13 October Blog

Cyberpunk Revolution – How to defeat the Artificial Intelligence (AI) Bot Overlords

DISCLAIMER:  As a lifelong technologist and security researcher, I take security very seriously and that the following techniques are from the interest of improving security and promiting awareness of vulnerabilities, and do not condone using any of the methods.

Artificial Intelligence (AI), a key driver of innovation for many years, has spurred countless improvements time and again, both at the bedsie and on the administrative front.

Self-driving cars, diagnostic machines in various medical applications, voice-activated homes, facial recognition are just a few examples of the technology that utilizes AI.

In some instances, Ai can identify more superior solutions than human-based decision mediums. That is because AI can identify complex engineering processes and deep patterns in large amounts of data that the human mind cannot. But while AI results in well-calculated solutions, the potential for sabotaging AI is very real.

Before AI can interpret information, the data presented to them must be faultless and should be coherent with how it was programmed for the system to read. Thus, the presence of tiny sticker triangles in a stop sign could hinder the leading pattern-recognition technology AI in a self-driving car from reading it correctly. There are many other ways in which AI bot overlords can be defeated through simple human manipulation. Anyone who has played with machine learning knows that the AI can make hilarious mistakes once in a while. In this article, we are going to dig deeper into how to defeat AI.

Camouflaging your car from adversarial attack

We all know how artificial intelligence has been using convolutional neural network-based detectors to detect objects in the wild. It seems that with this AI overlord, there was nowhere to hide against adversarial attacks until a new experimental study was conducted to hide cars from object detectors. 

Object detectors are high-end technology that detects all types of materials, including cars in the wild, but they can be disrupted using camouflage. The study aims in creating an individual adversarial patch that covers the planar parts of the vehicle’s surface. As result, it hides the car whenever there is an adversarial attack. Experiments have also shown that the 3D camouflage cannot only hide your car from the image-based detectors but also in different environments. 

In this example, a Toyota Camry CLE was camouflaged to escape detection and yielded an affirmative result. On the other hand, neither the plain-colored nor randomly camouflaged Camry escaped detection. The main goal is to disrupt the correct measurement and detection algorithm of object detectors using a specific camouflage cover, so they cannot find the object.  

AI algorithms and adversarial attacks continue to pose a very tangible threat to national security. When used against adversarial attacks during military strikes, camouflaging vehicles could be one way of hiding from enemies. 

Defeating facial recognition through a simple make-up

Facial recognition is everywhere. Imagine a world where you’re being watched, scrutinized, penalized or reward for your actions. You walk downtown or cross the street and you see the green light, take a few steps backward, but too late, the camera has already captured your action. Since most of these cameras have facial recognition technology, you will have the ticket on your way in just a matter of days.


It’s not only that… Facial recognition is used in tons of applications like marketing agencies know when you went to the shop, how long you lingered, and how much you bought. Doors can open or lock depending on your facial features as soon as you present your face in front of security cameras.


It’s not hard to imagine because the capability of facial recognition software has grown technology over the last decade and the aggregation of data is going to continue flourishing in the next few years. But the system is going to be more integrated and stronger. The question is, can you escape facial recognition? Is there a way to defeat it?


According to WoonSook Lee, Ph.D., a professor at the school of engineering and computer science in Ottawa, facial recognition is “almost perfect.” But there are few ways to defeat it and surprisingly, the techniques are kind of so much simpler than what you can imagine. Recent studies have revealed that advanced facial recognition techs can be defeated using digitally and physically applied makeup. After all, there is a limit to AI facial recognition technology.

It’s a very simple method for bypassing facial recognition but quite effective. Several studies shave shown how facial recognition systems can be bypassed digitally by creating “master faces” that could copy other people’s faces. In one of the studies, the person was made to wear a sticker face and the facial recognition system was able to detect another person’s face.

The same logic can be applied to another machine-learning algorithm. Tweaking images of people can possibly confuse an AI that is designed to spot a human pose into thinking that the person actually assuming a particular stance. The image-based recognition can be led to believe they are seeing things that are not really there – convincing AI machines to see fake figures or people, and consequently unlocking safety and security systems.

Manipulating Artificial Intelligence Systems through “AI Attacks”

The methods sustaining the cutting-edge AI systems are systematically vulnerable to a new type of cybersecurity attack called “artificial intelligence attack”. Unlike ordinary malware and cyberattack, AI attacks are not caused by human errors. They are embedded in AI algorithms that suffer errors and cannot be fixed. This weakness makes them vulnerable to manipulation. For example, AI attacks can be used on physical objects. A stop light can be turned into green light in the eyes of self-driving cars by simply tapping it with green material.

This vulnerability caused by the inherent limitations of AI methods leaves them open to devastating and insidious attacks. Five areas are most susceptible to AI attacks – content filters, military, law enforcement, and civil society itself.

Winning over AI during a job interview

Artificial intelligence is increasingly being utilized by various countries to match jobs, analyze employment trends, track illegal hiring, and even screen employees. It can analyze the personality and decision-making of applicants and offer advice to consultants on the best candidates. Soon you’ll realize that it’s you versus the robot.


