AI News: Touch Soccer Ball, AI in Cinema, Medicine and Military
The autumn draws to a close but data analytics and AI market has just come into full force. Here is your monthly summary of what the tech world was talking about. Let’s start with some eternal things, shall we?
The global sports event has started. With millions watching World Cup 2022 in Qatar there’s a multitude of innovations used at the stadium and around it. Take, for example, a sensor soccer ball made by Adidas. With motion sensors inside the ball, it reports precise location data 500 times per second. This is for sure a good aid for referees. To minimize the controversy for them even more, a “team” video assistant referees (VARs) based on algorithms and data points help on-field judges make accurate calls.
Off the stadium you can use benefits of an algorithm developed by the Alan Turing Institute in Britain. It predicts the winner in the World Cup by accumulating data sets with the results of all international football matches since 1872. The new model was run 100, 000 times by researchers. It goes on saying that Brazil has 25% chances of winning with Belgium and Argentina following it.
More on an art scene, artificial intelligence in the non-fungible token (NFT) space is becoming increasingly popular. Generative art – which means that the art piece is created by an autonomous system, - has recently become one of the main categories of the NFT market. But what drives this immense progress? The experts say the growth is due to three main factors: multimodal AI, diffusion methods and pretrained language models. Models like DALL-E or Stable Diffusion enable image and video generation from natural language. This lowers the barrier for specialists and novices alike to interact with new generative AI models.
Another November case in the art world is presented by world’s known Museum of Modern Art (MoMa) in New York. The artist Refik Anadol presents Unsupervised, the large-scale installation with digital artworks based on AI to reinterpret more than 200 years of art from MoMA’s collection. Fed by 138,151 pieces of metadata of the museum collection, AI generates new images as part of the larger concept where Refik has been collecting data from digital archives and public resources, and processing it with ML classification models.
Well, you can call films another large area of the art market. Here is how AI can help the film industry minimize costs. OpenAI's DALL-E, Midjourney and other AI tools can use images from the internet and select datasets to train their AI models to reconstruct similar yet wholly original imagery using text prompts. Thus, they generate a new actor to play in the next blockbuster with no privacy violations. Since these actors never existed. It means that Hollywood studios can use AI for casting.
In the animal world praise goes for the University of Florida. A team of researchers are using artificial intelligence to collect data of an animal's body language and posture to reveal specific ailments. The program analyzes videos with animals’ behaviour and detects patterns in how certain among them, - horses or cows, for instance, - walk, trot or run. As the program learns how an animal normally moves, it can spot abnormalities and symptoms of possible deceases. When farmers of Florida have the access to the system, this will be a great value since the AI program can significantly minimize costs for treatments of animals, which are in later stages of deceases.
While helping farm animals, AI also penetrates university grounds. The survey reveals that the professions for which universities prepare students will soon rely on AI. Or rely already. Journalism is just one of the examples. 71 media organisations from 32 countries found AI already a “significant part of journalism”, used for automatic news collection and filtering, news production (including automatic fact checkers and transformation of financial reports into texts) and news distribution, making it more personalized for each subscriber. Why have a bunch of reporters on a payroll, if a single AI can do their job?
On a more serious side, comes usage of machine learning and analysis in dealing with deceases. A collaboration study done by the MIT and the Cornell University found more accurate ways to anticipate the advances in Alzheimer’s disease. New models show that predictions of the decline for patients with mild cognitive impairment is more accurate than it is for cognitively normal individuals. On the contrary, predicting the development of illness for cognitively normal subjects is less accurate in the long run.
Science also goes ahead. Researchers from the George Washington University, Queens University, University of British Columbia and Princeton University discovered enhanced method of hardware reconfiguration for ML trainings. After one training step, the experts observed an error and reconfigured the hardware for a second training cycle followed by additional training cycles until a sufficient AI performance was reached. Previously photonic chips could classify and infer information from data. Now, photonic tensor cores and other electronic-photonic application-specific integrated circuits demonstrate a totally different capability which leads to more efficient photonic chip manufacturing for ML and AI.
Who knows a lot about manufacturing chips is China. But this November its current leader Xi Jinping made a special stress on another important thing. Speaking at the 20th National Congress of the Chinese Communist Party he stated the country would use all technology advances, including artificial intelligence for strengthening the national army. The journalists point out that the Xi Jinping mentioned the word “intelligent” three times in a single report. Isn’t it a new direction for developing the immense AI market in China?
Whereas China’s neighbour, South Korea, experiments with reinforcement learning algorithms to solve traffic signal control problems and reduce congestion. Chung-Ang University team points out to the recent advancements in AI and ML which help optimize traffic signal controls and make driving in busy urban areas less nervous.
Vendor news
IBM released a new business intelligence and analytics suite, Business Analytics Enterprise. It is designed to help companies break down data silos and barriers to collaboration caused by the use of varied sets of analytics tools across different divisions. Business Analytics Enterprise have a new Analytics Content Hub and Watson-powered versions of Planning Analytics and Cognos Analytics. The system is a ready-to-use business intelligence tool for budgeting, reporting, and forecasting data across multiple business units.
U.S. chip designer and computing firm Nvidia announced its cooperation with Microsoft for building a “massive” computer to handle intense artificial intelligence computing work in the cloud. The AI computer will operate on Microsoft Azure cloud, using tens of thousands of graphics processing units, Nvidia’s most powerful H100 and its A100 chips.
Analysts say…
This month revealed new insights from Gartner. Frank Buytendijk, distinguished VP Analyst at Gartner, wrote a list of recommendations for chief data and analytics officers who strive to learn from their advanced peers to become more data-driven companies. To improve their practices Gartner analyst advises to establish executive relationships, to maintain focus on business outcomes and to experiment with new techniques, to create a culture of collaboration, to refine their data management capabilities and to constantly optimise and innovate.
Term of the month
Let’s talk about synthetic data. It is that kind of data which is generated artificially or algorithmically and closely resembles actual data’s underlying structure and property. There are usually three types of synthetic data: text data, audio or visual data (for example, images, videos, and audio) and tabular data.
Synthetic data can be used in a number of cases for machine learning. These include use of synthetic text data for training NLP models. Good example is Alexa AI team at Amazon that uses synthetic data to finish the training set for their natural language understanding system. It provides them with a solid basis for training new languages without existing or enough consumer interaction data.
Another area is training vision algorithms. Here you can use GAN or some other generative network to generate realistic human faces and to generate as much data as we want from these algorithms without breaching anyone’s privacy.
The last use case is related to tabular synthetic data. It is an artificially generated data that mimics real-world data stored in tables. Structured in rows and columns, it can contain any data, e.g., a music playlist or a finance record.
With this, we are rounding off with November summary and hope to bring you more insights in a month.