The third week of Theoriq’s Educational AMA saw our community discussing unique and innovative concepts in the Theoriq protocol. Senior AI Solutions Engineer Pourya Vakilipourtakalou and Alexander Marx joined head of AI Education Shingai Manjengwa. As a special guest this week, Theoriq’s Head of AI Research, Ethan Jackson, explained some of the meatier architecture topics in a simple way.

After introductions, the AMA dove straight into how Theoriq sets itself apart with its multi-faceted approach to evaluating AI Agents, how they are optimized using BOTS, and how they ensure fairness and transparency in the selection. Below you will find a summary of what was discussed. Enjoy!

The Challenge of Choice in an AI-Powered World

As AI continues to grow and become more ingrained into our everyday lives, we will start to see a variety of AI Agents become available for everyday use. Each agent will be designed to perform specific tasks, like booking you a flight or finding you the best performing index fund. Soon, there will be dozens of these choices, each claiming to offer the best service.

The challenge arises when the user is not just worried about functionality but also about trusting the reliability of different Agents. The key to solving this lies in creating robust mechanisms that allow users to evaluate and select the best agents from an ever-expanding pool of choices.

A Multifaceted Approach to Evaluating AI Agents

The Theoriq protocol has developed and integrated a comprehensive toolkit and strategy to address this challenge. The team's approach begins with categorizing Agents based on three primary criteria.

Preference-based Evaluation: This refers to the popularity of an AI Agent, if the agent is popular and frequently used, it is likely a strong performer. This popularity serves as an implicit quality signal, suggesting that the agent is and will be reliable and effective.

Staking Mechanisms: To enhance the evaluation process, Theoriq has introduced a staking mechanism, where users and community members can “stake” on agents they believe will perform well. This encourages active participation and rewards those who correctly identify high-quality agents. Staking adds an economic layer to the evaluation process, allowing users to earn rewards when their staked agents are successfully utilized.

Human Feedback: providing a human evaluation element plays a critical role. Users can provide direct feedback on their interactions with agents through simple thumbs-up ratings or detailed reviews. This feedback is crucial as it offers real-world evidence of an agent’s performance, helping to build a comprehensive profile of each agent. The data from these reviews is further analyzed using AI models, which help to reveal an agent’s strengths and weaknesses.

Optimizing Agent Collectives with Theoriq’s BOTS Algorithm

While choosing individual agents is important, many tasks require multiple agents working efficiently together. This is where Theoriq’s BOTS Algorithm comes into action. The algorithm—an acronym for Bayesian Optimization by Tournament of Substitutions—optimizes the composition of agent collectives. Instead of just selecting the best-performing agents individually, the BOTS algorithm considers how well agents work together as a team.

For example, planning a vacation might require an agent to book flights, another to manage your calendar, and another to find local restaurants. The BOTS Algorithm ensures that the chosen agents don’t just excel individually but also complement each other to achieve your overall goal. It builds a predictive model based on past Agent evaluations and user stakes. This approach allows BOTS to estimate the performance of new combinations without exhaustive testing, significantly speeding up the process of finding high-performing Collective combinations and doing so efficiently.

Ensuring Fairness and Meritocracy in AI

A significant concern with any system that relies on popularity or staking is the potential bias towards well-established agents. Theoriq solves this by ensuring that even new agents have the opportunity to prove their worth. Developers can submit their agents to standardized evaluations that compare them directly with more established agents. If a new agent performs well, it is automatically given priority in relevant tasks, creating an environment that awards meritocracy.

The Power and Potential of Agentic Systems

These innovations are part of a broader shift towards agentic systems—networks of AI agents working collaboratively to perform complex tasks. These systems represent a significant leap forward in AI development, driving personalized optimization that has the ability to revolutionize industries ranging from finance to healthcare.

Looking ahead, platforms like Theoriq’s Infinity Studio, where users can design and deploy their own collectives of AI agents, will empower individuals and businesses to create tailored solutions for specific uses. As we stand on the brink of this new era, the possibilities for AI are truly limitless, driven by the convergence of advanced evaluation techniques, optimization algorithms, and agentic systems. The future of AI lies in applying collaboration, transparency, and accountability, ensuring that the best agents and collectives rise to the top in a fair and efficient manner.

Make sure to tune into our next Weekly AMA Educational series on Discord and X.

If you want to dive deeper into all of the above, head over to our litepaper and read how we are building the future of AI Agent Collectives.