The AI Trainer course by Conversation Design Institute was the course I had the biggest expectations of.
I have not had much experience in AI Training yet. Although the Drift Enterprise plan enables AI-powered Chatbot usage, their solution is slightly different than the ones I learned about in books or other courses. I tried my abilities with the Voiceflow tool during the Third Conversation Design Festival challenge, but I needed more time to learn sufficiently.
So what’s CDI’s AI Trainer Course?
The AI Trainer course consists of 3 modules that broadly cover all substantial information needed to train Conversational Assistants and AI models. The course is taught by CDI’s co-founder and CEO, Hans van Dam.
The Foundation → “Theory”
The first module starts by contrasting the role of the AI Trainer from 2 other roles in an ideal team working on a conversational interface. Hans van Dam describes the main daily tasks and challenges AI Trainers can expect. He explains the types of Conversational Interfaces and how the split of the responsibilities and the role's importance changes depending on a bot’s abilities.
Knowing that an AI Trainer needs to turn data into understanding, it is crucial to learn how to do it. Before, you must learn the theoretical aspects of AI and Conversation Design and several essential definitions. You will familiarize yourself with the Conversation Design workflow and the role of AI Trainer in each step.
A video about a value-irritation model was valuable for me from a business perspective. The presented approach will allow me to be more customer-centric while considering the company’s business goal.
The Core → “Build”
The second module focuses on more practical aspects of an AI Trainer, and helps to understand several nuances in the field. For example, you will understand why “all training phrases are utterances but not all utterances are training phrases.” Thus you will learn how to source, create and label the phrases. The CEO of CDI tells what it means to clean the phrases and how you can create more variants and improve them.
After the above work is done, the AI trainer needs to know how to measure a success or a failure of the efforts. It must be known how to realize if a given intent is overtrained or undertrained. After you get this knowledge you’ll learn about thresholds, confidence levels, and other important metrics.
I really appreciated the list of 7 items that should belong to the Master Flow from which the design starts. And then the information on how to link dialogue steps and conditions to it.
I also must mention very valid advice about testing any work done before deploying. As an experienced IT specialist, I know that a change in 1 place can have a butterfly effect and impact the system elsewhere. Hans van Dam provides five necessary testing dialogue steps in the second part of the training to avoid unexpected consequences.
The cherry on top → “Improve”
The last module was the most interesting for me. As mentioned, I already had a bit of experience with AI Training courses, but I still felt unfamiliar with ways of improving the model that may be used in various conversational platforms.
One of the important aspects mentioned by the trainer was recognizing business metrics and technical metrics. There are various aspects that may be measured:
- Traffic,
- Activation,
- Containment,
- Customer satisfaction or coverage.
I really loved the table picture that explained in a straightforward way how to understand the definitions of: true positive, false positive, true negative, and false negative. Course viewers are provided four formulas to use these data in calculations of different measures informing us about 4 points of view on the cognition of the virtual assistant.
With the calculated measures, an AI Trainer can improve the bot conversations. It should not be done blindly because there should always be a focus on one metric at a time. Moreover, these metrics should be chosen according to the proper priority.
I really appreciate the list of 6 priorities provided by Hans in the courses. I still feel I need more experience, and this was valuable guidance. (It can also be a good guide for experienced AI Trainers who used the wrong priorities before).
Next, the CEO of CDI provided the four ordered steps to improve the cognition of a virtual assistant. Regardless of all improvements, it is always crucial to maintain the training phrase discipline.
And again, thanks to the course, I could learn seven best practices to achieve it. If you choose this course you’ll get advice on how to fix over and under-trained intents, how to do blind and k-fold tests and how to analyze the confusion matrix.
What I learned from the AI Trainer course
Thanks to the AI Trainer course, I feel prepared for many challenges that I may encounter in future. I am not afraid to solve different kinds of technical problems or improve the metrics and deal with any platform. Knowing the basic rules, priorities and standards, I feel I can manage!
Dear CDI Team, thank you for that.
Who will benefit most from this course?
It should be a rhetorical question for those who want to crack into Conversation Design and feel that an AI Trainer is something they want to be. I want to inform you that not every company has enough vacancies and funds to build an ideal team proposed by the Conversation Design Institute. Sometimes one person needs to cover a few roles.
I saw programmers that needed to become copywriters. I can also imagine a copywriter that was to invent several phrases and suddenly was asked to fix the model trained by his phrases. Even if you decide to go deeper in another role it is also worth knowing the basics of all.
Mutual understanding and communication are better if team members know each other's challenges. Knowing AI Trainer duties, you will be prepared better in case of illness of the team member, role change, job change or any unexpected circumstances.
