Imogen West-Knights is absolutely right about us losing our brain power to the artificial intelligence bots (ChatGPT has its uses, but I still hate it – and I’ll tell you why). I too believe creative imagination is a muscle, which needs its exercise. She is also right that it can revolutionise scientific endeavour. My field of weather forecasting will soon be revolutionised by machine learning – a type of AI – where we recognise enough past weather patterns so that we can predict what weather will be coming. But writing best-man speeches, leaving speeches for work colleagues, letters to a dear friend? Do we really want to dissolve into brain-lazy folk who use AI to be the understudy to our own emotions?

If I say “I love you” to someone, would they like to hear it from me or a bot? There is also another concern: AI output has no audit trail, no clues to its source. Its source is the wild west. Anyone – good, bad, indifferent – can feed into it, program it, bias it. If, as you say Imogen, you do end up in the woods in an “analogue manner” with your ability to think intact, I’ll happily join you. Hopefully others will too.
Murray Dale
Hayle, Cornwall

Imogen West-Knights shares her hatred of offloading to ChatGPT the tasks that make us human. And while I share her concerns (I couldn’t have put them better myself), there is an additional one that troubles me: if students go through their entire school lives with this all-knowing and all-solving technology at their fingertips, how will their critical thinking skills develop?

Students in literature class are not given books such as The Great Gatsby so they can regurgitate the plot 20 years later at a dinner, but rather so that they can understand the interconnection of class disparities, wealth and the social atmosphere after the first world war, and so they can trace parallels with the present day.

They learn multivariable calculus not because they will need it to buy groceries but to make their brains strong and malleable, so that grasping and implementing new concepts and ideas will become easier, whatever the subject.

And they don’t learn history so they can repeat over and over “Victoria 1837, Edward VII 1901, George V 1910, Edward VIII 1936, George VI 1936, Elizabeth II 1952”, but to understand how sequences of events have led to wars, legislative changes and economic crises, and can do so again.

Technologies that make work easier have always been seductive, and always will be. AI usage is already rampant in secondary schools and universities.

But as ChatGPT turns three years old in a few months, preschoolers are also starting to go to kindergartens. And I wonder how in the years to come we will ensure that their answer to everything is not “I will ask ChatGPT.”
Ignacio Landivar
Berlin, Germany

PE is essential for the development of physical fitness, motor skills, mental health, and lifelong healthy habits in students; thus, it is a pillar of holistic education in contemporary institutions. As technology continues to improve at an alarming rate, AI algorithms are becoming increasingly common in the delivery, monitoring, and personalization of PE programs. Real-time performance monitoring, customized feedback, and data-based training programs can be achieved through AI-powered systems, including pose estimation tools, adaptive training optimizers, and interactive coaching assistants, which can dramatically improve learning outcomes and engagement. Nevertheless, the large number of possible AI solutions and the trade-offs between accuracy, adaptability, ease of use, and resource demands necessitate careful and systematic DM to choose the most appropriate AI algorithms for PE requirements. A practical, uncertainty-sensitive DM framework can help educators and administrators make informed, transparent, and sensible decisions, ensuring that the chosen AI technologies support educational objectives and bring the most value to both students and institutions.

To demonstrate the feasibility and practical value of the proposed q-RLDF WASPAS framework, a real-world case study focusing on the selection of the most suitable AI algorithm for enhancing PE programs is chosen. The goal of this case study is to assist PE instructors and institutional DMKs in choosing AI solutions that best align with their objectives of improving training effectiveness, ensuring student safety, and personalizing learning experiences. To provide a balanced and credible evaluation process, three domain experts/DMK were selected based on clear criteria: practical experience in PE technology, technical expertise in AI for sports data analytics, and strategic insight into educational technology policy. Each DMK has at least 5 years of post-qualification experience. This diversity ensures the DM process reflects operational, technical, and organizational realities. Based on recent trends in AI algorithms and DMK’s opinions, five widely recognized AI algorithms \(\:{{\Lambda\:}}_{i},i=\text{1,2},\dots\:,5\) with proven applications in sports science and physical training, are considered alternatives. These alternatives represent a broad spectrum of intelligent capabilities ranging from real-time posture tracking and motion analysis to adaptive training plan generation and personalized feedback delivery. The selected widely recognized AI algorithms with a proven track record in sports science are:

\(\:{{\Lambda\:}}_{1}:\) Convolutional neural network-based motion analysis (CNN-MA),

\(\:{{\Lambda\:}}_{2}:\) Reinforcement learning based training optimizer (RL-TO),

\(\:{{\Lambda\:}}_{3}:\) Expert system for exercise prescription (ES-EP),

\(\:{{\Lambda\:}}_{4}:\) Hybrid AI tutor with natural language processing (HAI-NLP),

\(\:{{\Lambda\:}}_{5}:\) Wearable sensor data mining algorithm (WSDMA).

