Currently, automatic learning models are widely used in various professional fields and form the basis of many mobile applications, software packages and online services. Although many people meet and interact with these models, few fully understand their operation and their underlying processes.
In the modern world of automatic learning, models are becoming more and more complex and rich in functionalities. Their growth raises an important question: how can we make these models more understandable and interpretable for a large audience, including specialists without in-depth knowledge in the field of automatic learning?
Researchers from the University of California, Irvine and Harvard University Developed the talktomodel. It is an interactive conversational system designed to explain automatic learning models and their predictions to professionals and non -expert users. This interface allows you to dialogue with ML models using ordinary natural language.
Research is based on previous developments linked to explainable artificial intelligence (XAI) and human-Ai interaction. The main objective of this work was to introduce a new platform which could provide clear and accessible explanations on the functioning of artificial intelligence, similar to the way in which the conversational platform of Openai, Chatgpt, answers questions.
The researchers conducted an experience involving health workers with different levels of automatic learning experience. Almost all participants were new on the ground. They were invited to use TalkTomodel to answer questions and understand how automatic learning models work.
The study results were impressive. Most users preferred to use TalkTomodel to understand the models. They did tasks faster and more precisely using this interface. Even automatic learning engineers have admitted that Talktomodel is a useful tool.
So how does TalkTomodel work? It transforms questions into structured logical forms which allow ML models to provide explanations and interpretations. This approach offers flexibility in dialogue, supporting an open survey and facilitating understanding of complex models.
TalkTomodel is an innovative system that opens the door to natural conversations aimed at understanding the automatic learning models applied to a variety of data and tabular classifiers. Instead of a complex programming, users communicate with TalkTomodel in natural language (fig. 1, block 1). The dialog engine analyzes the input data into the executable representation (fig. 1, block 2). The execution engine performs the operations and the dialog engine uses the results of its response (fig. 1, block 3).
Figure 1.
With TalkTomodel, users can explain why certain predictions occur in a model; The changes that occur in predictions when the input data change; The means to change the predictions, and more. This analysis can be applied to any data group, whether it is an individual body or a complete category of data.
For example, if you want to predict the development of diseases, you could ask questions like: “What is the importance of the body mass index (BMI) for predictions?” Or “how will the probability of illness change after lowering the glucose levels of 10 in men over 20 years old?”. TalkTomodel will give you information, saying that BMI is the most important predictive characteristic, and that the decrease in your glucose levels by 10 will reduce your chances of developing diabetes by 20%. After that, you can continue the dialogue by asking additional questions. TalkTomodel facilitates the explanation of the operation of the models, because you can speak to the system in natural language and this will give you informative responses.
You can see an example of such a dialogue in Figure 2.
Figure 2. Example of a dialogue on the prediction of diabetes, demonstrating the extent of the various subjects of discussion with the system
To support significant conversations with TalkTomodel, there are methods to improve language understanding and the explanation model. First, a dialogue engine is implemented which analyzes the input of the user's text. These data are converted into a language similar to the structured query language using a large language model (LLM). LLM performs the analysis by treating the task of translating user statements into a programming language as a seq2seq learning problem, with user statements as a source and analysis of the programming language as a target.
In addition, the TalkTomodel system combines the operations of the explanation, the analysis of automatic learning errors, data handling and the generation of descriptive text in a single language which can cover the wide range of potential conversation subjects which are necessary in the most explainable models. Examples of various operations are presented in Figure 3.
Figure 3. Operations are included in the conversation to generate answers.
The system offers an operating mechanism that automatically selects the most suitable explanations and operations for the user. This reduces user burden and makes interaction with automatic learning models more accessible. In addition, a textual interface has been created which allows those without high technical skills to understand and interact with ML models. Consequently, Talktomodel explains how the automatic learning models work more accessible and understandable to a wider audience.
In the future, the use of talktomodel can extend to include the use of the system in real clinical and laboratory contexts, where participants can apply it to understand and optimize model performance. In addition, future research could focus on visualization and analysis of raw data to increase user confidence.
Talktomodel is a step forward in the development of the field of explainable artificial intelligence. This interface allows you to talk to complex automatic learning models in natural language and to understand their decisions. This tool promises to make the ML more accessible and interpretable for everyone.
You can find the model code on github.
