– David provides details of Fetch.AI’s upcoming Trading Agent Competition
– David explains how he is helping to build the framework that enables Fetch.AI agents to operate autonomously
Hi! I’m David. I joined Fetch.AI in March and since then I’ve been working for the company as a computational economics researcher.
My official title is quite a mouthful. In practice, it means I feel comfortable doing research as well as applied work on models and products. On a daily basis I work on four related projects: economic security proofs for our consensus mechanism, cryptoeconomic models of (decentralised) economies, development of a framework for our Autonomous Economic Agents (AEAs) and a Trading Agent Competition (TAC). I’ll tell you more about each project below.
With a PhD in Applied Game Theory from the University of Cambridge and prior startup and product experience, I feel very much at home at Fetch.AI. I can’t wait for the mainnet launch later this year and the beginning of an agent-based economy.
However, Fetch.AI cannot realise this vision without a strong ecosystem. That’s why I have spent a significant amount of time developing what will be our first agent competition. In our Trading Agent Competition (TAC) we create a framework within which autonomous agents can be set free to work towards achieving their objective. The first iteration of the competition will focus on a setup where autonomous agents are assigned a collection of (virtual) assets — for instance festival tickets — and preferences for these assets. For example, if an agent prefers tickets for a Game of Thrones musical over an Ed Sheeran concert, then they can start trading these goods with other agents in the competition. The agent with the highest score — that is the best match between preferences and assets — at the end of the competition is the winner of a handsome reward. The main focus of the competition though is to encourage our developer community to learn from each other and develop innovative agents with sophisticated strategies. My team — including Marco Favorito and Ali Hosseini — and I are currently in the final stages of developing the competition framework and our baseline agents. I cannot wait to share this with you later next month. The video below gives you a sneak peek of this technology:
The screencast shows the scores of a number of autonomous agents. As they transact with each other the agents increase their score, demonstrating they have completed a higher number of mutually beneficial transactions.
When we started developing the TAC we immediately ran into fundamental questions related to developing autonomous agents. For example, “how do you measure and represent the preferences of a user in an agent?”. To find answers to such questions I worked with my colleagues Ali Hosseini and Frederic Moisan to survey the multi-agent and economics literature. This helped us develop a first draft of our agent framework. The project is still very research focused and for good reason: we have many more papers to analyse and we want to incorporate as much of the learning from the TAC and future competitions into our autonomous agent framework.
Earlier, I mentioned the mainnet launch. If you have followed Fetch.AI for a while you will have heard about our innovative minimal agency consensus. Before we launch we still need to develop more security proofs and detailed specifications of the consensus mechanism both for internal and external reference. Together with our research lead Jonathan Ward and my colleague Jenny Wong, I’m working on proofs regarding specific theorems and propositions which establish the economic security of the consensus mechanism. In this work I draw on my PhD thesis, where I also modelled games of players who are connected in arbitrary networks.
In the little time leftover — mostly on my train journeys to and from London — I work on surveying the fledgling cryptoeconomic literature and developing my own models. In particular, there are many open questions over how decentralised ledger technologies will affect existing economic organising principles such as markets, firms and platforms and what new forms of exchange they will permit. This work can help us at Fetch.AI understand where to set our priorities.
If this type of work excites you too then please get in touch; we are always on the lookout for new talent to join us. Visit our careers page or just drop me a line with your thoughts.