Artificial Intelligence 2022

From New Media Business Blog

Jump to: navigation, search



What is AI?

Here are a few different definitions:

“The ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings” -Britannica [1]

“Artificial intelligence leverages computers and machines to mimic the problem-solving and decision-making capabilities of the human mind” -IBM. [2]

“The theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision making, and translation between languages” -Oxford Language[3]

The general definition of AI is how computer systems perform human-like tasks. AI ranges from simple things such as speech recognition like Siri to more complex things such as a Roomba.

3 Types of AI [4]

Artificial Narrow Intelligence (Weak AI): Artificial Narrow Intelligence, also known as weak AI, is made to perform specific tasks. This is the only type of AI we see around the world today. Some common types of weak AI are speech recognition (Siri), customer service (chatbots on company websites), computer vision (self-driving cars), and recommendation engines (algorithms on social media).

Artificial General Intelligence (Strong AI): Artificial General Intelligence, also known as strong AI, is what the original goal of AI was, which is to have machines act the same as humans. This is the type of AI we see in movies where robots are working and thinking like humans, for example, C-3PO from Star Wars. Currently, there are no physical examples of strong AI, it is still in its developmental phase. We are seeing new inventions that are trying to become strong AI, but nothing has succeeded to date, it is so far only a concept.

Artificial Superintelligence: Artificial Superintelligence is the ability of machines to perform better than humans. However, this is not commonly used as this is only a theory presently. Currently, we are in the primitive stages of developing strong AI, which would need to be accomplished before even attempting Artificial Superintelligence.

Machine Learning vs. Neural Networks vs. Deep Learning [5]

Neural Networks: Before looking at machine learning and deep learning, we need to discuss neural networks. Neural networks are like our brain's networks, which help machines process data the way brains do. Neural networks help a machine classify data so that it can learn from the data classification. Neural networks are made of stacks of layers consisting of nodes. The neural networks are made of one input layer, followed by one or more hidden layers, and then an output layer [as in the image shown below] File:NeuralNetworks.png

Machine Learning: [6] Machine learning is the way machines can learn and process information. Machine learning is how machines use data to learn and grow. There are two types of machine learning: supervised and unsupervised learning. Supervised learning is when a machine does a process and requires human intervention when that process is wrong. In supervised learning, the machine learns from the inputs of the human and not by itself. In unsupervised learning, the machine learns from itself rather than through human intervention.

Deep Learning: Deep learning is a type of machine learning, where a machine learns from neural networks. If there is an error in its process, it can correct itself through the use of neural networks and further understand. Neural networks classify data, the machine uses the classifications from the neural networks to learn from itself.

History of AI[7] [8]

1950: Turing Test: Created by Alan Turing, the Turing test was a test to try and identify if a machine was capable of demonstrating human-like intelligence. In the test, a human evaluator would blindly judge text-only conversations between a machine and a human. If the evaluator could not differentiate between the machine and human, the machine passed the Turing test.

1956: Logic Theorist: Known as 'the first artificial intelligence program' Logic Theorist was the first program created to deliberately mimic the problem-solving skills of humans through mathematical formulas.

1958: Lisp: Still commonly used today, LISP is a high level programming language. It is one of the first programming languages used for AI development. It was developed by John McCarthy.

1967: Mark 1 perceptron: The Mark I perceptron, developed by Frank Rosenblatt, was the first computer that used neural networks to gain knowledge through trial and error.

1995: ALICE Alice was a chatbot program that was designed to engage in normal humanlike conversations with other humans.

2000: KISMET & ASIMO Designed by Cynthia Breazeal, Kismet was a robot head that could identify and simulate human emotions through facial features. Asimo was a robot released by Honda that could move around like a human.

2007: ImageNet Created by Fei-Fei Lang ImageNet was a website that organized images hierarchically. “It caused a paradigm shift in thinking about AI, while a lot of people are paying attention to models, let's pay attention to data" – Li [9]

AI in Government Administration and Politics

Artificial Intelligence (AI) and machine learning have revolutionized significant fields that matter to an average human on a day-to-day basis, for instance, healthcare, entertainment, etc.; however, there is a field yet to have a powerful influence from AI, Politics, and Government Administration. With the technological advancement in AI, scholars have started talking about the impact of this thriving technology on domestic and international politics. While governments have always been slow to adapt even the simpler technologies, the introduction of AI in administration can change the traditional forms of policymaking, service provision, and enforcement. With several human hours being wasted just fulfilling administrative needs, AI may help the government to focus on things that matter, as stated by Deloitte University Press in 2017 (Source:[10]).

Reshaping Governance

Governance, as a field, is always seen as an ever-evolving and dynamic sector. Leaders around the globe are looking out for modern methods to decode civil problems, enhance the constituency experience for the citizens, and boost governance. AI can process quintillions of data; therefore, it enhances the system to deepen insights into public needs. Data-driven policy making will create an all-new citizen experience by only filling the buckets that need water. For instance, if a man goes through an accident and is disabled to work, he need not apply separately for food and unemployment assistance, or other government services, since AI can create a single entry point connecting people with all appropriate administrative services in an AI-powered future state (Abillama, 2021)[11]. Furthermore, AI-based digital platforms are used to engage with the public, get their feedback, record their demands and opinions, and their existing level of information regarding political affairs.

