Analyse how digital systems change the ways people learn in formal, non-formal and self-directed education, and how science creates and validates new knowledge, using AlphaFold and real data.
Each is in the Digital Society glossary, with every other term in the subject.
An approach in which software uses data about a learner's responses to adjust the content, difficulty and pace of what it presents next.
A mode of education that combines face-to-face teaching with online learning, with some control for the learner over the time, place or pace of the online part.
Scientific research in which members of the public take part, for example by collecting observations, classifying data or analysing results, usually in partnership with professional scientists.
Obtaining ideas, labour, data or funding from a large and often anonymous group of people, usually through an online platform, rather than from a small group of employees or experts.
The skills needed to find, evaluate, create and communicate information safely and effectively using digital technologies.
The study and practice of teaching approaches that use digital technologies to support learning, chosen for how they help students learn rather than for the technology itself.
Learning delivered or supported through electronic devices and digital networks, such as online courses, educational apps and virtual classrooms.
A form of blended learning in which students study new content at home, usually through short videos, and use class time for practice, discussion and support from the teacher.
Define formal, non-formal, informal and post-formal education, and distinguish between them with real-world examples.
Describe how digital systems are used to design and deliver formal education, including remote, blended and flipped models.
Explain the impacts and implications of remote learning during COVID-19 school closures for different people and communities, using real data.
Outline approaches to skill training, competency development and self-directed learning, and analyse the role of MOOCs, micro-credentials and learning apps.
Explain what a digital pedagogy is and evaluate the evidence for and against specific digital pedagogies.
Describe approaches to scientific and technology research and development, including open-source and iterative development, simulation, citizen science, crowdsourcing, open access and AI.
Analyse how digital systems change the ways new knowledge is validated, including peer review and preprints.
Evaluate the impacts and implications of AlphaFold for scientific research and for the people and communities it affects.
Apply a step-by-step method to analyse an unseen real-world example in the human knowledge context.
The human knowledge context covers the ways people and communities learn and the ways they create new knowledge. Both have been reshaped by digital systems, from the learning apps on a student's phone to the AI models that now predict the shapes of proteins for laboratories around the world.
As a context, human knowledge situates the course's content and concepts. Learning platforms rely on data (3.1), algorithms (3.2) and networks (3.4); research increasingly depends on simulation and artificial intelligence (3.6). The context is also a natural place to apply the concepts of change, power, and values and ethics, because who learns what, who produces knowledge and who controls it are questions of fairness and influence. It connects with Theory of Knowledge (TOK), which asks how people know what they claim to know.
With a free account, Augment remembers where you stopped and what you have practiced.
Structured learning in an organised system of schools, colleges and universities that follows a set curriculum, is taught by qualified teachers and leads to recognised qualifications.
The use of game elements, such as points, badges, levels, streaks and leaderboards, in non-game activities such as learning, to increase motivation and engagement.
Learning that happens through everyday activities, conversations and media use, without a set curriculum, teacher or qualification.
Software that schools and organisations use to deliver, organise and track courses, including resources, assignments, communication and grades.
An online course open to unlimited numbers of participants over the internet, usually with video lectures, automated assessment and online discussion, and often free to join.
A certificate that records the completion of a short, focused course in a specific skill or competency, often earned online and designed to be combined with others.
Organised, intentional learning outside the formal school and university system, such as workplace training, community courses, coding bootcamps and online courses that do not lead to a full formal qualification.
Making research outputs, especially journal articles, freely available online to anyone, without subscription charges, often with permission to reuse them.
Teaching and learning materials in any medium that are free to access and released under an open licence, so that others can reuse, adapt and redistribute them.
The process in which experts in a field evaluate a research paper's methods, evidence and conclusions before it is published, to judge its quality and validity.
Learning that adults undertake after completing their initial formal education, such as professional development, retraining and learning for personal interest, often in non-formal or informal settings.
A complete version of a research paper that authors share publicly, usually on an online server, before it has been peer reviewed or published in a journal.
