The ConocimIA Seminar is pleased to present its first lecture, titled "The Disruptive Emergence of AI: An Exercise in Contextualization." This presentation will explore the context in which ChatGPT has emerged into society as an advanced Artificial Intelligence model, triggering a chain of disruptive reactions across the labor, professional, scientific, and academic sectors. The talk will address the origins and evolution of Artificial Intelligences, the technologies that have made them possible, as well as short- and medium-term perspectives related to AI.

  1. Date: November 17, 2023 / 12:00-13:00 h
  2. Location: Room V of Computer Science, Faculty of Documentation Sciences, UCM
  3. Admission: Free until capacity is reached
  4. Speaker: Prof. Manuel Blázquez, Department of Library and Information Science, Universidad Complutense de Madrid

Context and Purpose of the Lecture

This conference takes place at a unique juncture. At the end of 2022, OpenAI made ChatGPT publicly available—a language model that, while built upon decades of research in artificial intelligence, natural language processing, and information retrieval, emerged with an ease of use and responsiveness that surprised both users and experts alike. Within just a few months, public discourse shifted from asking “What is AI?” to questioning “How will AI change my work, my discipline, my life?” This conference arises from the conviction that, in the face of such a phenomenon, historical, technical, and social contextualization is not an academic luxury but a necessity. Disruption does not occur in a vacuum; it is the result of technological trajectories, research decisions, market dynamics, and geopolitical conditions that must be understood in order to anticipate, as far as possible, its future developments.

Development of the Presentation Content

1. The Long March Toward the Organization of Knowledge

The need to store, classify, and retrieve information is as ancient as writing itself. From the earliest libraries of antiquity—Alexandria, Nineveh, Pergamon—to modern classification systems such as the Universal Decimal Classification or subject heading lists, humanity has continuously developed systems to master the accumulation of knowledge. Each advancement in documentary media—from parchment to printed paper, from microfilm to digital formats—has brought new challenges in organization and access.

The emergence of web search engines in the 1990s marked a first major leap in accessibility. For the first time, a user could formulate a query in natural language and obtain, within seconds, a list of relevant documents distributed across the globe. However, this immediacy also fostered growing dependence on algorithmic intermediaries whose relevance criteria remained opaque. The user ceased to ask the librarian and began to ask the search engine, trusting that the machine would perform the selection task better and faster than any human.

2. From Expert Systems to Deep Learning

Artificial intelligence is not a recent phenomenon. Since the 1960s, expert systems have sought to encode human knowledge into logical "if-then" rules. These systems, widely used in medicine, geology, and engineering, demonstrated that certain reasoning processes could be automated, but their fragility was evident: beyond the narrowly defined domains for which they were designed, they lacked flexibility.

The shift toward machine learning transformed the paradigm. Instead of programming explicit rules, systems learned from examples. Supervised methods—classification, regression—and unsupervised methods—clustering, dimensionality reduction—enabled the resolution of problems previously considered intractable. Starting in the 2010s, deep neural networks (Deep learning) introduced the ability to learn hierarchical data representations through multiple layers that progressively extract increasingly abstract features.

Milestones such as IBM Watson—which in 2011 defeated human champions in the quiz show Jeopardy!—demonstrated the potential of these technologies applied to natural language processing. Watson combined information retrieval techniques, language processing, and machine learning to answer complex questions. However, its impact was then limited to specialized domains—medicine, finance—due to the cost of infrastructure and the complexity of implementation.

3. The Transformer Architecture and the Scale Shift

The true qualitative leap that led to ChatGPT occurred with the introduction of the Transformer architecture in 2017, in the seminal paper "Attention is All You Need". This architecture overcame the limitations of previous models—such as recurrent neural networks (RNNs) or long short-term memory (LSTM) networks—by enabling parallel processing of long text sequences and capturing dependencies between words regardless of their distance in the sequence. The attention mechanism allowed the model to weigh which parts of the input were most relevant for each element of the output.

From there, the race became one of scale. OpenAI released GPT-1 in 2018, GPT-2 in 2019—whose initial release was restricted due to concerns about malicious use—and GPT-3 in 2020, with 175 billion parameters. Each version exponentially increased processing capacity and output quality. The integration of reinforcement learning from human feedback (Reinforcement Learning from Human Feedback, RLHF) enabled fine-tuning of the models to follow instructions more accurately and generate more natural conversational responses.

4. The ChatGPT Moment: Convergence of Factors

When OpenAI launched ChatGPT in November 2022, several factors converged to make its impact qualitatively distinct from that of previous models:

  1. Accessibility: a free, conversational interface requiring no technical expertise, enabling anyone to interact with the model.
  2. Conversational fluency: specific training for dialogue generated coherent, contextually appropriate responses with a surprisingly human tone.
  3. Multitasking capability: the same model could draft texts, summarize documents, generate code, translate languages, solve mathematical problems, and much more, without requiring a change of tool.
  4. Demonstration effect: the ability for any user to personally verify the system’s capabilities generated viral diffusion unmatched by any previous technological product.