Applications in Administration

AI is making taxpayer-supported organizations more accessible and practical. Regardless of the technological advancement, governments, with the help of AI, are trying to make public services more competent and available and making the best use of the available resources and budgets. The following services have seen an increase in the use of AI in Public administration:

Fraudulent Claims: Governments offer continuous support to their citizens through EI, CERB, etc., but the scariest aspect of the support program is the number of fraudulent claims that cost governments billions. AI-enabled fraud detection can help public departments scale down the corruption of benefit programs by looking for patterns in claims such as previous claims and their reasons, phone numbers, and the application format (Dilmegani, 2022). For instance, Italy's insurance sector uses a risk-based triage approach to minimize whiplash claims by verifying the consistency of the reasons for the accidents, medical testing results, etc. (Clancy, Duranton, Mills, and Whybrew, 2022).

Healthcare: Data analysis through AI's learning algorithm can be used to prevent disease spread. Healthcare admins can analyze patient data to predict the risk factors so doctors can prioritize their choices of the sector to work in. China, during the COVID-19 outspread, was known to utilize the travelers' data to identify the possible carriers of the virus.

Optimizing Policing and Scaling down crime: AI algorithms can be used to explore the patterns in criminal activities and identify optimal police patrol presence. Governments have lately used AI-based surveillance to analyze images, videos, and data from CCTV. Although the ethical side of AI-powered surveillance is still controversial, features like facial recognition have enabled governments to track down a few criminals (Dilmegani, 2022)[12].

Military: Military weapons like UCAV (Unmanned combat aerial vehicles), also called autonomous military drones, contain a certain level of AI autonomy while the real-time control is still human. The latest examples include Azerbaijan using military drones against Armenia ((Dilmegani, 2022)[13].

Education: Education is one of the most important sectors for any nation. Machine learning algorithms can assess the personal needs of the students, regardless of the number, and detect any misalignments between what is offered and how it is transcribed. For instance, the University of Michigan provides students with immediate feedback on their writings, thanks to AI (Dilmegani, 2022)[14].

Public Relations: AI-based software is used to address common public issues through Chatbots, which not only answer FAQs but also help direct requests to appropriate departments. It enables the government to save human working hours, and AI helps collect data for future decision-making.[15]

Law, Politics, and AI

Let's consider an ethical approach to AI. It helps promote equality and asks people to participate in moral debates equally about the administration, thus helping in policy formation. The most standard form of AI in law is e-discovery, the process in which electronic information is scanned to obtain non-privileged information relevant to a case or a claim (Kern, 2021) [16]. Democracy thrives when the citizens' preferences are instilled in the judicial mechanism, and the data helps to create a structure that further helps to fulfill the public needs. Still, this process is not a one-way process. Democracy also means that the users of the AI are transparent while using the information, providing tracing ability to the public about the political decisions, upholding accountability, and society's inclusion in the legislation. So far, AI has just been visualized as an aid to Law enforcement rather than a law-making body. But there has also been another chain of thoughts that revolve around creating an AI-based political world where AI makes our decisions instead of bureaucrats. Hence the existing system, whereby civil servants are held responsible for their choices, will not operate in a sense that AI governance will be accountable for the political decisions taken by AI systems in the AI-based society (Zekos, 2022) [17].

On the contrary, it is argued that AI-based algorithmic governance will cause damage to democracy as AI systems might only allow electronic participation of AI entities, leaving normal citizens behind, thus creating a void in the system and harming all forms of democracy. Also, some scholars argue that AI lacks human qualities such as empathy, wisdom, and compassion, which are significant for making many human decisions. Moreover, the AI system employed in the current political state can add transparency, accountability, and impartiality, as long as it does not emerge as a replacement for the conventional political society for the reason that we still look for conventional human behavior in the application and the results of the administrative decisions.

Judiciary and AI: Can human judgment be replaced by AI?

Canadian Prime Minister Justin Trudeau shaking hands with a robot.

AI within the court refers to AI systems that support administrative parties and judges in their judicial capacity. AI helps judges to arrange and choose relevant jurisprudence, detect patterns in jurisprudence, or produce arguments to be used in court decisions (Zekos, 2022)[18]. For instance, the Estonian government in Europe has been developing an AI system that can take decisions on cases with little to no human intervention, surpassing human speed and saving time for the judges to work on more complex and crucial issues. Nonetheless, AI acting as a judge in all cases is still an idea miles away from reality due to technical problems, lack of human emotions, legal adaptations, political will, and ethical considerations. Presently, courts not only make decisions based on law, but they also construct law, and they do so in a way that they can alter the law according to the uniqueness of a situation. An AI humanoid judge may make decisions with much more capacity for data analysis, standardized operations, and a codified law system but cannot adjust the findings according to the exclusivity of the circumstances.

Future of AI in politics

It is worth noting that predictive AI tools are already being employed in the administration to make everyday decisions about the public, ranging from small claims cases, criminal sentencing, or budgeting the judiciary system, thanks to the data processing capacity of AI. AI promises considerable growth through global cooperation and the capability to automate or improve the speed, accuracy, and track of the decisions of humans through digitalization. But when it comes to AI taking the complete role of a law-making society or replacing our conventional politicians, the day seems too far from existing. Like it or not, we still need humans as our politicians to make decisions for us, on our behalf, for the decisions to have human consequences and be rectified by human reasoning. Even though computer systems are now capable of reconfigurations, they lack self-conscious awareness, a sense of pain, empathy, and humanity. AI politicians are not the future of politics; hybrid decision-making is (Nilsson, 2019) [19]. Even though this doesn't attract attention, hybrid is far more achievable, practical, and reaping.

