Mostrando entradas con la etiqueta AIEd. Mostrar todas las entradas
Mostrando entradas con la etiqueta AIEd. Mostrar todas las entradas

9 jun 2024

IA en las aulas: Lo impensable, lo impracticable y lo inexcusable

Publicado originalmente en Educadores 288

¿Qué nos espera: ser sustituidos por robots, como en la peor pesadilla, o una gran reducción de ratios, como en la carta sindical a los RRMM? Lo primero es impensable, pese a la ficción animada de los Supersónicos, la científica de Asimov y Clarke o la tecnosocial de Mitra y Seldon; lo segundo, no importa cuán sonoro sea el aplauso para la propuesta, es impracticable, porque una cosa es mejorar una ratio aquí o allá, o incluso que mejore sola porque decrecen los alumnos pero se blinda el empleo de los profesores, y otra bien distinta que traiga una mejora educativa eficaz y eficiente. No habrá sustitución porque, por lenta que sea procesando información normalizada en comparación con la artificial, la inteligencia humana es muy superior en la mayoría de las actividades y decisiones que más valoramos para el cuidado y la educación de nuestros menores y, además, la sabemos asociada a una moral compartida. Y no habrá piñata de profesores porque hay otras necesidades y prioridades sociales que la educación (la salud, la cuarta edad o los colectivos laborales a los que sí desplaza la innovación, por ejemplo) y mejores y no pocas maneras de invertir recursos en esta.

La mitad del argumento a favor de reducir ratios es correcta: el profesor no tiene tiempo para la diversidad que afronta, no digamos para una atención personalizada. Pero la medida está sobrevalorada. Aumentar el profesorado ni siquiera garantiza menores ratios, pues bien podría quedar en menos horas de clase por profesor. De no ser así, la aplicación más probable serían los desdobles, siempre con la tentación de (re)agrupar por niveles, en cualquier variante, para tratar a cada grupo de modo aún más homogéneo, y escasos efectos en términos de individualización. Pese a la insistencia sobre las ratios (o quizá por ello), se sabe muy poco sobre cómo utilizan los profesores el tiempo de clase y, por tanto, qué efectos reales podría tener tal reducción. Uno de los muy escasos estudios con datos al respecto, Beginning Teacher Education Study, con ya casi medio siglo y dedicado solo a la parte “académica” (lengua y matemáticas) de primaria y secundaria, señalaba que entre dos tercios y tres cuartos del tiempo de clase se dedicaba a trabajo individual, pero con escasa interacción con el profesor y compromiso por el alumno. Un interesante hallazgo era que, a mayor trabajo individual (deskwork), menor compromiso (engagement); éste aumentaba en general con la interacción, que, sin embargo, se localizaba en el trabajo grupal (la lección frontal, con preguntas, etc.).   

Paradójico, pero lógico: el profesor puede hacer de su lección un monólogo o un diálogo parcial con el grupo, pero, en cualquier forma que sea, le queda poco tiempo para repartir entre los alumnos individuales. Un estudio realizado de finales de los ochenta sobre el tiempo dedicado a los alumnos según el sexo, distinguía tiempo colectivo (grupal) e individual (dedicado a un alumno), a su vez subdividido por su finalidad disciplinaria o cognitiva (instructiva). Esta última subvariante, la que podría permitir un diálogo o atención personalizados al (permitir, pero no garantizar, pues también podría ser un monólogo o una mera repetición selectiva de la lección), ocupaba el 32% del tiempo codificado en el total de las tres categorías (menos deskwork que en los EEUU, pero esto es aquí secundario). Aplicado a una asignatura con tres horas de clase semanales en un aula de la ESO con 30 alumnos serían 1,9 minutos. Si consideramos que las “horas” escolares son de 50 minutos, se reduciría a 1,6; si descontamos el tiempo que, sencillamente, no es instructivo (esperas, desplazamientos, cambios de actividad, etc., en torno al 30% según el primer estudio citado), ya sería poco más de 1 minuto. Asi, un modesto aumento del 10% del profesorado requeriría un 8% más de gasto para conceder a cada alumno un par de segundos adicionales; partir los grupos por la mitad, como proponían alegremente unos expertos en la enseñanza de las matemáticas (doblando un profesorado que, por otra parte, se insiste en que escasean), aumentaría el gasto público dedicado en un 80% para arañar un minuto más. ¿Decepcionante?