5. The socioeconomic and geopolitical context of its emergence

The emergence of ChatGPT does not occur in a social vacuum. This presentation situates this phenomenon within a global context characterized by:

  1. Pandemic crisis: COVID-19 had accelerated the digitalization of many areas and sparked reflection on the future of work and automation.
  2. Geopolitical tensions: the rivalry between the United States and China, the war in Ukraine, and the reconfiguration of global supply chains have positioned AI as a strategic field of international competition.
  3. Regulatory debates: while the European Union advanced its AI Act, other countries adopted more permissive approaches, such as Japan, which declared itself favorable to deregulation to attract investment.
  4. Concentration of technological power: the development of large language models requires billion-dollar investments that only a few corporations—OpenAI backed by Microsoft, Google, and Meta—can afford, raising questions about the concentration of control over these technologies.

6. Implications for Documentation Sciences

The conference pays special attention to the impact of generative AI on the discipline. The current capabilities of these models are profoundly transforming traditional information professional tasks:

  1. Cataloging and classification: AI can assign access points, subjects, and classification codes with a precision that, with appropriate training, matches or exceeds human performance. Customized GPT tools enable fine-tuning these processes for specific domains.
  2. Document analysis: Automated summarization, descriptor extraction, and identification of key concepts are tasks performed rapidly by models, freeing professionals to focus on higher-value activities.
  3. Information retrieval: The conversational interface radically transforms the search experience. Users no longer need to formulate queries in structured search languages; they can express their needs in natural language, with AI acting as an intermediary between the question and the documents.
  4. Creation of information systems: tools such as Sutro, automated application generation, and GPT customization enable the creation of specialized information systems without requiring advanced programming, democratizing the production of documentary solutions.
  5. Reference services: chatbots can answer frequent questions and simple inquiries, but they can also act as assistants in complex research tasks by suggesting sources, formulating search strategies, and synthesizing information.

Far from announcing an apocalyptic substitution, the hypothesis is proposed of a transformation in which the documentalist becomes a specialized mediator, a curator of interactions with AI. The new required competencies combine traditional disciplinary expertise with the ability to:

  1. Design effective prompts to obtain high-quality results.
  2. Critically evaluate AI outputs, identifying potential biases or errors.
  3. Monitor and audit automated systems to ensure their accuracy.
  4. Select and curate the datasets that feed the models.
  5. Integrate AI into complex documentary workflows.
  6. Train users in the responsible use of these tools.

7. Risks and Opportunities

The presentation does not evade the questions raised by this transformation:

  1. Labor risks: repetitive and procedural tasks are at risk of automation. This does not necessarily imply the disappearance of the profession, but rather a reconfiguration of its functions, with increased demand for specialization and critical thinking skills.
  2. Bias and quality: language models learn from available data, which reflect historical biases, inequalities, and imbalances in knowledge representation. AI may perpetuate or even amplify these biases if adequate oversight is not exercised.
  3. Intellectual property: training on copyright-protected texts and generating content that may resemble existing works raises unresolved legal challenges, as evidenced by lawsuits against OpenAI by authors and publishers.
  4. Concentration of power: the control of generative AI technologies by a few corporations raises questions about equitable access to knowledge and technological sovereignty of nations.
  5. Digital divide: the capacity to leverage these tools—in terms of access, training, and competencies—may widen the gap between those who can benefit from them and those who are excluded.
  6. Autonomy and delegation: the risk that users delegate critical thinking to AI by accepting its outputs without verification is a central concern, particularly in educational and research contexts.

8. Future perspectives

In the short term, generative AI is expected to be integrated into all productivity tools—word processors, spreadsheets, email—as well as the proliferation of domain-specific assistants. The ability of models to process increasingly longer contexts—128,000 tokens in GPT-4 Turbo—will enable work with complete documents, not just fragments.

In the medium term, the evolution toward multimodal systems—integrating text, image, audio, and video—will transform the very nature of interaction with information. The combination of generative models with reasoning and planning capabilities could lead to systems with greater autonomy, capable of executing complex tasks with minimal human oversight.

In the specific domain of Documentation Sciences, new areas of specialization are emerging: "IAmetría"—the study of how AI processes, relates, and produces knowledge—or algorithmic curation—the design and supervision of automated document management systems.

Space for Dialogue

Following the presentation, a Q&A and idea exchange session will be opened. The format aims to encourage active participation, allowing attendees to raise questions, share their experiences with these technologies, and discuss the concrete implications for their education and future professional trajectories. The conference does not aim to provide definitive answers, but rather to articulate the right questions at a time when many certainties are being reassessed.

This conference is part of the activities of the ConocimIA Seminar, a space dedicated to monitoring and analyzing artificial intelligence in the field of Documentation Sciences.

Conference Materials

The materials used in this session are available for download in PPTX format. The presentation captures the ideas, references, and open questions raised throughout the conference and may serve as a starting point for further exploration of the topics discussed or for use in educational contexts, provided proper attribution is given.

The file can be obtained via the following link: conocimIA_mblazquez_2023-11-17_irrupcion-disruptiva-ia.pptx