, contextual and collective outliers at the end of each of their sessions. If the SMAC System detects an anomaly, the user will be flagged as cheating and all the reqards for the session will be erased.

We may further limit the user’s in-app functions including but not limited to — longer cooldown for shoe-minting, inaccessibility to in-app marketplace and slow down on energy refill.

SMAC system specifically targets the movement simulation by amending real walking/running data, thanks to our machine learning algorithm.

Cheating software video found on Youtube

Shoe-minting/Levelling by Script

SMAC System also detects any form of scripts such as buying/selling bots on our marketplace, minting bots or levelling bots. Once detected, the system will disconnect these bots from the app.

Minting bot video found on Youtube

Moonwalking

It is important to distinguish between moonwalking and cheating activities; moonwalking occurs when the GPS signal is poor or the system cannot detect valid motion data.

The moonwalking warning is meant to warn users to be aware of the situation and address it promptly. There are no repercussions for moonwalking.

SMAC system will kick in after the users long-press the STOP button and before going into the result page, SMAC system will analyse the users’ motion data and if an anomaly is detected, there will be repercussions.

Turing Score

In subsequent updates, STEPN will introduce the Turing Score (TS). Users will start with a score of 100/100. At the end of each session, the system will automatically add or deduct a user’s score. Until the tabulation is complete, users will not be able to start a new session.

When a user’s TS falls below 100, they are unable to interact with the in-app marketplace, and transfer between spending and wallet accounts will be suspended. In order to rescind this suspension, users can still go about their sessions, and if no cheating is detected, can slowly regain points to their TS.

For instance, if users are found to be multi mining (i.e. carrying multiple phones with the STEPN app opened), their earnings for the session will be nullified and a specified amount of points will be deducted from their TS. On the other hand, if a user moves regularly without cheating, their earnings will not be affected and they will receive additional points to their TS.

Conclusion

With STEPN’s model, it is not hard to see why long-term sustainability is usually one of the first questions raised. The longevity and stability of the game are at the forefront of the team’s priorities, evidenced by the focus on crafting robust tokenomics.

While updates on the anti-cheating front may not be the promises of wealth that some ask for, it is imperative that this issue be addressed. A high level of investment in anti-cheating infrastructure is to be expected. This is to ensure fair and equitable gameplay for users and to prevent cheaters from throwing the game off balance for the entire community.

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store