Following a thorough analysis of recent and relevant literature^{41,42,43,44,45,46,47,48}, eight criteria \(\:{\mathfrak{S}}_{j},\:j\:=\:1,\:2,\:\dots\:,\:8\:\)are identified as the most significant determinants of AI algorithm performance in PE. These requirements encompass various technical, pedagogical, and practical aspects, which play a crucial role in determining the effectiveness and applicability of AI models in dynamic PE settings. They are chosen based on both empirical results and expert opinion on which attributes have a significant impact on the performance of an algorithm in a real educational context. The detail of the criteria is given below:

\(\:{\mathfrak{S}}_{1}:\) Prediction accuracy (The extent to which the algorithm accurately classifies or predicts the physical performance of students, their fitness, or their injury risk).

\(\:{\mathfrak{S}}_{2}:\) Processing time (The duration that it takes to train the AI model and process large amounts of student movement or biometric data).

\(\:{\mathfrak{S}}_{3}:\) Interpretability (How easily coaches, PE instructors, and students can understand how the algorithm makes its predictions or recommendations).

\(\:{\mathfrak{S}}_{4}:\) Resource consumption (The computational and hardware resources needed to deploy and maintain the algorithm).

\(\:{\mathfrak{S}}_{5}:\) Scalability (The capacity to process more and more students, a variety of sports activities, and bigger data without a performance decline).

\(\:{\mathfrak{S}}_{6}:\) Personalization capability (The extent to which the algorithm can be adjusted to the individual physical state of students, their goals, and progress in the long term).

\(\:{\mathfrak{S}}_{7}:\) Integration with existing modules (The ease with which the AI solution will be able to integrate with wearable devices, learning management systems, or innovative campus platforms, and existing PE software).

\(\:{\mathfrak{S}}_{8}:\) Data privacy and security (The level at which the algorithm facilitates safe data processing and adheres to the privacy requirements of sensitive health or biometric data of students).

The interrelationship between the chosen criteria, available AI algorithms, and selection of the best AI algorithm for PE is shown in Fig. 3.

Step 1. In this step, each DMK was independently provided with a structured questionnaire containing eight selected evaluation criteria and five alternatives for AI algorithms. They rated the importance of each criterion and the performance of each AI algorithm against each criterion in the form q-RLDFNs. These preferences are further arranged in the form of decision matrices. The DMK’s preferences are arranged in the form of decision matrices and are given in Tables 1, 2 and 3.

1. a)

   \(\:\stackrel{-}{M}{}_{ji}=\frac{1}{\delta\:}\sum\:_{k=1}^{\delta\:}{d}_{ji}^{\left(k\right)}\), used to find group average matrix \(\:\stackrel{-}{M}=\left[\stackrel{-}{M}{}_{ji}\right]\).

2. b)

   \(\:{\zeta\:}_{k}=\sum\:_{j=1}^{\beta\:}\sum\:_{i=1}^{\alpha\:}{\chi\:}_{q-RLDF}\left({M}_{ji}^{\left(k\right)},\stackrel{-}{M}{}_{ji}\right)\), is utilized to find the total dissimilarity matrix where \(\:{\chi\:}_{q-RLDF}\left({M}_{ji},\stackrel{-}{M}{}_{ji}\right)\) represents the distance function between two q-RLDFNs given by:

1. iii)

   The consensus level (CL) is measured by using:

For predefined consensus threshold \(\:C{L}_{\text{t}\text{h}\text{r}\text{e}\text{s}\text{h}\text{o}\text{l}\text{d}}=0.8\), the computed \(\:CL=0.81\). Therefore, the relation \(\:CL\ge\:C{L}_{\text{t}\text{h}\text{r}\text{e}\text{s}\text{h}\text{o}\text{l}\text{d}}\) is satisfied, and consensus is met.