Education that takes place when learners and teachers are in different physical locations, delivered through any medium, from online platforms to radio, television and printed packs.
Systematic work carried out to increase knowledge (research) and to use that knowledge to create new or improved products, processes and services (development).
Learning in which individuals take the initiative to decide what to learn, find resources, set goals and judge their own progress, with or without help from others.
The use of a computer model to imitate the behaviour of a real system over time, so that researchers can test scenarios that would be impossible, unsafe, slow or expensive to test in reality.
Knowledge in this context has two sides. The first is learning: individuals acquire knowledge and skills that already exist. The second is knowledge creation: researchers, companies and communities produce, test and share knowledge that did not exist before. The two are linked, because new knowledge eventually becomes the content that others learn.
Digital systems store knowledge (digital libraries, databases), transmit it (video platforms), personalise it (adaptive algorithms), generate it (simulations and AI models) and judge it (plagiarism detectors, citation metrics). Each role brings opportunities and risks for different people and communities.
Content area | Role in the human knowledge context | Real-world example |
|---|---|---|
3.1 Data | Records of learner progress; large research data sets | Learning analytics in school platforms; the Protein Data Bank |
3.2 Algorithms | Personalising practice, ranking search results, detecting plagiarism | Spaced-repetition scheduling in Duolingo |
3.3 Computers | Devices for learners; supercomputers and cloud computing for research | Uruguay's Plan Ceibal laptops |
3.4 Networks and the internet | Delivering remote learning; sharing papers and data worldwide | arXiv preprint server |
3.5 Media | Video lessons, podcasts, open textbooks | Khan Academy videos |
3.6 Artificial intelligence | AI tutors; machine learning as a research method | Khanmigo; AlphaFold |
Digital systems from across the content areas play a part in learning and research.
Formal education is the main way societies pass knowledge to the next generation. It is usually organised by age and grade, publicly regulated and, at primary and lower secondary level, compulsory in most countries. The IB Diploma is an example.
Digital systems are now part of almost every stage of formal education. Their role can be divided into design (planning what is taught, how and how it is assessed) and delivery (how teaching reaches learners).
A learning management system (LMS) such as Moodle, Canvas or Google Classroom organises a course online. Teachers post resources and tasks, students submit work, and the system records grades and deadlines. Because every action is logged, an LMS also produces data about learners, which schools can analyse to spot students who are falling behind.
E-learning resources, such as interactive simulations, video lessons and online quizzes, are increasingly part of the design of courses rather than additions to them. Many are open educational resources (OER): teaching materials released under open licences so that anyone can use, adapt and share them. OpenStax, a non-profit based at Rice University in the USA, publishes free, openly licensed university textbooks, which lowers costs for students who might otherwise pay for printed textbooks.
Assessment design has also changed. Online tests can be marked automatically, and plagiarism detection software compares student work with large databases of texts. These tools save teacher time, but they also shape what is assessed: skills that are easy to mark automatically, such as factual recall, may receive more attention than argument or creativity.
Formal education can be delivered in several modes. Two questions separate them: do teacher and learners share the same place, and do they learn at the same time? Learning at the same time is synchronous (a live lesson or video call); learning at different times is asynchronous (recorded videos, tasks completed at the learner's own pace).
Remote learning is broader than online learning. During COVID-19 school closures, many countries used television and radio because large numbers of households had no internet connection. E-learning, by contrast, refers specifically to learning delivered through electronic devices and networks, whether in a classroom or at home.
Blended learning combines face-to-face teaching with online activities. A common form is the flipped classroom: students watch a short video or read material at home before the lesson, and class time is used for discussion, problem-solving and feedback.
The flipped model rests on simple reasoning. Listening to an explanation is something a student can do alone, at their own speed, with the option to pause and replay. Applying an idea to a hard problem is where students most need a teacher. Flipping moves the easier activity out of the classroom and the harder activity into it.
