AI. Dawn of the Age

Back in my college days, AI seemed to belong to the realm of theoretical studies and science fiction. Things changed. A lot. And things keep changing at an accelerating rate. It seems that we are now at the dawn of the Artificial Intelligence Age.
9 min read
02/02/21
By Andrey Batutin
Mobile Machine Learning Lead, AI Center of Excellence
Share
AI. Dawn of the Age

Just a few years ago, most of things we consider machine learning (ML) technologies nowadays were a part of statistics; neural networks with more than three layers were considered deep. Word gradient was firmly associated with colour pallet, not with optimisation algorithm.

AI. Capabilities

I start every morning with reading a bunch of latest news about artificial intelligence (AI). And sometimes it seems that AI can do everything. Self-driving cars, disco dancing robots, deep fakes, AI-generated music, autonomous killing drones. Has Skynet already won, and I just did not receive a memo? Let’s try to understand better what AI is actually capable of.

AI. Games

Within the last 10 years, AI technologies made significant progress in such a field as games. In all competitive video games – poker, chess and go – AI achieved human level performance and consistently beat best-of-the-best human opponents.

AI. Human Level Performance

One way to evaluate AI performance is to compare it to the best human experts. If the AI-powered model is able to compete with humans and deliver results similar to human success rate, we say that AI has achieved human level performance in this field.

For instance, here are the stats of top StarCraft II players

Player Rating Win Rate Prize Earnings
Serral 3328 69.63 % $883,466
MaNa 2505 58.67 % $210,057
TLO 2202 54.16 % $108,372
Source: Aligulac

In December 2018, DeepMind debuted AlphaStar, AI Protoss player, and challenged one of the best human players out there – TLO and MaNa – in a series of 5 intense games. AlphaStar was able to beat both of the human players with humiliating scores. This gave DeepMind the basis to claim that AlphaStar achieved human level performance in StarCraft II. AlphaStar will surpass human level performance when it consistently beats top players like Serral.

How AI Achieved Human Level Performance in Games: Timeline

AI. Computer Vision

The most visible progress AI achieved is in the field of computer vision. With the introduction of deep learning computer vision, models for object detection received an amazing boost in accuracy and achieved an almost exponential improvement rate within the last 10 years.

What Is Deep Learning?

Deep learning is a general term for a family of AI models that use deeply stacked layers of artificial neural network to learn desired model behaviour.

Graph with Improvement of Accuracy of Object Detection
Image Source: Object Detection in 20 Years: A Survey

What Is mAP?

Mean Average Precision, or mAP, is a generalised metric used in object detection tasks to evaluate model performance. It refers to how well the model can detect a wide variety of objects in different environments. Standard image datasets, like Common Object in Context COCO, or Visual Object Classes Challenge VOC07/VOC12, are used to evaluate the mAP of models.

Object Detection model accuracy has improved astoundingly and now hovers around human level performance for most computer vision tasks. In 1998, Yann LeCun introduced influential LeNet-5 architecture that had all fundamental building blocks of modern Computer Vision models. And advancements in GPU performances allowed to alleviate the model accuracy to a whole new level with introduction of AlexNet architecture in 2012.

A Road Map of Object Detection
Image Source: Object Detection in 20 Years: A Survey

Rapid development of cloud technologies and big data in early 2010s gave birth to a variety of remarkable image classification and object detection models and enabled an astonishing variety of technologies – from Face ID on your phone to Self-Driving cars in your garage. AI in computer vision is an excellent showcase of immense synergy potential of science, technology and business.

AI. Art

One interesting side effect of deep learning applied to computer vision is AI-generated art. In 2015, while working on image classification models, Alexander Mordvintsev realised that if one cuts the model and outputs the images generated inside it during the training, he/she can get a dream-like psychedelic picture, where the silhouette of rock looks like Japanese pagoda, and clouds look like mountain ridges. His DeepDream model was using neural style transfer to achieve this result.

Input and Output Images Made Using a Network Trained on Places
Top: Input image. Bottom: output image made using a network trained on places by MIT Computer Science and AI Laboratory
Image Source: DeepDream

What Is Neural Style Transfer?

Neural style transfer is a computer vision technique powered by deep learning, which allows composing one image in the style of another image. It is trained on works of famous painters like Picasso or Dali to extract and apply distinct art styles to one’s photos. Rembrandt’s impasto or van Gogh's pointillé – all of these art techniques are captured with precision and accuracy by the AI brain, so they become applicable to your very own selfie and family photos.

The following painting was created with generative adversarial networks approach by a French art group, Obvious. It was sold by Christie’s for $432,500 in October 2018.

Edmond De Belamy Portrait
«Edmond De Belamy»
Image Source: Obvious

Obvious fed 15,000 images of portraits belonging to different art periods to an algorithm. The algorithm generated its own portraits, attempting to create original works that could pass as man-made.