AI in Healthcare

As the world’s population continues to increase, health care will continue to be a critical sector. As seen from the impacts of the COVID-19 pandemic, the lack of supply of available medical professionals impacted an already strained global healthcare industry. When adopted appropriately, AI could play a prominent role in transforming both the present and future of the industry and could also mean the difference between life and death for some end-users.

Importance of AI in Healthcare

The healthcare industry is typically slower to adopt digital transformation due to high-risk levels and the complexities of tasks. However, according to surveys conducted by McKinsey, we could see the following three phases of scaling of AI in the healthcare industry:

  • Repetitive and administrative tasks (e.g. imaging) that require the time of healthcare professionals will be transformed by AI [20].
  • Increased usage of NLP in hospitals and at home; patients begin to take more ownership of their healthcare with hospital-based care being shifted to home-based. They may use solutions such as AI-powered alerting systems, remote monitoring, or virtual assistants. [21].
  • AI begins to be adopted in clinical practice. In this stage, AI will be a key factor in the healthcare system, helping the industry learn, analyze, and implement direct change to improve lives. [22]. Usage of AI technology in healthcare has proven to result in a direct increase in both efficiency and productivity for both healthcare professionals and end-users (patients). AI does this through machine learning algorithms and deep learning. These processes can recognize patterns in behavior and create their own logic. To gain useful insights and predictions, machine learning models must be trained using extensive amounts of input data.

    Healthcare Today

    While we are still leaps and bounds away from the desired outcome with AI usage in healthcare, there are presently several ways AI is shaping and improving the healthcare industry.


    Usage of AI alongside the Internet of Medical Things (IoMT) in consumer health applications is already making breakthroughs and helping consumers daily. Healthcare applications foster healthy lifestyles and put users at the forefront of their well-being.

    By analyzing the trends and needs of patients, AI helps healthcare professionals better understand their patients and provide better support for healthy living.[23].

    Early Detection

    According to the American Cancer Society, 1 in 2 healthy women have been told that they have cancer due to high rates of false results from mammograms. AI is currently helping transform the understanding of mammograms 30 times faster with approximately 99% accuracy, reducing the need for multiple unnecessary biopsies.[24].

    Additionally, AI usage alongside consumer wearables is helping doctors and caregivers to monitor life-threatening heart-disease episodes at a more treatable stage to eliminate fatality rates. [25].


    AI has also been enhancing the diagnosis of health concerns globally. Examples can include Google’s DeepMind Health, which is currently working with clinicians using machine learning and systems neuroscience to build general-purpose learning algorithms into neural networks that mimic the human brain. IBM’s Watson for Health is also transforming the industry by facilitating the storage of medical journals, symptoms, and case studies faster than any human.[26].

    Google DeepMind application

    Decision Making

    Using pattern recognition and predictive analytics, AI has enhanced decision-making in the healthcare industry for both complex and administrative tasks. AI has also been used to identify at-risk individuals who may develop a condition due to different factors like lifestyle or the environment[27].


    Robotics have been utilized in the healthcare industry for over 30 years. This technology includes laboratory robots to complex surgical robots that can assist a surgeon or conduct an operation by themselves. Robots have also been used in hospitals and labs for administrative tasks, rehabilitation, and in physical therapy to support long-term conditions.[28].

    End-of-life Care

    Robots have also been used to revolutionize end-of-life care to reduce old-age conditions such as dementia. AI combined with humanoid design enhancements has helped robots hold social interactions with older aged individuals to keep their minds working and sharp[29].

    Research and Development

    According to the California Biomedical Research Association, it takes approximately 12 years for a drug to travel from the research lab to the patient. Additionally, only five in 5,000 drugs, that begin preclinical testing, make it to human trials, and just one of these five is ever approved for usage. Furthermore, it will cost a company US $359 million to develop a new drug from the research lab to the patient.[30].

    By directing AI advancements to focus on research and development, time to market and costs can be reduced.


    Training is a repetitive task that has seen some use cases by AI. AI computers can draw from large databases of questions and scenarios to challenge a trainee in ways that a human would not be able to. Challenges can also be adjusted based on learning needs and familiarity level to ensure that the individual received the appropriate training or refresher courses[31].

    Clinical Application

    Although AI has not completely replaced human expertise, medical professionals may already have access to this technology to a certain extent to guide and provide a supplemental resource for making key decisions.


    Presently, AI has been applied to cardiovascular health in four key ways: precision medicine, clinical prediction, and cardiac imaging analysis.

    With precision medicine, AI has been helping patients be fully aware of medication reminders, disease counseling, remote follows ups, and detecting early warning of symptoms. On the clinicians' end, AI has helped collect a large amount of information and reduce administrative workloads.

    AI has also leveraged machine learning and data analytics to make more accurate decisions for patients through the analysis of large amounts of data. Research done by Dawes TJW found that AI could help predict the time of death for heart disease patients by monitoring MRI scans and blood tests of patients and identifying abnormal conditions that could lead to death.[32].