The COVID-19 pandemic produced the largest experiment in remote learning in history. In April 2020, closures affected more than 1.5 billion learners in over 190 countries, around nine in ten of all students enrolled worldwide. By January 2021, UNESCO calculated that schools had been fully closed for an average of about 14 weeks worldwide, rising to about 20 weeks in Latin America and the Caribbean.
Governments responded with whatever systems they had. Some used online platforms; others broadcast lessons on television and radio or distributed printed materials. Access to these systems was very unequal. A UNICEF analysis in August 2020 estimated that at least 463 million schoolchildren, about 31% of those affected, could not be reached by any digital or broadcast remote learning.
The regional pattern in the chart follows the digital divide: regions with lower household access to electricity, internet, computers and televisions had the largest share of students who were not reached. Younger children were also more affected; UNICEF estimated that about 70% of pre-primary children could not be reached, compared with about 18% of upper-secondary students, because remote learning for young children depends heavily on adult support.
Remote learning affected different groups in different ways.
People and communities | Impacts and implications |
|---|---|
Students in poor and rural households | Often no device, no connection or one shared phone; many fell behind or left school, widening existing gaps |
Girls in some low-income countries | Greater pressure to do household work while at home; higher risk of not returning to school |
Students with disabilities | Some platforms lacked captions, screen-reader support or sign language; others gained flexibility from learning at their own pace |
Teachers | Had to redesign lessons for new platforms quickly, often with little training; workload and stress rose |
Parents and carers | Became supervisors of learning, which was harder for those working outside the home or with limited education themselves |
Ed-tech companies | Rapid growth in users; platforms such as Zoom and Google Classroom became part of everyday school life |
Earlier investment left some countries better prepared. Uruguay had given a laptop to every public school student through Plan Ceibal since 2007 and had a national learning platform, CREA. When schools closed in March 2020, a programme called "Ceibal en casa" (Ceibal at home) moved teaching online; within weeks, use of CREA rose sharply. Telecom operators also agreed that the platform could be used without counting against students' mobile data allowances.
Through the concept of change, COVID-19 is often described as a moment of discontinuity: a sudden break that forced schools to adopt technologies they had been slow to use. Some changes lasted, such as online submission of work; others reversed once schools reopened, which suggests that the pandemic speeded up existing trends more than it transformed formal education.
The evidence on technology in classrooms is mixed. The OECD's PISA 2022 survey found that, on average across OECD countries, 65% of 15-year-olds reported being distracted by using digital devices in at least some mathematics lessons. Students who were distracted by other students' devices scored 15 points lower in mathematics than those who rarely were, even after taking social background into account, which the OECD describes as about three-quarters of a year of learning. This is a correlation, not proof that devices cause lower scores, but it has encouraged many countries to restrict phones in schools.
UNESCO's 2023 Global Education Monitoring Report, Technology in education: A tool on whose terms?, reached a similar conclusion from a wider review. It found that robust evidence on the added value of education technology is limited, and much of it comes from the companies that sell the products. The report recommended that technology be introduced only where evidence shows it is appropriate, equitable, scalable and sustainable, and that it should support, not replace, teachers.
Not all learning happens in schools. Education researchers usually distinguish three forms.
Informal learning is the third form. It is unplanned and often not intended as "study": a teenager watching a YouTube video to fix a bicycle, a gamer learning English through online play, or a patient reading about a diagnosis. Digital systems have expanded informal learning enormously, because explanations of almost any topic are now a search away.
Post-formal education refers to learning after a person has completed initial formal schooling and higher education. It includes training at work, retraining for a new career, professional development and learning for personal interest in later life. The idea of lifelong learning underlies it: in economies where technologies change quickly, skills learned at 18 may be out of date at 40.
Skill training teaches a specific, practical ability, such as using a spreadsheet, coding in Python or operating a machine. Competency development is broader: a competency combines knowledge, skills and attitudes needed to perform well in a role, such as project management or digital collaboration. Employers increasingly describe jobs in terms of competencies rather than qualifications.
Digital systems support both. Employees complete online training modules through workplace platforms such as LinkedIn Learning; simulations train pilots and surgeons in safe virtual environments; and short courses are packaged as micro-credentials.
