According to 2020’s McKinsey Global Survey on artificial intelligence (AI), in 2020 more than 50% of companies have adopted AI in at least one business unit or function, so we witness the emergence of new AI trends. Organizations apply AI tools to generate more value, increase revenue and customer loyalty. AI leading companies invest at least 20% of their earnings before interest and taxes (EBIT) in AI. This figure may increase as COVID-19 is accelerating digitization. Lockdowns resulted in a massive surge of online activity and an intensive AI adoption in business, education, administration, social interaction, etc.
This article aims to overview new and current AI trends that emerged in 2020 and are still increasing in 2021. Based on trends, companies can make projections of the AI future in 2022 and successfully mitigate risks. Saying AI, we mean neural networks, machine learning, deep learning, computer vision, and other subfields of Artificial Intelligence.
AI Adoption level differs depending on the industry. Using the data mentioned in the McKinsey Global Survey on AI, we can highlight four leading sectors: high-tech, telecom, automotive, assembly.
Companies apply AI for service operations, service or product design, advertising, and sales. Regarding investments, the area of drug discovery and development received the highest amount of money — in 2020, the total sum of investments exceeded 13.8 billion dollars, 4.5-fold higher than the year before.
AI drives the highest revenue growth if applied in inventory and parts optimization, pricing and promotion, customer-service analytics, sales, and demand forecasting. Use cases that reported cost decrease are related to optimization of talent management, contact-center automation, and warehouse automation.
In 2021 and the following years, AI will be leveraged to simplify operations and make them more efficient. Businesses should try to benefit from the commercial application of Artificial Intelligence through improving IT infrastructure and data management. But not each deployed AI model could be useful for companies and appropriate for performance monitoring. We’ll focus on AI trends 2021-2022 that are likely to become mainstream.
AI techniques have already been applied to face recognition, voice identification, and video analysis. These techniques form the best combo for surveillance and biometric authentication. So, in 2021, we can foresee the intensive exploitation of AI in video surveillance.
Artificial Intelligence is beneficial for a flexible setup of security systems. Previously, engineers spent a lot of time configuring the system because it was activated when a specific number of pixels on a screen changed. So, there were too many false alarms. These alarms were caused by falling leaves or a running animal. Thanks to AI, the security system identifies objects, which contributes to a more flexible setup.
AI in video surveillance can detect suspicious activity by focusing on abnormal behavior patterns, not faces. This ability enables creating more secure spaces, both public and private, through identifying potential threats. Such AI-driven video solutions could be also useful for logistics, retail, and manufacturing.
Another niche that provides promising perspectives for the AI application is voice recognition. Technologies related to voice recognition are able to determine the identity. By identity, we mean the age of a person, gender, and emotional state. The principles on which voice recognition for surveillance is based can be the same as in the case of Alexa or Google Assistant. A feature that is suitable for security and surveillance is a built-in anti-spoofing model that detects synthesized and recorded voice.
One of the most crucial technologies for security is biometric face recognition. Different malicious applications try to trick security systems by providing fake photos instead of real images. To defend against such cases, multiple anti-spoofing techniques are presently being developed and used at large scale.
The challenge for processing of real-time video streams is handling data pipelines. Engineers aim to ensure accuracy and minimize latency of video processing. And AI solutions can help to achieve this goal.
To implement an AI-based approach in live video processing, we need a pre-trained neural network model, a cloud infrastructure, and a software layer for applying user scenarios. Processing speed is crucial for real-time streaming, so all these components should be tightly integrated. For faster processing, we can parallelize processes or improve algorithms. Processes parallelization is achieved through file splitting or using a pipeline approach. This pipeline architecture is the best choice since it doesn’t decrease a model’s accuracy and allows for use of an AI algorithm to process video in real-time without any complexities. Also, for pipeline architecture, it’s possible to apply additional effects implying face detection and blurring. You can find more information on the subject in our article dedicated to AI in real-time video processing.
Modern real-time stream processing is inextricably linked to the application of background removal and blur. The demand for these tools has increased because of COVID-19 contribution to the emergence and popularization of new trends in video conferencing. And these trends will be actively developed because, according to GlobeNewswire, the global video conferencing market is expected to grow from USD 9.2 billion in 2021 to USD 22.5 billion by 2026.
There are different ways to develop tools for background removal and blur in a real-time video. The challenge is to design a model capable of separating a person in the frame from the background. The neural network that is able to carry out such a task could be based on existing models like BodyPix, MediaPipe, or PixelLib. When the model is chosen, the challenge remains for its integration with an appropriate framework and organizing the optimal execution process through the application of WebAssembly, WebGL, or WebGPU.
Modern AI models are able to generate text, audio, images in a very high quality, almost indistinguishable from non-synthetic real data.
At the heart of text generation stands Natural Language Processing (NLP). Rapid advances in NLP have led to the emergence of language models. For instance, BERT model is being successfully used by Google and Microsoft to complement their search engines.
How else does the development of technologies related to NLP boost companies? First of all, combining NLP and AI tools allows the creation of chatbots. According to Business Insider, the chatbot market is expected to reach USD 9.4 billion in 2024, so let’s emphasize the ways businesses benefit from AI-driven chatbots implementation.