    Deep learning has also been utilized to analyze coronary angiography, echocardiography, and electrocardiogram (ECG) to provide clinicians with better advice and historical data for decision-making and treatment plans.[33].


    AI has been utilized in a number of use cases across the radiology and oncology sector.

    AI has been clinically applied in multiple radiology and oncology tests. Applications can include thoracic imaging, abdominal and pelvic imaging, colonoscopy, mammography, brain imaging, and radiation. With all of these listed cancers, AI has helped identify both benign and malignant tumours through pattern detection, allowing clinicians to speed up diagnosis and follow-up processes[34].

    Google has developed an AI dermatology assistant, allowing users to upload photos of their skin to detect possible abnormalities


    AI has been utilized to determine possible skin conditions. We can see this with Google’s launch of an AI-powered dermatology assist tool in 2021, which allows users to simply take a photo using their phone’s camera to take a photo of the skin at different angles. Thereafter, the application prompts the user with questions about the conditions and duration of symptoms, and the AI model beings to analyze the information and determine possible conditions from its database[35].

    Melanoma detection remains the most successful application of AI in the dermatology sector thus far. Some cases of AI detection have surpassed that of human dermatologists. Since dermoscopy images are more standard than digital pictures of the skin, machine learning has been successful at identifying lesions and causes of concern[36].

    Future and Implications of AI in Healthcare

    Human Alternative

    The use of artificial intelligence in the healthcare industry can perform just as well or better than humans in certain procedures, like diagnosing diseases. An implication of AI is that there is a possibility that some jobs can be transferred over to artificial intelligence over the next few years. With algorithms continuously improving, AI can help process all the information about our health, lifestyle, and the environment we live in, which can help give suggestions on how to live a healthier and longer life.[37]

    Clinical Trials

    In the future, many businesses and labs will move away from traditional controlled experiments and toward more virtual experiments for research and clinical trials. Therefore, AI will take on the difficult work and tasks in clinical trials that used to require costly human efforts. AI will also shape the future of pharmaceuticals by improving candidate selection processes for clinical trials. By quickly analyzing all patients and identifying the best patients for a given trial, AI can help provide trial opportunities to the most suitable candidates within the candidate pool.[38]

    AI being utilized in hospitals

    Big Data

    AI systems can ease the strain on researchers, doctors, and patients while also assisting in our overall health and decision-making. The ability to comprehend the enormous volumes of scientific data that have been gathered over the years of research. Therefore, with the help of AI being able to process big data, it will ultimately benefit medical professionals to pinpoint accurate medical information. [39]


    One of the greatest implications of AI in the healthcare industry is whether the technology will be adopted to help everyday clinical practice. In time, many medical professionals may move towards the use of AI to increase efficiency within their practice; however, there will be a few that will refuse to adapt to this new system. The healthcare professionals who will lose out on the full potential of AI maybe those who refuse to work alongside it as they believe that the change is not needed. Therefore, communicating efficiency and change may help increase the adoption of AI within the healthcare industry. [40]

    Benefits of AI in Healthcare

    AI has undoubtedly benefited the healthcare industry. A few examples are that AI provides real-time data, increases efficiency, assists medical research, and streamlines tasks.

    Provide real-time data: An important part of diagnosing and resolving medical issues is getting correct data as quickly as possible. With AI, specialists and other clinical experts can leverage immediate and precise data to assist and improve medical diagnoses. This will lead to fast and accurate results that can lead to improved preventative steps, cost savings, and shorter patient wait times.

    Increases Efficiency: With AI automating more of its critical processes, medical professionals now have more time to thoroughly examine patients and make accurate diagnoses. AI is speeding up processes to save medical experts valuable productivity hours. Thus, Ai can reduce expenses significantly and utilize medical experts' time more effectively.

    Assists research: AI assists medical experts in research by going through mass amounts of data from various sources. This enables medical professionals to quickly gather information from many different sources ultimately leading to an effective analysis of the illness.

    Streamlines tasks: With AI in healthcare, it allows more tedious and meticulous tasks to be streamlined. For example, intelligent radiology technology has been used to identify significant visual markers which resulted in saving many hours of intense analysis. There are also other automated systems in this industry like automatic appointment scheduling, patient tracking, and care recommendations. [41]

    Risks and Concerns of AI in Healthcare

    Although AI in healthcare has several benefits, there are also some risks and concerns that need to be taken into consideration.

    Medical Errors: AI heavily relies on diagnosis information available from millions of medical cases. A wrong diagnosis is possible when there is limited information available about a certain illness. The risk of medical errors is important when a patient is provided the wrong medication because of a medical misdiagnosis.

    Data privacy and security: Due to AI's reliance on numerous data networks, there is a possibility that AI systems could be vulnerable to cyber security threats. Cyberattacks will be common as the advancement of both AI and cyber crimes are increasing thus it is harder to forecast and avoid data being stolen.

    Data access limitations: Medical data is generally hard to access and gather. Medical professionals often resend data collected which leads to data often being incomplete. Without access to big, high-quality data sets, it can be challenging to create a reliable AI system. Thus, the adoption of the technology by healthcare professionals can be delayed.