Governments also fund post-formal learning directly. Singapore's SkillsFuture programme gave every citizen aged 25 or over an opening credit of S$500, introduced in 2015, to spend on approved courses. In 2024 it added a top-up of S$4,000 for citizens aged 40 or older to support mid-career retraining. Through the concept of change, such schemes respond to automation and the rise of AI, which alter the skills that workers need throughout their careers.
A massive open online course (MOOC) is an online course open to anyone, usually free or cheap to join, with video lectures, quizzes and discussion forums that can serve thousands of learners at once. 2012 was called "the year of the MOOC": Coursera was founded by two Stanford computer scientists, and edX by Harvard University and MIT. Coursera reported 162 million registered learners by September 2024.
MOOCs were first presented as a way to bring elite university teaching to anyone with an internet connection. The results have been more limited. A 2019 study in Science by Justin Reich and José Ruipérez-Valiente, using data from Harvard and MIT courses on edX, found that completion rates remained very low over six years, that most learners did not return after their first year, and that growth in participation was concentrated in the world's most affluent countries. Many MOOC providers have since shifted towards paid professional certificates and online degrees.
Claim about MOOCs | Counter-claim and evidence |
|---|---|
They open elite education to anyone | Most learners are already well educated and live in wealthy countries |
They are free | Courses are often free to view, but certificates and degrees cost money |
They let people learn at their own pace | Self-paced courses have very low completion rates without deadlines or support |
They help people change careers | Certificates are valued by some employers, but less than formal degrees in many sectors |
Sources: Reich and Ruipérez-Valiente, Science, 2019; Coursera company reports, 2024.
Self-directed learning is the oldest form of learning, but digital systems have made it far easier. Language apps, video tutorials, podcasts, online forums and AI chatbots allow learners to follow their own interests at their own pace. Duolingo, a language-learning app launched in 2012, had an average of 116.7 million monthly active users and 40.5 million daily active users in the last quarter of 2024.
Duolingo shows how design shapes self-directed learning. Lessons are short; algorithms schedule review of words just before a learner is likely to forget them; and features from games, such as daily "streaks", points and leaderboards, encourage learners to return each day. These features keep many learners motivated, but critics argue they reward daily activity more than deep learning, and that the app is better at building vocabulary than at developing conversation.
Self-directed learning also depends on digital literacy: the ability to find, evaluate and use information with digital tools. A learner who cannot judge the reliability of a video or recognise a confident but wrong chatbot answer may learn misinformation as easily as knowledge.
Pedagogy is the method and practice of teaching: how a teacher organises learning so that students understand. A digital pedagogy is a teaching approach designed around what digital systems make possible, which means more than simply using technology in lessons.
The same tool can support very different digital pedagogies. A tablet used to read a digitised textbook changes the medium but not the teaching; a tablet used to collect field data for a shared class map changes what students do.
Digital pedagogy | How it works | Real-world example | Main implication |
|---|---|---|---|
Flipped classroom | Content at home by video, practice in class | Khan Academy videos used before lessons | More class time for support; depends on home access |
Adaptive learning | Algorithms adjust questions and pace to each learner's answers | Duolingo's review scheduling; adaptive maths platforms | Personalisation; learner data collected continuously |
Gamification | Game features (points, streaks, badges, levels) added to learning | Duolingo streaks and leagues | Motivation; risk of rewarding activity over understanding |
Collaborative online learning | Students build knowledge together on shared platforms | Shared documents, class wikis, online discussion | Develops collaboration; unequal contributions |
Mastery learning | Students move on only after showing mastery of each step | Khan Academy mastery system | Fills gaps; harder to keep a class together |
AI tutoring | A chatbot guides students with questions rather than giving answers | Khanmigo (Khan Academy, 2023) | One-to-one support at scale; accuracy and dependence risks |
Adaptive learning works through a feedback loop. The system presents a task, records the response, updates its estimate of what the learner knows and chooses the next task. Through the concept of systems, the learner and the software form one system in which each output becomes the next input.
