Generative AI in Educational Publishing

A Transformative Journey

Integrate Generative AI?

Generative AI is a specific category within AI that has become democratized, integrating algorithms and API layers to make it not just useful, but most importantly user-friendly. Its integration unquivocably propels us into the cloud and into an era of true digital transformation.?

Generative AI enables the creation of new products and services, streamlines editorial processes, and guides publishers towards a data-centric culture, where market trends are decisive and timely.Harvard Graduate School of Education Professor Houman Harouni emphasizes that schools cannot and should not ignore the use of generative AI in the classroom. Many argue that the advent of generative AI will rewrite the fundamental rules of modern education.

In this context, it is crucial to have a well-defined strategy for integrative this transformative technology to unlock its full potential in the Edtech world. According to Gartner, this data-centric transformation culture will lead to better, less intuitive decision-making, thereby boosting productivity, and, consequently, competitiveness.

2. The Framework

The foundation of any operational AI integration lies in a semantic framework involving various actions:

1. Creation of content pieces (chunks) where linguistic units are vital for the learning process.
2. Automated metadata incorporation.
3. Vectorization (embedding), transforming text into numerical data used by machine learning algorithms (ML).
4. AI model selection, usually a combination of several models (GPT, Llama, Falcon, etc.) followed by data adjustment and fine-tuning for specific tasks.

3. What are the pillars of Generative AI for publishers?

As of today, the key aspects of generative AI for educational publishers can be summarized as:

1. Text-to-Data Conversion. AI's ability to transform text into structured data, making it easier to process and analyze, enriched with the defined semantic layer.
2. Text Generation. AI's capacity to create content automatically, including translation and more.
3. Conversational Capabilities. Enabling understanding and assistance through natural language conversations.
4. Feedback. Enhancing response quality through continuous feedback loops.
5. Code. Automatically generating tasks related to its use.
6. Image Processing. Generating descriptions, enabling automatic labeling, and creating visual content.

4. Challenges in Implementation

Several editorial challenges need addressing:

• Optimizing Editorial Processes. Current editorial processes are slow and lack feedback mechanisms concerning product consumption. In regards to the learner’s experience, there is no content generation, knowledge management or learning which is entirely counterintuitive. It’s essential to clarify that job loss is not a risk, we are simply reducing tasks and processes that can be automated, with the same quality. The aim here is to free up time to devote to other projects, more initiatives and higher returns on investment that increase productivity and profitability.?
• Hybrid Editing. It’s not a question of forgetting about paper printing. The focus is on establishing hybrid approaches where generative AI plays a transformative role in your entire production flow. This includes emerging trends, such as personalized print-on-demand, which is gaining momentum in the industry, thanks to generative IA. It requires proper intellectual property (IP) management and optimal teacher training aligned with quality standards. Encouraging the initiation of Proof of Concept (POC) as a focus for experiential feedback.?
• Generating new services and products. According to Ted Mo Chen- Vice President, Globalization at ClassIn, generative AI is already saving teachers time in lesson planning, instructional guidance and personalized feedback. Already many Edtechs are reimagining the entire learning flow through AI.?

5. Requirements for Successful Implementation

To navigate this transformative landscape, educational publishers need:

• Clearly defined applications and objectives.
• To work on content labeling and the semantic layer vectorization (embedding).
• Use the most appropriate LLM functional model (either GTP, Llama, Falcon, etc) to pre-train (fine-tunning) the model above all to avoid unexpected costs and ensure alignment with the business.?
• Give the right context that the publisher wants, creating the "promts" that align with the objectives. In other words, offer clear instructions to obtain specific answer.
• Interoperability and integration. Essential components to incorporate in your workflows.?
• Compliance with security and privacy standards.

In conclusion, as the Packard Law predicts, success or survival in this transformative landscape depends on internal knowledge and talent, a resource currently scarce in many companies, but imperative for short and medium-term success.

If you’d like to discuss the topic of generative AI in educational publishing and you’re attending the Frankfurt Book Fair, schedule a meeting with me here.