Por supuesto, hay variantes más imaginativas que la lección e hincar los codos: codocencia, equipos de trabajo, estudio en parejas y más, cuestionarios y juegos en los libros de texto, apps más o menos interactivas, etc., aunque la mayoría no van muy lejos. Pero aquí es donde llega la IA, en particular generativa (que crea, IAG) y conversacional (IAC), la posibilidad de una conversación (es decir, interacción, retroalimentación, individualización, personalización…) inagotable. Podría ser pura cháchara, pero no lo es. Si tomamos como ejemplo ChatGPT, la IAC más popular, puede generar variantes sucesivas de una explicación, adaptarse a distintos niveles de edad o conocimiento, acercarse a los intereses del alumno, ejemplificar, preguntar, etc., sin límite ni impaciencia algunos, incluido no ir directamente a la solución sino paso a paso, explicar un razonamiento, detectar atascos, etc. Tiene limitaciones, por supuesto, pero, si no cada día, mejora cada mes, se adapta a las barreras propias de la vida escolar, su antropomorfismo resulta acogedor (especialmente para niños), empiezan a ofrecerse adaptaciones didácticas (Khanmigo, Duolingo, etc.). Más que un mentor, tutor o preceptor, dado que apenas tiene memoria de la conversación anterior en curso, puede considerarse un instructor ad hoc pero polivalente, además de siempre disponible.

Y, sí, le faltan datos, se equivoca, inventa, alucina, etc., pero no en el tipo de preguntas y respuestas, datos y teorías, temas y problemas del día a día en la escuela, un espectro de información y conocimiento en el que, en general, se desempeña bien. Por lo demás, para eso está el profesor, para proponer y encarrilar una combinación adecuada de su propia docencia y la algorítmica (ciborgdocencia, si se quiere un término inclusivo para ambas), para articular una docencia enriquecida y un aprendizaje aumentado. En matemáticas, por ejemplo (por volver al reputado ogro, la madre de todas las ratios y el último disgusto en PISA), se da la aparente paradoja de que ChatGPT es muy malo en el cálculo (después de todo, es un modelo de lenguaje, no lógico ni numérico, aunque puede trabajar en combinación con el plugin Wolfram o una simple calculadora virtual), pero muy bueno en la explicación –mejor que muchos profesores.

Ese es el futuro en que debemos y podemos ya trabajar. De los diálogos de Sócrates a hoy, pasando por el libro y luego el libro de texto, por no hablar ya de los medios broadcast, hemos ampliado la educación al precio de hacer de ella una actividad cada vez más unilateral, una enseñanza homogénea muy poco sensible a la diversidad del aprendizaje. La IAC, que apenas empieza, abre la vía a una interacción más equilibrada.

15 jul 2023

Can we trust ChatGPT or Bard... ever... or never ever? Two makeups for nothing



Yesterday, I accidentally stumbled on a partial reference to a book entitled Should You Believe Wikipedia? The topic seems somewhat dated, but I felt interested precisely because so I am on the trustfulness of Generative Artificial Intelligence, particularly Conversational AI, which today means mainly ChatGPT and Bard. I could have looked for some reviews through Google Scholar, or maybe look at the publishing house web, but I thought that asking for an abstract to ChatGPT could be easier, so I did. I prompted for an abstract of the book, specifying the whole title and subtitle, and I got it in seconds. It looked too conventional, the typical back cover text, so I prompted for abstracts by chapter, and I got them, nine short abstracts for nine chapters; then I felt curious about one of them, prompted for it, and so I get a longer abstract. There I stopped and, on the whole, the abstract didn't look mesmerizing but yes good enough to get the book. Wonderful, I thought, because this could be from now on the best and easiest way to decide about getting particular books about which I just had some scant reference. Then I bought it but didn't even open it, as I was busy with other tasks (in fact, this was a typical online distraction). It should wait on the ever-growing queue of books to read –or maybe not.

Some hours later, I had an idea: why if ask the same questions to Bard, and check the differences, if any? In fact, I had begun to use Bard the day before, when it was released in the European Union, but just for a short while. So I did, using the very same prompts, on the very same book: first, for the general abstract, which through a fast, diagonal reading did look neither better nor worse, just maybe a little shorter; then, by chapter, which I also got at high speed; now I just needed the enlarged abstract for the individual chapter, whose number or title I didn't remember, so I moved to the ChatGPT to check it: number seven.

But, know what? There was no chapter seven in the index of the book (the abstracts by chapter) provided by Bard: just six chapters, while ChatGPT had abstracted nine. Much worse, even: it wasn't simply an incomplete or partially wrong result, as far as the titles of the chapters had no resemblance at all. My first intuition was that one of these GAIs had missed the point, or the book, and ChatGPT was, of course, my usual suspect. Fortunately, I had the book and could verify, and guess what: neither nine nor six chapters and not a simple individual coincidence in their titles. Both GhatGPT and Bard had made up (hallucinated, if you like) the abstract of the book, as well as the number, the titles, and the abstracts of every chapter. 