1. d)

   Since there is consensus among DMK’s preferences, consensus-based weights are computed using \(\:{\psi\:}_{k}^{{\prime\:}}=\frac{\raisebox{1ex}{$1$}\!\left/\:\!\raisebox{-1ex}{${\zeta\:}_{k}$}\right.}{\sum\:_{k=1}^{\delta\:}\raisebox{1ex}{$1$}\!\left/\:\!\raisebox{-1ex}{${\zeta\:}_{k\_}$}\right.}\), where \(\:\sum\:_{k=1}^{\delta\:}{\psi\:}_{k}^{{\prime\:}}=0\).

The weights of DMKs and criteria are computed using the above relations, which ought to be \(\:{\psi\:}_{k}^{{\prime\:}}={\left\{\text{0.35,0.40,0.25}\right\}}^{T}\), and \(\:{\psi\:}_{j}={\left\{\text{0.13,0.10,0.15,0.11,0.12,0.13,0.11,0.15}\right\}}^{T}\) respectively.

Step 3. The DMK’s preferences are aggregated to obtain a single combined decision matrix, which collectively exhibits the preferences of all the DMKs. The q-RLDFWA AO used for this purpose is given as:

For \(\:q=5\), the aggregated combined decision matrix is provided in Table 4.

Step 4. The criteria are assessed and considered as cost type or benefit type. On assessment, it is observed that \(\:{\mathfrak{S}}_{2}\) and\(\:\:{\mathfrak{S}}_{4}\) are cost-type criteria, while the remaining are benefit-type. The cost type criteria are normalized by inverting the MG and NMG. The normalized decision matrix is denoted by \(\:{M}^{{\prime\:}}\) and shown in Table 5.

The relative significance of alternatives using WSM is provided in Table 6.

The relative significance of alternatives using WPM is provided in Table 7.

The calculated score values are given in Table 9. Furthermore, the graphical illustration is shown in Fig. 4.

The proposed q-RLDF WASPAS approach provides a structured mechanism for evaluating, ranking, and recommending AI algorithms for PE by systematically capturing the interplay of multiple performance criteria in uncertain situations. Based on the final ranking obtained through our proposed methodology, the CNN-MA emerged as the most effective alternative, demonstrating superior capability in real-time motion tracking and posture analysis, which directly supports physical skill development and form correction. The RL-TO ranked second, reflecting its strong potential to personalize training regimens and adapt workout plans dynamically according to individual learner needs. The HAI-NLP secured the third position, showing promise in enhancing interaction and providing customized feedback through natural language interfaces. The remaining alternatives, WSDMA and the ES-EP, are positioned lower due to relatively limited adaptability and narrower application scopes within dynamic PE contexts. This ranking provides practical guidance for educators and administrators to prioritise AI solutions that best align with optimizing training effectiveness, learner engagement, and real-time performance monitoring in PE environments.

The Wizard of Oz is back. Audiences no longer just see the tornado that ripped Dorothy Gale from Kansas to Munchkinland. Now they can experience it thanks to powerful fans creating hurricane-force winds that tousle the audience’s hair and lead some to cover their faces, while the room is covered in confetti like debris inside the eye of the storm. The Las Vegas Sphere, the enormous entertainment venue opened two years ago in the American casino city, has injected new life into the influential 1939 classic with the promise of turning what was once a Technicolor trip to another world into a sensory experience. To look to the future, Hollywood is turning to the past.

Judy Garland, who died 56 years ago, is no longer the big star of this new era. The actress shares the spotlight with artificial intelligence, a tool that dared to imagine what was off-screen in Victor Fleming’s Oscar-winning film. The premiere of the new version last Thursday will fuel the existential debate about the future of cinema in the age of AI, a topic that sparked a 148-day strike in Hollywood in 2023.

“We’re breaking a lot of ground with this. We truly believe we’re opening a new chapter in the way stories are told. AI will change the film industry. And I think what we’ve done is push it to its limits,” Jim Dolan, CEO of Madison Square Garden Entertainment and owner of the Sphere, told EL PAÍS in Las Vegas. The company invested $100 million, more than double the initial budget, in making this version, shattering the original’s $2.8 million ($76 million at today’s exchange rate).

The Wizard of Oz was first shown at Grauman’s Chinese Theater in the heart of Hollywood. In the late 1930s, the theater had a seven-meter-high screen. It was enough to dazzle audiences with its transition from sepia tones to brilliant colors thanks to the then-revolutionary technology of Technicolor, which combined three color negatives into one. That first showing would be like watching a movie on a cell phone today. Oz now exists on a screen 111 meters high and 157 meters wide.