    Automation of jobs: A recent study found that within the next 10 to 20 years, AI can replace up to 35% of jobs within the healthcare industry. However, it was determined that no healthcare occupations had yet been eliminated by AI. If AI were to automate healthcare-related jobs, occupations involving digital information, radiography, and pathology would be most at risk of automation, as opposed to those involving doctor-patient interaction. [42]

    Ethical Considerations

    Perhaps the greatest long-term challenge with AI application in the healthcare industry would ultimately occur when AI replaces human physicians. This can pose to be a huge liability risk in the future. Joseph Carvalko, BSEE, JD, chair of the Technology and Ethics Working Research Group at Yale’s ICB, stated that physicians may feel pressured to give AI full authority due to fear of being legally liable for overriding a machine that's considered to be the "latest" technology. [43]. Given that AI is also continuously learning and creating new algorithms, it is currently being debated whether the party that deploys the system should be liable, or if the machine itself should be [44].

    Unconscious Bias

    Although humans are generally prone to making unconsciously biased decisions, AI bias can also exist and pose to be a threat. The technology may skew data, resulting in the algorithm being inappropriate for larger amounts of data. This bias is often challenging to detect due to its systematic nature. [45]. Certain medical AI systems may contain biases that correspond accordingly to the geographic region or society in which their creators were educated; this might make the system challenging to be globally implemented. Give the complexity of AI, scientists can begin to ensure diverse datasets are being used but the ethical aspects of AI in healthcare is definitely worth keeping an eye on as time progresses [46].

    AI in Entertainment

    Artificial Intelligence has also seen applications in the entertainment industry. Below are just some examples of AI applications in various segments of the entertainment industry.


    Song Analysis

    HITLAB is a Montreal-based digital media and AI company that employs a Digital Nuance Analysis (DNA) tool to classify and predict the potential of a song’s chart success.[47] DNA utilizes a historical database comprised of millions of songs, both successful and unsuccessful, to identify and compare trends.[48] CEO Michel Zgarka states that “any song, in any language and musical genre, is made up of 84 mathematical parameters,” which DNA studies to predict future charting songs.[49]

    Vocal Models

    Vocal models are “a species of deep neural network” trained to mimic a specific voice.[50] Holly+ is a vocal model trained in musician Holly Herndon’s voice capable of synthesizing the vocal input of another user to match the vocal timbre of Herndon’s voice.[51] Holly+ was first demonstrated at the 2022 Sonar Festival in Barcelona, in which artist and researcher Matthew Dryhurst, Herndon’s husband, performed in front of a crowd using Herndon’s voice.[52]

    Music Creation

    ‘Lost Tapes of the 27 Club’ is a collection of AI-generated tracks in the styles of musicians who died at age 27, such as Amy Winehouse, Jimi Hendrix, and Kurt Cobain.[53] The project was produced by Over the Bridge, a Toronto-based nonprofit seeking to spread mental health awareness in the music industry.[54] The project utilized Magenta, an open-source AI research project, which analyzed a library of the respective artists’ songs to study details like vocal melodies, chord progressions, and lyrics to create an approximation of new compositions.[55]

    Music Personalization

    Music streaming apps such as Spotify implement AI models to enhance user listening experience. AI in the Spotify app collects and stores user information such as songs played, keywords searched, used devices, and frequently played songs among others.[56] Playlists such as Discover Weekly recommend new songs and artists based on recent listening activity,[57] while Blend creates a personalized playlist between a user and their friends.[58] The Discover Weekly playlist utilizes AI models such as Collaborative Filtering (recommendations based on user preference and behaviour) and Natural Language Processing (NPL).[59]

    Film and Television


    The script for 'Sunspring', a 2016 sci-fi short film, was written by an AI model named Benjamin.[60] Sunspring was a collaboration between filmmaker Oscar Sharp and researcher Ross Goodwin, who trained Benjamin with hundreds of sci-fi scripts from films and television shows.[61] In 2019, comedian and writer Keaton Patti trained a bot to write a Batman movie script based on 1,000 hours of Batman movies.[62] In 2020, media production company Calamity AI trained AI tool Shortly Read to write a screenplay for a short film called ‘Solicitors,’ which follows a Jehovah’s Witness telling his life story to an unwilling resident.[63]

    Decision Making

    Cinelytics is an LA-based startup that implements AI in the decision-making side of film production.[64] In real-time, the AI model can show film studios and executives how decisions with casting and scripts can impact a movie’s risk profile and revenue potential.[65] Essentially, Cinelytic acts as a business intelligence dashboard for filmmakers.[66]

    Content Recommendation

    One notable application of AI in streaming services such as Netflix is content recommendation. Using data collected from users’ watching habits and watching habits of viewers with similar tastes, Netflix can better suggest content that a user would find interesting enough to stay on their service.[67] This recommendation system continues to learn and adjust as more data gets collected.[68]

    Visual Effects

    Marvel Studios used AI to model Josh Brolin's facial expressions

    Growing demand for more content is quickly outpacing the number of available visual effects artists, who are now implementing AI to help automate time-consuming tasks such as motion capture tracking removal.[69] A VFX team for Marvel Studios’ WandaVision developed a neural network to quickly remove motion-tracking dots on actor Paul Bettany’s face during editing.[70] Another example of AI in visual effects is the making of Thanos’s character in Marvel Studios’ Avengers: Infinity War and Avengers: Endgame. Actor Josh Brolin was recorded to create a library of facial expressions and movements, which helped train a neural network.[71] This neural network allowed VFX artists to pull these facial expressions from the library and incorporate them into their 3D models.[72]