Chatbot tries to understand the intentions of people, instead of just performing standard commands. Companies working in different areas use the AI-driven chatbot to provide their clients or users with human-level communication. Applications of chatbots are widely observed in the following business domains : healthcare, banking, marketing, travel and hospitality.
AI-driven chatbots help to automate admin tasks. For instance, in healthcare they reduce the amount of manual work. Here, chatbots help to organize appointments, send reminders related to taking meds, and provide patients with answers to queries. In other areas, chatbots are introduced to deliver targeted messages, improve customer engagement and support, and provide users with personalized offers.
Besides chatbots, NLP lies at the heart of other cutting-edge technological solutions. One of the examples is NLP text generation that can be used in business applications. An NLP-based Question Generation system presented in the video below is used in a secure authentication process.
The recent arrival of the GPT-3 model allows AI engineers to generate an average of 4.5 billion words per day. This will allow a tremendous range of downstream applications of AI for both socially useful and less useful purposes. It is also causing researchers to invest in technologies for detecting generative models. Note that in 2021-2022 we will witness the arrival of GPT-4 — “artificially generally intelligent AI”.
Coming back to Generative AI, we want to pay attention to GANs, or Generative Adversarial Networks, that are now capable of creating images indistinguishable from human-produced ones. That could be images of non-existent people, animals, objects, as well as other types of media, such as audio and text. Now is the best moment to implement GANs gaining from their abilities because they can model real data distributions and learn helpful representations for improving the AI pipelines, securing data, finding anomalies, and adapting to specific real-world cases.
The most remarkable branch of Computer Vision is AI inspection. In recent years, this direction has been prospering because of the increasing accuracy and performance. Companies started to invest both computational and financial resources to develop сomputer vision systems at a faster rate. The intensive development of AI inspection is also connected with a rapid progress in the domain of object detection in video frames.
Automated inspection in manufacturing implies the analysis of products in terms of their compliance with quality standards. The methodology is also applied to equipment monitoring.
Here are few use cases of AI inspection:
- Detecting defects of products on the assembly line
- Identifying defects of mechanical and car body parts
- Baggage screening and aircraft maintenance
- Inspections of nuclear power stations
- Use cases of AI inspection
The next trend related to the implementation of AI in the healthcare industry has been intensively discussed over recent years. Scientists use AI models and computer vision algorithms in the fight against COVID-19, including areas like pandemic detection, vaccine development, drug discovery, thermal screening, facial recognition with masks, and analyzing CT scans.
To counteract the spread of COVID-19, AI models can detect and analyze potential threats and make accurate predictions. Also, AI helps to develop vaccines by identifying crucial components that make them efficient.
AI-driven solutions may be applied as an efficient tool in The Internet of Medical Things and for handling confidentiality issues specific to the healthcare industry. If we systematize use cases of AI in healthcare, it becomes clear that they are united by one aim – to ensure that the patient is diagnosed quickly and accurately.
No-code AI platforms have enabled even small companies to apply powerful technologies that were previously available only to large enterprises. Let’s find out why such platforms are a key AI trend for businesses in 2021.
Developing AI models from scratch requires time expenditure and relevant experience. Adoption of the no-code AI platform simplifies the task because it reduces the entry barrier. The advantages are:
- Fast implementation — compared with writing code from scratch, working with data, and debugging, time saving reaches 90%.
- The lower cost of development — through automation, the businesses eliminated the need for large data science teams.
- Ease of use — drag-and-drop functionality simplifies software development and enables the creation of apps without coding.
No-code AI platforms are in demand in healthcare, financial sector, and marketing — though produced solutions couldn’t be highly customized. Among the most sought-after no-code AI platforms, you can find Google Cloud Auto ML, Google ML Kit, Runaway AI, CreateML, MakeML, SuperAnnotate, etc.
Enterprise-sized companies, as well as mid-size businesses, leverage no-code platforms for software solutions aiming at image classification, recognizing poses and sounds, and object detection.
The lack of diversity in AI can contribute to the emergence of racial and gender biases. By diversity, we mean a variety of people who develop AI models. According to NYU’s research, 80% of professors involved in AI development are men, and only 10% of researchers who work with Artificial Intelligence at Google are women. The same research shows that not even 5% of staff at Google, Facebook, and Microsoft are Black workers.
The number of female graduates of AI PhD programs and computer science faculties has remained at a low level for a long time. But the need for diversity in AI should influence this situation, which is one of the emerging trends. Moreover, women in AI can make big decisions influencing the development and implementation of AI systems. So, if you want to know more about females in AI, read the article dedicated to the brilliant career path of an AI engineer at MobiDev.
Trends show that the future of Artificial Intelligence is promising because AI solutions are becoming commonplace. Autonomous cars, robots and sensors for predictive analysis in manufacturing, virtual assistants in healthcare, NLP for reports in media, virtual educational tutors, AI assistants, and chatbots that can replace humans in customer service — all these AI-powered solutions are moving forward with huge steps.