Dolan opened the Sphere in 2023 with an immersive show by the Irish group U2. Since then, the venue has focused on musical acts, from veteran rockers the Grateful Dead and the Eagles, to the jam band Phish, to the current Backstreet Boys residency. The site has previously screened a film, Darren Aronofsky’s Postcard from Earth, but this is the first time a classic has been adapted for this screen, the size of three football fields. According to the entrepreneur, 200,000 tickets have already been sold for screenings over the coming weeks. The cheapest costs $104. The movie will have three showtimes and will be available at the new Abu Dhabi Sphere, which is due to begin construction soon.

To adapt The Wizard of Oz to this giant format, with 16K definition, the Sphere enlisted Ben Grossmann, a visual effects supervisor who won an Oscar in 2012 for Martin Scorsese’s Hugo. Grossmann hired an army of 1,000 artists, who went to the MGM archives and reviewed the original script, technical shoot guides, production reports, and outtakes. All of this was then entered into Google’s DeepMind program. Nearly another 1,000 engineers and creatives spent more than two years shaping this new digital world until arriving at the version that premiered last week.

“We needed to see things that were there in the original, but that you couldn’t see through the camera. It’s not about photography, it’s about taking you there. We adopted a philosophy as if we had invented a time machine that transported us to the world they were imagining,” Grossmann, who serves as producer, explained.

The technology broadens the audience’s view, allowing them to see, to their left and right, animals on Dorothy’s farm that didn’t exist in the original film, or characters who had been left out of frame in important scenes. The castle interiors in this version feature ceilings that are tens of meters high, making the audience feel as if they’re right there with the protagonists. The Wizard’s face has the imposing size of a small building, which successfully conveys to the audience the fear Dorothy and her friends felt.

The experience is also complemented by snowflakes falling, a shower of rubber apples, pyrotechnic effects, and a series of enormous blue flying monkeys hovering above the audience in one of the film’s most iconic scenes. The audience’s seats vibrate in the tense moments.

Some visual effects, however, take people away from the story or prove distracting. The digital makeup sometimes makes Dorothy and the actors in costume look like caricatures. Most disturbing are the faces of the extras and the new people digitally created on the sides to fill the enormous screen. Some have familiar faces. Others are blobs reminiscent of Borgia’s Ecce Homo.

Dolan confesses that his face briefly appears over one of these characters. The production team also inserted the face of executive David Zaslav, president of Warner Bros. Discovery, a key figure in approving the use of AI in the film. In old Hollywood, products were inserted into the background of films. In the age of artificial intelligence, it’s the faces of moguls.

Grossmann points out that the design team’s work isn’t over. “There’s no finish line. The original filmmaking team didn’t finish the 1939 project either; they just abandoned it and moved on when shooting time ran out. And as I say this, people are still working on our film. Not because we’re not ready for the premiere, but because we have to keep working on it for next month and the month after that,” he said in an interview in the Sphere.

Because of this constant digital retouching, Grossmann isn’t bothered by the catastrophic claims that artificial intelligence will eventually replace filmmakers. “We ended up giving more work to more artists on this project than Hollywood has employed in recent years,” he says. The special effects specialist has worked on Sin City, Shutter Island, and Alice in Wonderland, but he’s never worked on a production as large as this one in his 30-year career.

Grossmann also has to defend the integrity of the new work. The machine creates the information that was left out of the original frame. “It does this because this model knows what Judy Garland’s eyelashes looked like that day, what scene it was,” he notes. “I think these tools will allow for new waves of creativity and expression. Art will remain art.”

An impact that is still valid

The allure of the Oz universe for Americans is far from fading. In December of last year, one of the four pairs of slippers Judy Garland wore during the filming of the 1939 movie was auctioned off. The shoes sold for $28 million, making them the highest-valued piece of film memorabilia, shattering the record set in 2011 by the white dress worn by Marilyn Monroe in The Seven Year Itch. Auction house Heritage estimated the slippers would fetch $3 million, but the bidding surpassed that figure in mere seconds.

Last year, Universal released Wicked, the Oscar-winning film version of the Broadway musical inspired by The Wizard of Oz. The film grossed $755 million at the global box office, becoming the sixth highest-grossing title in the studio’s history. The sequel, also directed by Jon M. Chu, will be released on November 21. This one will have to compete with the original, which was scaled up thanks to digital anabolics.

Sign up for our weekly newsletter to get more English-language news coverage from EL PAÍS USA Edition