    Video Games

    AI development in video games has been relatively stagnant, with many developers sticking to predictable AI.[73] This is mainly due to worries that continuously learning AI opponents would eventually lead to unexpected and unpleasant experiences for the player.[74] Video game developers instead choose to expand the number of AI systems in a single video game, all still relatively simple but all interconnected and interacting with each other.[75]

    However, some AI research has started to bleed into video game development. Procedural Generation, recently popularized through the video game No Man’s Sky, is a method of data creation that utilizes AI algorithms.[76] This process is mainly used for level generation, but researchers are speculating on complete game generation through procedural generation.[77] Generative AI applications can also generate rendered game environments and faces.[78]

    Generative AI

    Generative AI is a branch of artificial intelligence that creates new content by taking from existing textual, visual, or auditory data.[79] Generative AI models seek to find patterns related to the inputted data in order to create similar content.[80]

    Generative AI Models

    Generative Adversarial Network

    A generative adversarial network (GAN) is one kind of generative AI model. Generative adversarial networks are comprised of two different neural networks: a generator and a discriminator.[81]

    The generator in a GAN is responsible for generating new data or content.[82] This new data or content is based on source data that the generator is trained on.[83]

    The discriminator in a GAN is responsible for differentiating between the source data and the new content that is generated by the generator.[84] The discriminator then provides feedback to the generator network to help improve future generated content.[85]

    The goal of a generative adversarial network is to create new data or content that is near indistinguishable from the source content.[86] Generative adversarial networks are semi-supervised, meaning that some aspects of this model are monitored by humans, such as manually labelling training data for the generator.[87] However, GANs also implement unlabeled data for unsupervised learning purposes to help make predictions beyond labeled data.[88]

    Generative Adversarial Network (GAN) Diagram

    Diffusion Model

    Diffusion models are another kind of generative AI model that has found success in applications such as image generation.[89] Recent diffusion models have been able to outperform generative adversarial networks (GANs) in use cases such as converting text descriptions to images, inpainting (completing images that have been erased in some areas), and image manipulation.[90] Diffusion models work by successively adding Gaussian noise to a piece of data (most often, images) in steps, and then learning to recover that data by reversing the noise process.[91] Diffusion models seek to generate new data that is coherent to the source data while starting from a canvas of Gaussian noise.[92] Diffusion models are implemented in research projects such as OpenAI’s DALL-E 2 and Google’s Imagen.[93]

    Diffusion adds Gaussian noise to an image, then seeks to reverse it

    Applications of Generative AI

    Face generation from existing celebrities

    Image Generation

    Generative AI models are capable of producing realistic photographs of faces, objects, and scenes.[94] In 2018, a study by NVIDIA experimented with GANs to find a new training methodology for generators and discriminators that would generate higher quality images with better variation.[95]

    Image-to-Image Conversion

    Day to Night

    Generative AI can translate one image to another,[96] such as:

    • Black and white photographs to colour
    • Day photographs to night photographs
    • Photo-realistic image to stylistic image
    • Satellite image to map image
    Photo-realistic to stylistic image

    Film Restoration

    Generative AI can upscale old images and films to high resolutions, by generating in higher frames per second, eliminating noise, and adding colour.[97]

    Face Photo Manipulation

    Aging and De-aging using AI

    Generative AI can manipulate photographs of faces,[98] such as:

    • Generating frontal face photos from side profiles
    • Turning face photos into emoji
    • Aging or de-aging faces
    • Deep fakes

    Text-to-Image Synthesis

    Generative AI can produce images from text-based prompts.[99] Notable text-to-image synthesis projects include:

    DALL-E 2

    Images generated by DALL-E 2

    DALL-E 2 is owned by OpenAI.[100] This AI model can generate both realistic and stylistic images from text, as well as create variations of original artworks.[101]

    Google Imagen

    Google’s Imagen uses a diffusion model with research focused on photorealism and language understanding.[102]

    Images generated by Google Imagen


    Images generated by Midjourney

    Midjourney is an independent research project operated by a small team of eleven researchers.[103] The Midjourney beta version can currently be accessed by the public through a subscription model and on their Discord server.[104]

    DALL-E mini

    ‘court sketch of Darth Vader on trial’ by DALL-E mini

    DALL-E mini is a free text-to-image model available for everyone.[105] Although relatively unsophisticated compared to previously mentioned models, DALL-E mini exemplifies the experience of using text-to-image software. Quality of artwork generated by DALL-E mini can vary depending on the prompt. The challenge for users to find the most outlandish prompts has allowed DALL-E mini to be adopted into internet meme culture.[106]

    Risks and Concerns

    Fraud and Security

    Generative AI could be used to scam people, steal their identities, or produce disinformation.[107] As deep fake technology continues to become more sophisticated, the challenge to discern this technology from reality, as well as its legal implications, becomes more relevant.[108]

    Unexpected Behaviour and Misuse

    Depending on the data set used in training, AI models could generate data with cultural and linguistic biases in data; unintended racial and gender discrimination; and generation of hateful images.[109] In 2021, a team of researchers found image and text pairs of problematic content such as pornography, racist stereotypes, and racial slurs within the LAION-400M dataset, used in some text-to-image models.[110]

    Intellectual Property

    As AI models become increasingly better at mimicking the style of artists, questions as to who the work of art begins to emerge. Some artists, such as James Gurney are open to the idea of AI models mimicking the styles of artists, provided that artists have the ability to opt-in or opt-out artwork from datasets.[111]

    Automation of Jobs

    Text-to-image models have the ability to produce artwork faster than skilled artists. The sheer speed of these models to produce artwork can have negative impacts on the freelance artist community, especially once text-to-image models continue to develop and become available to arts and graphic design companies.[112] While some very skilled artists could avoid job automation, most artists in the freelance community will need to adapt.

