top of page

We Go Beyond

the State of the
Art in Assessing

AI Trustworthiness



Fairness in AI includes concerns for equality and equity by addressing issues such as bias and discrimination. THEMIS acknowledges differing perceptions of fairness among users. Using innovative AI training methods to combat disparities in AI systems will promote transparency and trust. Co-creation and innovation with users is at the forefront of what we do. 

Decision Intelligence


Decision Intelligence (DI) is an engineering discipline that improve the decision-making in a socio-technical system and provides a means to overcome the mismatch between the sophistication of socio-technical systems’ decision-making practices and the complexity of situation in which those decisions must be made. An invaluable part of understanding that socio-technical context in which decisions take place is to assess how these decisions impact the core metrics affecting the quality of the socio-technical system through the discipline of DI.


In this way, humans should be able to reach a holistic view on the quality of the decision support and be able to reflect on it and improve fairness and technical accuracy of the providing AI systems.


Reinforcement Learning (RL) is the type of learning guided by a specific objective: an agent learns by interacting with an unknown environment, typically in a trial-and-error way; the agent receives feedback in terms of a reward (or punishment) from the environment, then, it uses this feedback to train itself and collect experience and knowledge about the environment. RL methods have been very successful on a variety of complex tasks, often far exceeding human-level performance. 


Although a large portion of these successes can be attributed to recent developments in deep reinforcement learning, many of the core ideas employed in these algorithms derive inspiration from findings in animal learning, psychology, and neuroscience.


Reinforcement Learning

Flexible tool that AI users can use to enhance trust in their AI systems through risk management processes that promote fairness, transparency, and accountability. 

Framework for self-managing the trustworthiness of AI

IPT has developed the conceptual model ME.So.T.IS (Moral Enhancement via Socratic AI for Training and Intelligent Self-coaching), it provides a theoretical framework to empower moral agent decision-making through Socratic AI in the context of applied AI ethics.

ME.So.T.IS will be extended in THEMIS to enhance human agency in moral values assessment. 

Conceptual model for moral evaluation

The agent will be developed as a Rasa chatbot. Rasa is an AI open-source framework which allows developers to builder conversational agents.


In THEMIS, the agent will be extended with an interactive, multi-criteria Reinforcement Learning algorithm with which the agent will learn from human evaluative feedback i.e. evaluations of the quality of the agent’s behaviour provided by a human user, or advice/instruction.

Human AI conversational agent 

Trustilo has developed an extended attackers' profile model using five categories of human traits with specific attributes: personality, social-behavioural, technical capabilities, motivation and trigger. Trustilo has also developed a scoring system to measure the strength of the profile.


THEMIS will enhance this profiling model and scoring system to address the specificities of AI trustworthiness attacks

Human profiling model and scoring system


bottom of page