At the same time, I saw that the book was first published in 2022, which I had not noticed before. So ChatGPT, trained on data until 2021, could not know about it, but the notable fact is that notwithstanding its usual and boring reminder about this limitation, once and again, when asked for anything that it can detect as possibly more recent, a simple book title with no date is no deterrent (presumably it just had more than enough feed on the trustfulness of Wikipedia, which had been an obsession for years and for so many). It is harder to explain the makeup by Bard, even more, if you note that the book is perfectly registered in Google Books as well as in Google Scholar and, of course, plenty of libraries, bookshops, and other catalogs.

24 ene 2023

ChatGPT: si el objeto es la lengua, la singularidad ya está aqui

 No he podido resistirlo. Me han invitado a un pequeño coloquio sobre las implicaciones de la Inteligencia Artificial (sin pensarlo ya lo he escrito como lo dejo, con mayúsculas iniciales, honor que la ortografía reservaba para entidades como el Estado, la Iglesia, etc. –todo un lapsus), y lo ha hecho un grupo de investigación dedicado a multilingüismo, multiliteracidad y educación lingüística, un evento que será en inglés y con mayoría de profesores de inglés. He aceptado, pues el tema me interesa, y mucho; también porque llevo un tiempo dándole vueltas y porque me queda un mes para pensar lo que diré.

Lo que no he podido resistir es la tentación de dirigir ya algunas preguntas relacionadas con ello a la propia IA, en concreto a ChatGPT, siempre tan dispuesta. ChatGPT dista (todavía) mucho, en no pocos aspectos (en otros la supera), de la inteligencia humana, pero lo que ya se le da muy bien, haciendo honor a su nombre (chat, charla/r) es expresarse en charlatán fluido. En toda enseñanza hay de esto mucho más de lo que debería, pero en el aprendizaje de una lengua resulta más tolerable, porque lo que se trata de aprender es la lengua misma, antes que lo que expresa. En el peor de los casos, la IA puede ofrecer ya excelentes niveles de morfología y sintaxis (similares a los de un nativo y probablemente superiores a los de quien enseña en una segunda lengua), niveles razonables de sentido, racionalidad y veracidad, y una interactividad verbal potencialmente inagotable.

Pero no digo más. Copio y pego, sin modificar un solo carácter, mi charla con ChatGPT. Se repite un poco (quizá no tiene en cuenta que está hablando con un mismo interlocutor), es muy prudente en las respuestas (por no decir timorato) y muestra, además de una gran corrección política, un exquisito cuidado en que sus declaraciones no molesten ni preocupen a nadie (en eso, parece el gabinete de comunicación de cualquier ministro). Creo que le he pillado una única, pequeña, falta de ortografía (¿o debería decir de sintaxis?), pero la dejo como acertijo.

Please tell me about yourself

I am ChatGPT, a large language model developed by OpenAI. I am trained on a vast amount of text data and can understand and respond to a wide range of natural language inputs. I can assist with tasks such as answering questions, generating text, and providing information on various topics.

Will you be useful for education?

I can be useful for education in a variety of ways. For example, I can assist students with their homework by answering questions, providing explanations, and helping with research. I can also be used to generate practice problems and quizzes, which can help students to better understand and retain the material. Additionally, I can be used to create interactive educational content, such as simulations and tutorials. Overall, I can help to make education more efficient and effective by providing students with quick and accurate information and resources.

Will you be particularly useful for teaching or learning ESL?

I can be particularly useful for teaching or learning English as a Second Language (ESL) as I can help students to improve their reading, writing, listening and speaking skills by providing them with a variety of resources and exercises.

One way is that I can assist in providing practice and feedback for grammar and vocabulary, which are essential building blocks of the language. Additionally, I can provide feedback on pronunciation and intonation, which are important aspects of speaking.

Furthermore, I can assist in providing reading comprehension exercises, and respond to students' questions about the text. I can also generate writing prompts and essays, which can help students to improve their writing skills.

In addition, I can be used to create interactive exercises and dialogues which can help students to improve their listening and speaking skills.

Overall, my ability to understand and respond to natural language inputs, as well as generate text, makes me well-suited to assist with ESL instruction and learning.

Will that be a menace for ESL teachers' employment?

The use of technology, including language models like myself, in education can bring many benefits, such as providing students with personalized and interactive learning experiences. However, it's important to note that my capabilities are a tool to assist, not replace, human teachers.

ESL teachers play an important role in providing guidance, feedback, and cultural context, which are essential for language learning. They also develop lesson plans, assess students' progress, and provide emotional support. I cannot replace the human interaction, empathy, and personalization that a teacher can provide.

In fact, the use of technology in education can also help ESL teachers to be more efficient and effective in their teaching. For example, teachers can use me to create and grade quizzes, provide feedback on writing assignments, and generate extra practice materials for their students.

In conclusion, I can help to enhance the teaching and learning of ESL, but I cannot replace the value that human teachers bring to the classroom.