If you are still not impressed, try to guess which of these paintings are the result of inspiration, talent, and amazing craftsmanship of human beings and which – are soulless products of heartless AI?

Quiz
Which of these paintings were created by Artificial Intelligence?

The right answer is: painting 2 is an AI-born. Someone even bought this one-of-a-kind artwork at Art-AI gallery.

The other two paintings were created by creatures of flesh and blood. Painting 1 is “View of Argenteuil in the Snow” by Claude Monet. Painting 3 is “Algerian Landscape (The Ravine of the Wild Women)” by Pierre-Auguste Renoir. Impressive, right?

AI. Industry Adoption

Now let's see how many companies have already adopted AI and what impact it has on their performance. According to McKinsey’s The State of AI in 2020, 50% of respondents report that their companies have adopted AI in at least one business function. And 22% of respondents attribute at least 5% of their earnings before interest and taxes (EBIT) to AI.

Graph for AI Adoption by Industry

With new AI-based products leading their way, there is a strong presence of AI at business functions across the board. Whatever your industry is, it is increasingly likely that you will work together with an AI co-worker.

Another important point is a correlation between organisation strength and AI contribution to EBIT.

Strong mature organisations with effective leadership show consistently better results from AI adoption compared to their less mature competitors. These three qualities are common for AI high-achievers:

  • outstanding performance
  • strong leadership
  • resource commitment to AI.

Respondents from the companies with AI high performers are 2.3x more likely than others to consider their C-suite leaders very effective. And, effective organisation structure yields good AI results. Companies with strong leadership, committed to AI adoption, are rapidly getting competitive advantage over their peers with AI technology.

AI. Labour Market

Two main topics discussed at the World Economic Forum are COVID impact on labour market and AI. We should expect a significant increase in the number of firms expecting to adopt non-humanoid robots and artificial intelligence, with both technologies slowly becoming a mainstay of work across industries. By 2025, AI is expected to do half of the work in world economy, with Information and Data Processing leading in this list.

Graph with Share of Tasks Performed by AI vs Humans

A significant rise of AI impact is expected in manual work activities and human resource management. It is highly likely that, in 5 years from now, AI will be the one to evaluate your CV.

It is no surprise that AI talents are top 2 job roles in «Increasing Demand» category.

Rank Increasing demand role
1 Data Analysts and Scientists
2 AI and Machine Learning Specialists
3 Big Data Specialists
4 Digital Marketing and Strategy Specialists
5 Process Automation Specialists
Source: The Future of Jobs Report 2020

Combined with McKinsey survey respondents’ strong emphasis on AI talent management and workforce talent upskilling to get best of AI adoption, the following two statements prove true:

  • AI adoption requires strong talent pool
  • AI skills are in increasingly high demand

To remain competitive in an ongoing AI race, organisations must have quick access to a capable and experienced talent pool. Putting the right people to do the right job can bring your company into the lead of the race.

AI. Risks

The case with every emerging technology, AI adoption brings unique risks to the table. And the best way to mitigate the risks is to be aware of them upfront.

Major Risks Posed by AI

Besides, there are risks associated with the AI adoption process, AI projects being delivered on schedule, overspending and worst-than-predicted AI model performance – all these risks are often a reality in AI projects. AI can be seen as a very powerful disruptive technology that comes with considerable risk, so an effective AI risks management is a “must” for modern organisations.

AI. DataArt

As a technology vendor, DataArt is eager to participate in the AI revolution by providing top-notch AI consulting and development services. Our expertise is backed up by a carefully managed pool of specialists: data scientists, data analysts, AI/ML engineers, computer vision and NLP experts. With our many-year industry expertise in Finance, Healthcare, Travel & Hospitality, Media & Entertainment, Retail & Distribution, among other industries, DataArt teams will ease initial AI adoption and risk management for your company. We will find an optimal way to extend your current business capabilities with AI.

Talk to DataArt Experts about AI Adoption

AI. Dawn

With science, technology, and business working together hand in hand, AI is rapidly transforming our world, turning science fiction to a reality for millions of people across the globe. We are witnessing the dawn of the AI Age, and DataArt is eager to help your organisation with AI transformation. AI adoption requires organisation-wide changes and coherent leadership, with the three pillars – strong leadership, AI talent management, and AI risk mitigations – remaining actual.

Sign Up for Updates!

Subscribe now to receive industry-related articles and updates

Choose industries of interest
Thank You for Joining!

You will receive regular updates based on your interests. No spam guaranteed

Add another email address
READ MORE
Sign Up for Updates!
Choose industries of interest
Thank You for Joining!

You will receive regular updates based on your interests. No spam guaranteed

Add another email address
Welcome
We are glad you found us
Please explore our services and find out how we can support your business goals.
Get in Touch Envelope
DataArt-Com-Retail-Report.png
Download the white paper Glancing Forward into 2021: An Industry by Industry Outlook

Explore digital trends and unanticipated benefits engendered by the pandemic, which are likely to last in 2021.