Generative AI has added a new digital pedagogy. Khan Academy, a non-profit founded by Sal Khan in 2008, launched Khanmigo in 2023, an AI tutor built on a large language model. It is designed to ask students questions that lead them towards an answer instead of simply supplying it, following the Socratic method.
YouTubeSupporters argue that AI tutors could give every student the personal tutoring once available only to wealthy families. Critics raise several counter-claims. Chatbots can produce hallucinations, confident statements that are false, which a struggling student may not detect. Students may use AI to avoid thinking rather than to support it. AI tutors also collect detailed data about children's learning and mistakes, which raises questions of privacy and ownership. Through the concept of power, the spread of AI tutors may also shift influence over what and how children learn from teachers and education ministries to a small number of technology companies.
New knowledge in science and technology comes mainly from research and development (R&D). Universities and public research institutes carry out much of the basic research, which aims to understand how the world works without a direct application. Companies carry out more applied research and development, which turn knowledge into products and processes.
R&D is funded by governments, companies, charities and international bodies. The source of funding matters, because funders shape which questions are asked and who owns the results. Publicly funded research is often expected to be openly shared; company research is often protected by patents or kept secret to gain commercial advantage. AlphaFold, discussed in the case study, sits between the two: it was built by a company, Google DeepMind, on decades of publicly funded data.
Technology R&D follows several approaches of its own. Corporate laboratories employ researchers to work on long-term problems; Bell Labs in the USA, for example, invented the transistor in 1947. Open-source development spreads the work across volunteers and companies who share their code openly, as with the Linux operating system, started by Linus Torvalds in 1991 and now maintained by thousands of contributors. Iterative development builds a simple prototype, tests it with users and improves it in short cycles, so products change in response to evidence about how people actually use them.
Digital systems have made iteration much faster. Online services can run A/B tests, showing two versions of a feature to different groups of users and keeping the one that performs better. This produces knowledge quickly, but it is knowledge owned by the company and rarely published, and users are often unaware that they are taking part in an experiment. Through the concept of values and ethics, this raises questions about consent that formal scientific research handles through ethics committees.
Scientists use several approaches to research, and digital systems have changed each of them. Many fields now combine several approaches, so that results from one approach are checked against another.
Approach | What it involves | Role of digital systems | Example |
|---|---|---|---|
Experimental | Controlled tests of a hypothesis | Sensors, automated laboratories, data logging | Robotic high-throughput screening of drug candidates |
Observational | Measuring the world without changing it | Satellites, telescopes, environmental sensor networks | Sky surveys that image billions of objects |
Theoretical and computational | Building mathematical models | Supercomputers solve equations that cannot be solved by hand | Climate models |
Simulation | Running a model of a system to test scenarios | Virtual experiments that would be impossible, unsafe or too slow in reality | Epidemic models during COVID-19 |
Data-driven and AI | Finding patterns in very large data sets | Machine learning, cloud computing, big data | AlphaFold predicting protein structures |
Participatory | Involving the public in collecting or analysing data | Online platforms, smartphone apps, games | Galaxy Zoo; Foldit |
The research process can be seen as a cycle. Each stage now relies on digital systems, and validation is the stage that decides whether a result is accepted as knowledge.
A simulation is only as good as the model behind it and the data it uses. Every model simplifies reality, so its outputs are projections under assumptions, not observations. In March 2020, a model from Imperial College London projected that an unmitigated COVID-19 epidemic could cause around 510,000 deaths in Great Britain. The projection influenced the UK government's decision to introduce a national lockdown. Critics argued that the code was poorly documented and its assumptions too pessimistic; an independent check later reproduced its results, but the debate showed how much power a model can have over public policy.
Climate science depends on simulation even more. Results from many international modelling centres are compared to judge how confident scientists can be in projections of future warming.
Digital networks allow very large numbers of non-professionals to take part in research.