    AI in Marketing

    Marketing Challenges

    According to Forbes, 70% of millennials feel frustrated when companies send irrelevant emails, and 74% of customers are frustrated when the content of a website is not personalized to them [113]. Therefore, one of the biggest challenges of marketing is all about personalization. Marketing is about how the brand can reach its customers with the right messages and offers, at the right time and through the proper channels. To attain this accomplishment, the brand must be familiar with its target demographic to provide more compelling and interactive content, run tailored marketing campaigns that correspond with customer segmentation, and accommodate specific individuals’ expectations for better engagement. Approximately 29% of marketers used AI in 2018, however, that number surged to 84% by 2020. This captures a 186% increase in adoption in just only two years. As data proliferates and customer engagement increases digitally, the second-ranked marketing priority is how to engage customers in real-time [114]. It is imperative for brands to have a clear understanding of customers' needs and preferences as the digital economy continues to evolve rapidly.

    AI-powered solutions that are based on consumer data offer more accurate insights that facilitate and accelerate the creation of content customized to consumers’ needs and offer better segmentation. It can be incorporated into marketing-driven applications and how its advancements can forward firms with more effective and efficient marketing strategies, and improve the customer journey in the way a business attracts, nurtures, and converts prospects or leads. Furthermore, it can be used to automate processes that were once manual and dependent on humans.

    AI Marketing

    AI marketing is initiated from marketing automation, which is the technology that executes tasks based on if-then statements. To illustrate, if a business wants to send a call-to-action (CTA) email, it can be sent automatically to the customers, however, these automated emails will not guarantee any actions from customers. On the other hand, AI in marketing is the technology that can execute with the outputs to be more returnable. AI in marketing can execute if-then statements to determine if specific actions in return are feasible or not. For example, a software can leverage an algorithm that evaluates all the relevant aspects (time zone, cookies, digital touch-points) of each customer and then decides when is the best time to send CTA emails so the rate of people taking actions is increased. From the illustration, these two technologies in marketing often work in parallel, where marketing automation allows customer interactions to be more efficient, and AI in marketing can empower each interaction to be more customized and relevant.

    How AI-Enabled Solutions Address Marketing Challenges

    The top highest-rated use-cases of AI in marketing that provide moderate-to-high values for marketers are [115]:

    - Examine the current digital content for any gaps or possible opportunities.

    - Select keywords, key phrases and topic areas for content optimization strategies

    - Identify customer segmentation based on interests, needs, goals, and behavior

    - Discover data-driven content strategies

    - Unearth insights for top-performing content initiatives and marketing campaigns

    Content Generation

    AI can help to generate content better, faster, and cheaper. It excels at a task known as language modeling, which predicts the next word in a given sentence. Generative Pre-trained Transformer 3 (GPT-3) is an autoregressive language model that leverages deep learning to auto-complete the text. GPT-3 was introduced in May 2020, which is the latest version of a series of language-generating AI autocomplete tools designed by San Francisco-based OpenAI. GPT-3 can generate any sort of text, even with guitar tabs such as songs and technical manuals like computer code [116]. AI tools such as GPT-3 has helped marketers in drafting light and reasonably informative content such as articles, blogs, posts, emails, social media updates, and email subject lines. Given a headline or a topic, AI tools can fully or partially write the rest of the article, complete with fake quotations and statistics. Even though AI writing enables the possibility to generate standard languages based on a specific topic, AI writing is not at the point of comprehensive rationality and logicality just yet.

    Writing & Grammar Assistant

    AI can help to mitigate manual errors that frequently occur in writing. Grammarly is a tool to auto-correct spelling mistakes, grammar, and typos. It also can interpret sentence structure and provide feedback on tone adjustments, word choice, fluency and cohesion, formality level, plagiarism detection, and more. Grammarly’s AI system incorporates machine learning with a mixture of natural language processing (NLP) models [117]. A large corpus of well-written texts is organized with patterns and labeled to train AI algorithms. The AI systems also learn from interactions and feedback. For example, when many users click on the “Ignore” button to decline a recommendation, Grammarly’s linguistics and researchers will need to re-analyze and make adjustments to the algorithms to make error-prone suggestions more accurate and relevant.

