Artificial intelligence (AI) has gained tremendous traction over the last couple of months. Since the end of 2022, AI has become a household topic due to the mainstream adoption of OpenAI’s chatbot “ChatGPT” and its immediate, worldwide impact across industries and people's lives.
In 2022, consultants at McKinsey found that AI adoption had stagnated over the past few years. However, with the arrival of ChatGPT, adoption has increased significantly. According to OpenAI’s founder, Sam Altman, ChatGPT crossed over 100 million users in just two months, a milestone it took Facebook 4.5 years, Instagram 2.5 years and Twitter five years to achieve.
This article is part of CoinDesk’s “BUIDL Week.” Marcello Mari is chief executive officer of SingularityDAO and Rafe Tariq is a senior quant researcher at SingularityDAO Labs.
As we start 2023, we see that Microsoft and Google are engaged in a fierce battle for AI dominance. They are competing with rival chatbots, search optimization and more – and it appears Microsoft is leading the way. The software giant gave OpenAI $1 billion in the initial stages of ChatGPT's development, taking a 46% stake in the company, and plans to integrate ChatGPT into its web browser Edge and search engine Bing, both of which are likely to revolutionise search and internet browsing.
When you think about it, AI may finally allow Microsoft to outcompete Google in a space the latter has dominated for years. OpenAI predicts that ChatGPT will generate revenue of $200 million by the end of 2023 and $1 billion by the end of 2024. It’s quite possible that by 2030 AI will become the number one industry in terms of revenue generation and market cap.
See also: Crypto AI Needs a Showcase to Know What's Real | Opinion
As we move towards a future where AI is everywhere, inevitably replacing many human jobs, it is interesting to consider how this powerful form of computing can be used to maximize opportunities in the crypto industry. AI can be applied to make crypto more efficient, and blockchain technologies can also be used to solve problems unique to machine learning.
Traditional AI methods applied to crypto
Sentiment analysis and cognitive distortion detection in social media
Sentiment Analysis is a technique in which natural language processing algorithms (NLP) are able to analyze text and attribute meaning to it, helping humans to understand whether there is a positive or a negative sentiment regarding a particular asset class.
In traditional finance, sentiment analysis was typically performed over news media. However, in the crypto market, by the time an update reaches the news, it's usually already too late to make money from trading. This may explain the adage "buy the rumor, sell the news," meaning a new market trend must be spotted on social media as it happens or even before it happens.
As we know, crypto markets without volatility wouldn't be as attractive. The unpredictable movements in the crypto market play a crucial role in its dynamics. Therefore, there is a need for further development of AI and data frameworks to facilitate price prediction studies and applications.
These frameworks should be capable of collecting sentiment data from various channels, whether they are crypto-related or not, and should have an AI analytical framework that can integrate the latest developments in sentiment analysis research. It should also be able to distinguish a real person from a bot as well as real conversations from orchestrated ones.
These frameworks will be able to detect so-called cognitive distortions on social media, such as catastrophizing (exagerating the importance of a negative event: “because of this everything will dump”), fortune-telling (pretending to know about the future: “this will definitely happen”) and mind reading (pretending to know what other thinks: “everyone knows that.”)
Predicting market movements
AI has been used for decades in traditional finance to detect market dynamics before they occur. Traditionally, this has been achieved through sentiment analysis. However, in the field of cryptocurrency, we can rely on statistical correlation between major coins or categories of coins. For instance, in localized ecosystems like the decentralized exchange Curve or AI-focused SingularityNET, which have multiple tokens, we see lagging and correlative trading patterns emerge.
Due to rapid technological advancements in hardware used to secure and mine decentralized networks (i.e., the rise of GPU-based computation), the use of large-scale deep learning models has become increasingly valuable for understanding price fluctuations. Expanding machine learning and deep learning methods used in traditional finance to predict price fluctuation or identifying market regimes (i.e. whether we in a bear or bull market) is one of the key areas of exploration for AI use cases in crypto.
A further area of research regards the application of reinforcement learning, an AI technique that learns without supervision from humans (aka unsupervised learning) to better understand the impact of its actions. This has applications for predicting slippage and price impact when assets are traded.
Trading bots/AI-based market making
The AI team at SingularityDAO has conducted exploratory studies in the field of market simulation and backtesting to improve the state of the art in quantifying market dynamics. One promising technology we have explored is the "adaptive multi-strategy agent" (AMSA) for market making. This basically provides an environment where different AI algorithms can buy and sell assets and backtest those trades, while evaluating the performance and effect trading has on the market.
These self-reinforcing trading algorithms can be seen as the next step evolution of traditional trading bots already widely adopted by traders and market makers on centralized exchanges. In other words, AI is being developed to help create more sophisticated automated market maker systems. This contributes to the adoption of more robust decentralized trading systems, and can help traders to rebalance their multi-asset portfolios.
Crypto native AI problems
Effective monitoring of dynamic position and entity risk
Due to the increasing frequency in crypto markets of black swans (unpredictable events with potentially severe consequences), traditional methods to evaluate risk in trading positions have become outdated. In crypto, analysts need to evaluate risk associated with liquidity movements across protocols and this is virtually impossible to do manually given the large amount of data to be analyzed.
An AI approach, once again, can extend human decision making.AI algorithms can be used alongside other methods commonly used to monitor the health of on-chain positions across all protocols, like analysis of large wallet holders and liquidation risk. By gaining expertise and experience in both AI and decentalized finance (DeFi), it is possible to create new metrics that can provide easy-to-read signals about risk exposure taken across different protocols.
Further, AI offers a substantial amount of value and support to human analysts as the crypto industry become increasingly multiprotocol (with development across blockchains happening even in the bear market), leading to a significant increase in complexity. Predictive and correlational risk methodologies are essential to prevent future black swan events, such as those that occurred with crypto exchange FTX and lending platform Celsius Network.
An emphasis on flow analytics, correlation and predictive analysis
Following the fallout of Celsius and FTX, there was an increased need to develop methodologies for monitoring events and factors that could lead to similar cases. Crypto analysts and data scientists explored a range of approaches, from classical alerting signals based on wallets and entities to more advanced AI-based capital flow aggregations.
Twitter vigilantes are already using AI-based analytics platforms to uncover news stories before they break to mainstream crypto news. However, a lot can be done to simplify and expand these tools in order to be adopted by the wider market.
AI techniques for malicious entity labelling and detection on-chain
In the crypto market, there is a constant game of identifying malicious entities on-chain, which requires the use of extremely large datasets. AI plays a crucial role in this transparency effort, using state-of-the-art clustering, genetic programming and neural networks to pinpoint these malicious entities to their aliases on-chain.
See also: Why Crypto Trading is Essential for the Cryptocurrency Industry | Opinion
As malicious users become more sophisticated in hiding their obligation to an entity, we rely on advanced AI algorithms along with geographical and behavioral data to identify these wallets.
Far away and here today
Although AGI (artificial general intelligence) or an AI that is sentient is still far away, progress in the field in the last few years has been remarkable. I strongly believe that in the future, artificial intelligence will manage our crypto funds and ensure the safety and health of our wallets.
The integration with large language models like ChatGPT has significantly expedited this process and will make it easy and accessible to anyone. Crypto has the potential to create a new inclusive financial ecosystem, and we have a once-in-a-lifetime opportunity to lead the way in this and compete with Big Tech companies.