How intelligent is AI? We are in the midst of a third major revolution. The information revolution has the potential to change society as profoundly as did the agricultural revolution and the industrial revolution in the past. A notable difference, however, is that while the impact of those revolutions took centuries to be felt, the information revolution is transforming society within a few decades. We have moved from the first PCs to a world where everyone is connected all of the time in thirty years. Today, one of the more pronounced impacts refers to the machines’ intelligence.
We live in an era of rapid sophistication of autonomous agents — they are precisely the carriers of what we call Artificial Intelligence (AI). The scientific foundations have been laid some time ago, but only recently has the technology begun assuming a critical role in the everyday life of ordinary citizens. Famous examples start with IBM’s chess player Deep Blue and Google’s AlphaGo, which were very successful in the confined environment of a game, and span a wide spectrum of breakthroughs in medical diagnosis and legal judgments, up to self-parking and fully self-driving cars. If you are still hesitant of a robot offering a diagnosis or taking a legal decision, think of the amount and variety of experience a smart mechanical agent may possess, as well as the speed in which it assimilates new information. When it comes to more mundane matters, as AI and Machine Learning gain traction in our daily lives, smart agents may run essentially all microtransactions and everyday tasks.
Ideally, autonomous agents shall be very efficient in learning and refining their understanding of the world based on their own experiences, or from the feedback provided by humans. A related aspect nowadays is, of course, the availability of massive data which, by means of Machine Learning, aspires to improve the quality and robustness of autonomous decision-making. The aforementioned examples were cited in order of increasing difficulty, going beyond structured problems and predetermined environments to heterogeneous and unpredictable settings, where making decisions and judgments involves the complex nature of human agents. In the latter realm, a concrete question, which still represents a challenge even for advanced AI systems, is to enable the smooth functioning of decentralized communities. Let us focus on the case of Digital Asset Exchanges.
Decentralized Exchanges disrupt digital transactions.
After the first decade of the digital asset marketplace, a number of issues hinder wide adoption of cryptocurrencies: malicious hacks, technical traps, unreliable governance, stolen stakes, or even flat-out fraud. In particular, looking at exchange platforms, the digital equivalent of a stock market, some major centralized exchanges have pulled back dramatically, typically because of software attacks that made users lose money. Therefore, it seems inevitable for seasoned investors as well as everyday citizens to turn to Decentralized Exchanges (DEXs) that leverage the power of blockchain technology: The exchange platform itself becomes transparent, and the funds are held by each user in their own personal digital wallet. As in so many domains of today’s activity, one chooses to rely on peer-to-peer transactions than put trust on some unknown, and often dubious, centralized authority. In a nutshell, a DEX is a distributed exchange that emphasizes security, privacy, liquidity and speed in a permissionless, decentralized framework. Besides having no single point of failure, DEXs have a reduced risk of server downtime.
DEXs rely on development and governance implemented by a Decentralized Autonomous Organization (DAO), which is managed by the community according to a constitutional code of conduct. This structure is spreading decision making power evenly among the DAO’s members, thus ensuring that the interests of all voters are adequately represented. As diversity increases and the system matures, it becomes increasingly autonomous, healthy, and self-sustainable, hence the increasing relevance of an autonomous agent that learns to safeguard the constitutional code by implementing the principles of Machine Learning and exhibiting the virtues of AI, in particular lack of bias. A particular example is when the DAO discusses projects that ask for funding, and the autonomous agent must first confirm that they are congruent with the core values and goals, in other words the constitution of the DAO. Hence AI is the key that shall lead a DAO to self governance.
However, a major challenge here is to seamlessly handle the conflicting goals of the human agents.
The dawn of Collective Intelligence
The power of AI in structured problems and predetermined environments is well-established; some major showcases were already discussed above. But assisting a DAO is a significantly harder task because, like driving cars or making medical decisions and legal judgments, it involves balancing the human factor, learning while the behaviour of actors may be selfish or conflicting and, finally, managing human agents with personal, and often selfish agendas. This is todays’ major challenge for AI in supporting the autonomous operation of decentralized communities. Game theory in the past couple of decades has focused on the computational hardness of synchronizing such groups of agents, in particular when they wish to engage into transactions. It has demonstrated the so-called Price of Anarchy, in other words the overhead of the overall system incurred by lack of coordination among its members.
Enter Collective Intelligence, where groups of individuals act collectively in an intelligent manner. This of course includes software, people, or agents that switch between the two states over time. It is precisely the case of DAOs, but also of many other instances of human-machine interaction. Eventually, the goal is to seamlessly integrate machine and human intelligence at a large scale in order to yield the so-called mixed-initiative systems, one of the holy grails of AI.
In this respect, the current efforts of the community target technology that is both intelligent and sensitive in such environments, where autonomous and human agents coexist. The practical aim is to maximize the Machine Learning benefit emerging from interaction with human insights, failures, and struggles. This project lies at the heart of our work in Volentix (https://volentix.io/), where we are supporting an integrated ecosystem to provide the infrastructure in the whole lifecycle of a digital asset, centered around a DEX and managed by a DAO, assisted by AI in the framework of Collective Intelligence. In particular, Volentix develops the decentralized exchange platform VDex, which employs a collection of smart EOS.IO contracts to establish quick and secure transactions, liquidity, and user anonymity, as well as its own digital KYC tool called Blocktopus. All of these aspects rely on AI in order to gain self-governance since they are eventually managed by a DAO, hence the need of combining smart agents with collective intelligence.
The future is bright but we do have some work to do.
Professor at the University of Athens