Galaxy Zoo, launched by researchers at the University of Oxford in July 2007, asked volunteers to classify images of galaxies by shape. Its organisers hoped for 20,000 to 30,000 volunteers; more than 150,000 took part within a year. In 2007 a Dutch schoolteacher, Hanny van Arkel, noticed an unusual green object in one image; now called Hanny's Voorwerp, it turned out to be a large cloud of gas lit up by radiation from a nearby galaxy's black hole. Galaxy Zoo grew into the Zooniverse platform, which had more than 2.7 million registered volunteers by 2025 and hosts projects from transcribing historical documents to counting penguins.
Citizen science is one kind of crowdsourcing, which more broadly means obtaining ideas, work or data from a large, open group of people over the internet.
Foldit, an online puzzle game created at the University of Washington, shows the power of crowdsourcing in a field close to the AlphaFold case study. Players fold three-dimensional models of proteins and earn points for stable shapes. In 2011, players produced models of an enzyme from a monkey virus related to HIV, the M-PMV retroviral protease, in about three weeks, after scientists had failed to solve its structure for more than a decade. The players were named as co-authors in the journal article.
Citizen science has limits. Volunteers are not evenly spread around the world, so data can be biased towards wealthy and urban areas. Tasks such as classifying galaxy images are increasingly automated using labels that volunteers created, which raises questions about recognition for volunteer work.
The traditional route for sharing research is a journal article. Many journals charge readers or libraries for access, which prices out researchers and the public in low-income countries. Open access makes research free to read online.
Open access takes two main forms. In "gold" open access, the journal makes the article free, often charging authors an article processing charge; some prestigious journals charge several thousand euros per article, which shifts the barrier from readers to authors. In "green" open access, authors place a copy in a free online repository. In 2018, a group of research funders called cOAlition S launched Plan S, requiring research they fund to be published open access from 2021.
A preprint is a version of a paper shared publicly before peer review. The arXiv server, started in August 1991 for physics, now covers mathematics, computer science and other fields; it received 24,226 new submissions in October 2024 alone, and its total passed 3 million papers in 2026. Preprints spread findings quickly, which proved valuable during the COVID-19 pandemic, but they can also spread errors before experts have checked them.
Peer review is the main way the scientific community decides whether a claim counts as knowledge. It is not perfect: reviewers are unpaid volunteers, reviews take months, and studies in several fields have failed to replicate when other researchers repeated them. Digital systems help (online submission, plagiarism checks, open data that lets others re-run analyses) and also add strain. In late 2025, arXiv's computer science section began requiring review articles and position papers to have passed peer review before submission, because moderators could not cope with a flood of low-quality survey papers, many apparently written with generative AI.
Machine learning is now a research method in its own right. Instead of starting with a theory and testing it, researchers train a model on large amounts of data and use it to predict results, which are then tested experimentally. This approach has produced results in materials science, weather forecasting and biology, where AlphaFold is the best-known example.
AI-based research raises questions about explanation. An artificial neural network can make accurate predictions without giving reasons a human can follow, which makes it a kind of black box. Scientists value understanding why something happens, not only what will happen, so the rise of accurate but opaque models challenges what counts as scientific knowledge.
Proteins are large molecules that do most of the work in living cells. They digest food, carry oxygen, fight infection and copy DNA. Each protein is a chain of building blocks called amino acids, which folds into a precise three-dimensional shape. The shape determines what the protein does, so knowing it helps scientists understand diseases and design medicines that fit into a protein like a key in a lock.
In 1972, the chemist Christian Anfinsen received a Nobel Prize for showing that a protein's amino acid sequence determines its shape. In principle, then, it should be possible to predict the shape from the sequence. In practice the problem proved extremely hard, because a chain can fold in an astronomical number of ways. For about 50 years scientists determined shapes experimentally, using methods such as X-ray crystallography and cryo-electron microscopy, which often took months or years of work for a single protein.