You’re no longer pleasing big companies but the machines in front of you. AI in interviews utilizes gamification to analyze different factors that affect the likelihood of an employee fitting into a particular position. It can get as specific as figuring 37 capabilities, including your personality, character, and conflict resolution strategies.

AI can analyze data by examining facial expressions and processing word choices. For example, rolling your eyes or moving too much and staring upward to think for answers may send negative signals to the AI that signifies you’re not prepared enough for the interview. AI can also detect the slightest movement in your body such as shaking, which could send signals that you’re nervous.

AI is like a shrink in the machine. If you know that an AI can analyze characters, then you should focus on your facial expressions and avoid forcing a smile. Another type of AI used to weed out employees is through an Applicant Tracking System (APS), intelligent software that can go through massive pools of data to dig for more information about your person.

One major way to win over this AI is to improve your presentation to the APS by streamlining your resumes. Keep formatting of your resumes simple, use the right keywords, get rid of career objects, and avoid using too many acronyms. Of course, don’t forget to run a grammar and spell check.

Meanwhile, the book by Shlomo Kashani called “Deep Learning Interviews” has presented hundreds of fully solved job interview questions from a wide range of key topics in AI. The book is specifically for job-seekers who want to defeat AI bots during the interview and pass the system. Here, you can learn how to speak confidently and quickly on any relevant topic that AI bots are programmed to ask and discuss. This is a powerful, indispensable advantage to have when walking into the interview room.

Using electronically modulated voice on voice-activated smart homes

artificial intelligence in voice recognition apps

One of the coolest advancements in home automation is the growing integration of voice control with all kinds of smart devices. It’s hard to beat the convenience of voice-activated technology like Amazon Alexa, Apple HomeKit, or Google Assistant.

These types of AI are programmed to execute a sequence of commands by listening to and measuring your vocals. Albeit voice-activated AI performs commands using applets from “if that, then this” formula – such as “if you hear this voice command, then switch the lights on” – it is not ultimately fool-proof. Similar-sounding voice commands can send recognizable signals to the AI, disabling alarms and opening doors.

Nowadays, many methods can change your voice electronically so that you will sound like someone else’s. Some software like Voxal Voice Changer or MorphVox can modulate audio data and change the frequency of audio signals to make your voice sound completely different. Recording your morphed voice and using specific voice commands can easily unlock voice-activated AI.

In some many ways, voice is more potent and valuable than DNA and fingerprints since it doesn’t only carry information about the speaker, but it also informs the receiver of the person’s current state and their surroundings. Unlike humans, AI only decipher information based on what they’re engineered for. That means, if the AI is made to recognize a specific voice pattern, then the same pattern could be replicated to manipulate a voice-activated machine. Using electronically available tools, one can easily change their voice and make it sound like another person. To defeat an AI it only takes voice-morphing machine that doesn’t even require payment for use.

However, in more complex world, while researchers are trying to resolve the flaws of neural networks in Artificial Intelligence, several researchers have also studied how to profile humans from their voices. In addition to DNAs, voice profiles collected through various voice recording devices and other data are being experimented in using human voice as evidences driving forward litigations and various other cases in a myriad of unprecedented ways.

Beating AI in patient diagnosis through a doctor’s secondary analysis

The medical world experienced groundbreaking transformations in the past 20 years to include AI in its medical processes. Many hospitals, nowadays, are using AI’s advanced software to detect diseases and diagnose patients by reading deep layers of data. As a result, AI helps save time between tests and treatments.

Recent advances in AI have been applied in pathology as well, through computation algorithm and slide imaging platform. The AI works in transforming pattern recognition and image-based interpretation to make diagnosis and prognosis to further aid or enhance the judgment of pathologists.

IN 2016, an AI-programmed computer only took 10 minutes to analyze and diagnose a 60-year-old patient to have a rare type of leukemia. AI has topped doctors in various ways such as the capability to have speedy health evaluations and precision in their diagnostic method.

However, AI in its medical process is not ultimately foolproof. A benign tumor could be misclassified as a malignant one, leading to unnecessary treatment for the patient. To defeat AI, secondary or confirmatory tests must be done by medical practitioners. Diagnosis from AI machines should only be used as a preliminary diagnosis and a second eye before a more precise diagnosis by doctors themselves.

 

Conclusion – Defeating Artificial Intelligence

When most people hear the term artificial intelligence, the first thing that comes to mind is robots. That’s because big-budget films weave stories about humanoids wreaking havoc on earth. However, artificial intelligence is based on the principle that human intelligence can be demonstrated in a way that machines can easily mimic and execute tasks. Researchers and developers in the field are making surprisingly rapid strides in creating machines that are able to mimic humans in learning, reasoning, and perception. It is not necesasrily a humanoid or a “terminator” robot that hunts humans in a cyberpunk world. It is more on how technology is being developed to let machines perform more-human skills to help businesses and companies offer more advanced tech solutions and make processes easier for humans.

Our team has established various solutions to integrate AI into your unique business needs. Additionally, we have vast experience in customizing software development and architecture unique intelligent software and apps for your business.

Our multi-domain expertise allows us to cater to all kinds of needs for technological advancement in your organization, and develop intuitive software that makes use of Artificial Intelligence, cognitive computing, and predictive learning. We understand how critical AI can be for every business. Find out more about our revolutionary data science and technology solutions.