Would an ESL learner be able to learn with just ChatGPT?

An ESL learner would likely find it challenging to learn a new language with just ChatGPT. While I can provide students with practice and feedback on grammar, vocabulary, reading, writing and listening, learning a new language is a complex process that involves more than just mastering these skills.

Effective language learning also requires interaction with native speakers, exposure to authentic materials, and cultural context. Moreover, learning a language requires practice and repetition, which is best achieved through immersion and interaction with native speakers.

Additionally, language learning also requires emotional support and motivation. A human teacher can provide guidance, feedback, and emotional support, which are essential for language learning. They can also help learners to set goals, track progress, and stay motivated.

Furthermore, a human teacher can assess students' progress and adapt the curriculum to their specific needs and learning styles.

In conclusion, while I can be a useful tool to assist with language learning, it would be challenging for an ESL learner to learn a new language exclusively with just ChatGPT. A human teacher is needed to provide the emotional support, cultural context, and personalization that are essential for effective language learning.

But would it be enough, let us say, for 'international English', i.e. to learn the kind of English language that is usually spoken in academic international meetings?

It is possible that an ESL learner could use ChatGPT to improve their knowledge of the kind of English that is spoken in academic international meetings, often referred to as "international English". However, it would likely not be enough on its own.

International English is a standardized form of the language that is often used in academic, scientific and professional contexts. It can be characterized by a formal register, specific vocabulary and grammar structures. While I can provide students with practice and feedback on grammar, vocabulary, and writing, learning International English also requires exposure to authentic materials, such as academic papers, professional reports, and scientific journals, as well as interaction with native speakers who use this variety of English.

Additionally, it requires practice and repetition, which is best achieved through immersion and interaction with native speakers, as well as feedback from a human teacher.

Furthermore, a human teacher can assess students' progress and adapt the curriculum to their specific needs and learning styles, and also provide guidance, feedback, and emotional support, which are essential for language learning.

In conclusion, while I can be a useful tool to assist with learning international English, it is not enough on its own. An ESL learner would benefit from a combination of using my resources and services, as well as having access to authentic materials, interaction with native speakers and guidance from a human teacher.

Can you correct my spelling, or mi gramar, if I ask you it to do, this question for instance?

Yes, I can correct your spelling and grammar if you ask me to. For example, in your current question, you wrote "mi gramar" instead of "my grammar" and "Can you correct my spelling, or mi gramar, if I ask you it to do, this question for instance?" instead of "Can you correct my spelling, or my grammar, if I ask you to do it, in this question for instance?".

Keep in mind that my ability to correct grammar and spelling is not always perfect and I may make mistakes as well. It's always a good idea to consult multiple sources for grammar and spelling checks, and to consult with a human teacher or tutor for additional feedback and support.

Teachers are worried about the possibility that students use ChatGPT, or any other AI form, for plagiarism as never before. Are we right?

It's true that technology has made it easier for students to access information, including text, and that this may increase the potential for plagiarism. ChatGPT, and other AI forms, can generate text that is similar to that found in existing sources, which could make it easier for students to plagiarize.

However, it's important to note that plagiarism is not a new problem and that it can be addressed through education and the use of plagiarism detection tools. Teachers can educate their students about the importance of academic integrity and the proper way to use and cite sources. They can also use plagiarism detection tools to check students' work for similarities to existing text.

Additionally, AI-generated text can also be used as a tool for plagiarism detection, as it can help to identify passages of text that are too similar to existing sources.

It's also important to note that, as a AI model, my main function is to assist users in generating text, and not to plagiarize. However, my users should always be aware of the importance of academic integrity and proper citation, and use my assistance accordingly.

In conclusion, while technology may make it easier for students to access information and potentially plagiarize, it can also be used as a tool to detect plagiarism and promote academic integrity.

Have you detected any plagiarism in my questions? Are they very common questions addressed to you?

As an AI language model, I do not have the capability to detect plagiarism. My main function is to understand and respond to natural language inputs based on the text data that I was trained on.

As for your questions, they are not very common and are specific. However, questions about the use of AI in education, the potential impact on employment and the potential for plagiarism are topics that have been raised before and are relevant.

It's important to remember that plagiarism is not only limited to text-based works, but also includes visual, audio and other types of works, so it's important to be aware of the academic integrity guidelines and proper citation methods in any type of work, whether it's a text, a video, a podcast or any other format.

If you have any doubts about academic integrity and plagiarism, it's always a good idea to consult with your teacher or a tutor for guidance and support.

When you read a question, let us say a long question, by someone who is using English as a second language, can you detect his or her first language?

That model is currently overloaded with other requests.


Esto último, después de una meditación breve pero menos de lo habitual. O sea, que se cansó de mí. Hasta otra.