The results of these experiments were collected in the Protein Data Bank (PDB), a public archive started in 1971. By the end of 2024 it held more than 220,000 experimentally determined structures, contributed by laboratories around the world and mostly paid for by public research funds. That is a tiny fraction of the hundreds of millions of known protein sequences.
To measure progress on prediction, scientists set up CASP (Critical Assessment of protein Structure Prediction) in 1994. Every two years, teams predict the structures of proteins whose shapes have been determined experimentally but not yet published, and independent assessors score the predictions.
Google DeepMind, an AI research company owned by Alphabet, entered CASP in 2018 with the first version of AlphaFold and won. At CASP14 in 2020, its new system, AlphaFold 2, achieved a median score of 92.4 out of 100 on the main accuracy measure (GDT). A score of about 90 is generally treated as comparable to experimental methods. The CASP organisers described the protein structure prediction problem as largely solved for single protein chains.
AlphaFold 2 combines several digital systems.
Input data. The system takes a protein's amino acid sequence, a string of letters.
Database search. It searches large sequence databases for related proteins in other species. Amino acids that change together across evolution are likely to sit close to each other in the folded shape, which gives clues about the structure.
Neural network. A deep learning model, trained on about 170,000 structures from the PDB, processes the sequence and the evolutionary clues and predicts the position of every atom.
Confidence score. For each part of the structure, AlphaFold gives a confidence score (pLDDT, from 0 to 100), so researchers know which parts to trust.
Distribution. In July 2021 DeepMind published the method in Nature and released its code. With the European Bioinformatics Institute (EMBL-EBI), it launched the free AlphaFold Protein Structure Database, which grew in July 2022 to more than 200 million predicted structures, covering almost every catalogued protein known to science.
In 2024, Demis Hassabis and John Jumper of Google DeepMind shared half of the Nobel Prize in Chemistry "for protein structure prediction". The other half went to David Baker of the University of Washington for computational protein design; his laboratory also developed an open alternative, RoseTTAFold, and the Foldit game.
YouTubePeople and communities | Impacts (what has happened) | Implications (opportunities and risks) |
|---|---|---|
Structural biologists | Predicted models used as starting points for experiments; structures that resisted experiments solved faster | Shift from determining structures towards explaining function; some fear a loss of experimental skills and funding |
Researchers in low- and middle-income countries | Free access to structures that once required expensive equipment | Opportunity to work on local diseases; still need computing power, training and laboratories to test predictions |
Patients with neglected diseases | The Drugs for Neglected Diseases initiative (DNDi) uses AlphaFold in work on medicines for Chagas disease and leishmaniasis | Potential new treatments for diseases that receive little commercial investment; benefits are years away and not yet proven |
Environmental researchers | The Centre for Enzyme Innovation at the University of Portsmouth uses AlphaFold to engineer enzymes that break down plastics | Possible new recycling methods; laboratory results still need scaling |
Pharmaceutical companies and Alphabet | Isomorphic Labs, an Alphabet company founded in 2021, uses AlphaFold-based methods for drug discovery and signed agreements with Eli Lilly and Novartis in 2024 | Commercial value created from tools trained on public data; questions about who benefits |
Students and teachers | Free, visual protein models for classrooms | Learning biology through predicted rather than measured structures; need to understand the limits of predictions |
AlphaFold also has clear limits. It predicts a single static shape, while many proteins move and change shape as they work. Some regions of proteins have no fixed shape at all, and AlphaFold gives them low confidence scores. It does not explain the physical process of folding. Its predictions are hypotheses that still need experimental testing, especially before they are used to design medicines.
Through the concept of change, AlphaFold represents a discontinuity in structural biology. For five decades, knowledge of protein structures grew one experiment at a time, reaching about 200,000 structures by the early 2020s. Within two years, predicted structures for almost every known protein became available. Some scientists describe this as a transformation of biology into a field where structure is a starting point rather than an end goal. Others argue it is better seen as an evolution: AlphaFold built on 50 years of experiments, databases and earlier computational methods, and laboratory work remains essential.