    Search Engine Optimization (SEO)

    Since not all web traffic is created equally, AI in SEO analyzes the relationship between sites, content pages, and search engine ranking to produce fast and actionable tactics to optimize with the right traffic to result in conversions. A variety of companies leverage AI to analyze ranking opportunities and identify market gaps in their strategies to create content that increases their search traffic. AI SEO can spot patterns within massive data sets to know precisely what the target audience is looking for on Google or other search engines such as Bing or Yandex. Text mining AI tools learn and identify words, phrases, topics, and other variables that pique users’ interests in recent times, link to higher traffic, or get the most online searches. Based on these insights, it then suggests trending content plans based on relevance, competition, and popularity. For instance, SEO marketers concentrate on SEO algorithms to prompt their content ranking on the first page of search results for huge traffic gains. With the ubiquitousness of voice assistants (Alexa, Siri, Cortana) on all Apple, Android, or Windows devices, more and more users bypass the search screen by using voice search to process their queries, which makes voice search one of the most established AI-enabled technologies today [118]. Voice search is based on AI technologies such as processing-to-function and natural language generation (NLG). Instead of a list of search options displayed on the screens for a selection, the voice search provides only one choice, often with no attribution to the source of the results. The non-selection will significantly influence how small and entrepreneur businesses pull new customers to their websites and establish brand awareness. In addition, most users prefer using voice search at home or in their vehicles, where they are more likely to open up about personal subjects’ inquiries. Hence, consumers’ voice search behavior will significantly and directly impact the SEO priorities, which prompts local searches to be more discoverable for B2C businesses.

    Intelligent Insights

    AI can be utilized to centralize, process, and analyze massive amounts of complex operational data sets from multiple different business functions to generate insightful reports that might take weeks, if not months, to complete. Hence, with AI’s advanced capability, the firm can transform all the data into real-time and actionable predictions. For example, if a machine learning program is to understand the performance of Facebook posts, it can create an algorithm to determine which post titles have the most clicks for future posts.

    Specifically, machine learning algorithms are used to learn, examine and group unlabeled data sets without human supervision. The unsupervised machine learning identifies hidden patterns, discovers the inherent structure in the unlabeled input data sets, and classifies them based on their dissimilarities and similarities to return a response [119]. Significantly, in shorter cycle times, AI is used to target customer segmentation and help generate personalized advertisements by understanding user behavior and preferences, demographic, profiling, loyalty and retention rates, or customer lifetime value. For instance, Facebook, Twitter, and Instagram all utilize algorithms to filter out irrelevant and poor-quality posts and analyze user behavior to prioritize compelling and engaging content on the users’ news feed [120].

    A Case of AI in Social Media

    A great example is TikTok. In December 2020, TikTok surpassed Facebook as the most globally downloaded app, with over a billion active users monthly [121]. The way TikTok has used AI to directly show videos to consumers based on their recommendation system has been a major factor in the company's success. TikTok is an excellent example of gathering information from AI and letting the data drive innovative actions. Not only with highly effective algorithms, but TikTok also designed the interface to apply the recommendation immediately instead of just making a suggestion.

    TikTok's Content Flow

    Video Analysis & Categorization

    TikTok leverages computer vision to analyze the videos' many features, effects, and other traits (color, shape, texture) of people and objects. Then, NLP extracts the audio information to identify patterns and classify the data. In addition, users will be the one who provides metadata such as descriptions, captions, and hashtags in the videos. Metadata is the data that gives information about other data but not the content of the data [122]. With all components mentioned above, TikTok does analyze every part of the video to have a comprehensive understanding of its content and context.

    TikTok's Video Analysis

    Audience Prediction

    After video analysis, it comes to audience prediction based on the prophecies about one user's preferences, with the applications of two models, content-based filtering and collaborative filtering.

    When users open the app, TikTok presents a bunch of videos across varying topics to users. Depending on the level of first-time engagement, TikTok’s algorithm can then apply content-based filtering to look for likenesses between suggested videos and recent videos that a user just engaged with. The estimated engagement rate is based on TikTok's ranking matrix considering the following activities: likes (2 points), comments (4 points), shares (6 points), completion (8 points), and re-watch (10 points). The decision matrix is established with two closely related factors - “retention” and “time spent,” as TikTok wants to keep its users' on as long as possible. Based on the score hierarchy, TikTok will immediately understand its user’s behavior and deliver content correspondingly [123].

    Collaborative filtering, which is to feed users videos based on the behavior of similar users. If User A engages with videos 1, 2, 3, and 4 and user B engages with 2, 3, 4, and 5, TikTok is likely to identify the similarities between these two users (videos 2, 3, 4) and cross-suggest video 1 to user B and video 5 to user A [124].

    Audience Recommendation

    TikTok uses machine learning to curate a For You feed for each user. The feed is a stream of videos that are personalized to every user’s unique interests [125]. TikTok also evaluates its users' positive and negative performance signals based on the interactions with its video recommendations following the ranking matrix mentioned above. The good signal is deciphered from viewers liking the videos, following its content creators, watching it till the end, or re-watching. A negative signal is interpreted from users swiping away, stopping the videos, exiting the app, or even from the speed at which the users immediately swipe away [126]. So if it is a good signal, the video will be forwarded back to the audience prediction stage to be analyzed and promoted to different user clusters with the same interests.


    Davianna Chien Manesh Sandhu Vaibhav Arora Amandeep Boparai Leah Dinh Tristan Perales
    Beedie School of Business
    Simon Fraser University
    Burnaby, BC, Canada
    Beedie School of Business
    Simon Fraser University
    Burnaby, BC, Canada
    Beedie School of Business
    Simon Fraser University
    Burnaby, BC, Canada
    Beedie School of Business
    Simon Fraser University
    Burnaby, BC, Canada
    Beedie School of Business
    Simon Fraser University
    Burnaby, BC, Canada
    Beedie School of Business
    Simon Fraser University
    Burnaby, BC, Canada


    41. ​​
  • Personal tools