Through the concept of power, AlphaFold raises questions about who controls scientific knowledge. The training data came from the public PDB, built by thousands of researchers, yet the model that turned those data into a breakthrough was owned by one of the world's largest companies. DeepMind chose to release AlphaFold 2's code and database freely, which spread its benefits widely. With AlphaFold 3, announced in May 2024, it at first offered only a restricted web server and no code. Many scientists signed an open letter arguing that this prevented others from checking and building on the work. DeepMind released the code for academic use in November 2024, but the model's parameters are licensed only for non-commercial research.
Through the concept of values and ethics, the case raises the question of what openness in science requires. Science depends on results being checkable; a claim that cannot be reproduced is weak evidence. Companies, however, need returns on their investment, and some argue that limiting access also reduces the risk that protein design tools are misused, for example to engineer harmful toxins. The tension is between the scientific value of openness, the commercial value of control and the safety value of caution.
Through the concept of systems, AlphaFold is one element of a larger research system. Its accuracy depends on the PDB (input), its usefulness depends on EMBL-EBI's free database and on laboratories that test predictions (outputs), and its results feed back as new experimental structures enter the PDB. Removing any element, such as public funding for the PDB, would weaken the whole system.
AlphaFold is one of the clearest cases of AI creating new scientific knowledge. Its impact on research is large: the database is used in more than 190 countries, and the Nobel committee described it as fulfilling a 50-year-old dream of predicting a protein's structure from its sequence. Its impacts on people outside research, such as patients, are still mostly implications rather than outcomes, because drug development takes many years.
The decision to release AlphaFold 2 and its database openly did more to spread its benefits than the technology alone. The more restricted release of AlphaFold 3 shows that this openness depended on choices by one company and is not guaranteed for future tools. A balanced judgement is that AlphaFold greatly expanded what scientists know about proteins and who can access that knowledge, but it also concentrated influence over a key research method in private hands, and its predictions remain hypotheses that experimental science must still validate.
A real-world example in this context may be unfamiliar: a new learning app, a national education platform, a citizen-science project or an AI research tool. A clear method makes it possible to analyse any example in depth.
Step | Question to ask | What to produce |
|---|---|---|
1. Identify | What is the digital system, who runs it, where and since when? | A precise description of the system |
2. Place | Is it about learning (formal, non-formal, informal) or creating and validating knowledge? Which content areas make it work? | The context link and the technical vocabulary |
3. Stakeholders | Which people and communities are affected, directly and indirectly? What claims does each make? | A list of named groups with their claims |
4. Impacts and implications | What are the positive and negative, intended and unintended effects, now and in the future? | Balanced points, each linked to a group |
5. Concepts | Which concept (change, power, values and ethics, systems, space, identity, expression) gives the most insight? | One or two concepts applied, not just named |
6. Evaluate | Under what conditions does the system succeed? For whom? Compared with what alternative? | A qualified judgement |
Paper 2 uses unseen sources; this question uses a short practice source.
I can define formal, non-formal, informal and post-formal education and classify a real-world example of each.
I can describe how digital systems such as learning management systems and open educational resources are used to design courses.
I can distinguish remote learning, e-learning, blended learning and the flipped classroom.
I can use UNICEF and UNESCO data to explain the unequal impacts of remote learning during COVID-19 school closures.
I can explain why Uruguay's Plan Ceibal helped formal education continue during closures.
I can outline skill training, competency development and micro-credentials, with an example such as SkillsFuture.
I can evaluate MOOCs using evidence on completion and on who participates.
I can explain self-directed learning and the role of digital literacy, using Duolingo as an example.
I can define digital pedagogy and evaluate adaptive learning, gamification and AI tutoring.
I can describe approaches to scientific and technology research and development, including open-source and iterative development, simulation and AI.
I can explain citizen science, crowdsourcing, open access and preprints with real-world examples.
I can explain how peer review validates knowledge and why digital systems put it under pressure.
I can analyse AlphaFold's impacts and implications for different people and communities through at least two concepts.
I can apply the six-step method to analyse an unseen real-world example in the human knowledge context.