How Generative AI is Looking to Change Financial Services

In June 2023, Acuity’s CEO Andrew Lane hosted a special LinkedIn Live event, you can re-watch now:

How does Generative AI relate to NLP and NLG?

Generative AI is a broader field that encompasses Natural Language Processing (NLP) and Natural Language Generation (NLG) as specific areas of focus. Generative AI refers to the use of artificial intelligence techniques to generate new content, such as text, images, music, and more.

NLP specifically deals with the processing, understanding, and manipulation of human language by computers. NLP techniques enable computers to analyze and interpret text, speech, and other forms of natural language data. NLP encompasses tasks such as text classification, sentiment analysis, named entity recognition, machine translation, and more.

NLG, on the other hand, is a subfield of NLP that specifically focuses on the generation of human-like text. NLG techniques use AI algorithms to automatically generate text that is coherent, fluent, and contextually relevant. NLG finds applications in areas such as content creation, data reporting, personalized recommendations, and more.

Generative AI can utilize NLP techniques to understand and analyze existing text data, and then use NLG techniques to generate new text based on the learned patterns. For example, a generative AI system may use NLP to analyze a large dataset of customer reviews and then use NLG to generate summaries or responses based on the analysis. Similarly, a chatbot or virtual assistant may use NLP to understand user queries and NLG to generate responses in natural language.

In summary, generative AI is a broader field that includes NLP and NLG as specific areas of focus. NLP enables computers to process and understand human language, while NLG specifically focuses on generating human-like text. Both NLP and NLG are important components of generative AI, enabling systems to understand and generate text in a wide range of applications.

How does Generative AI relate to NLP and NLG?

What about the evolution of Natural Language Generation (NLG)?

• 1950s and 1960s: The early days of NLG were focused on developing systems that could generate simple text, such as telegrams and weather forecasts. These systems were often rule-based, meaning that they were programmed with a set of rules for how to generate text.
• 1970s and 1980s: In the 1970s and 1980s, NLG systems became more sophisticated. These systems began to use statistical methods to generate text, which allowed them to produce more natural-sounding output.
• 1990s and 2000s: In the 1990s and 2000s, NLG systems became even more advanced. These systems began to use machine learning to generate text, which allowed them to produce even more realistic and engaging output.
• 2010s and 2020s: In the 2010s and 2020s, NLG has continued to evolve. Recent advances in artificial intelligence have led to the development of NLG systems that can generate text that is indistinguishable from human-written text.

Today, NLG is used in a wide variety of applications, including:

• Chatbots: NLG is used to create chatbots that can interact with humans in a natural way.
• Virtual assistants: NLG is used to create virtual assistants that can help humans with tasks such as scheduling appointments, making reservations, and providing information.
• Machine translation: NLG is used to create machine translation systems that can translate text from one language to another.
• Text summarization: NLG is used to create text summarization systems that can create brief summaries of longer pieces of text.
• Question answering: NLG is used to create question answering systems that can answer questions posed in natural language.

What are some popular use cases of Generative AI?

Content Creation: Generative AI can be used to automatically generate content for journalism, advertising, and marketing

Image Generation: Generative AI can generate images from scratch or modify existing images. Think Dalle 2

Music Generation: Generative AI can create new music compositions or assist musicians in generating melodies, harmonies, and rhythms.

Video Game Design: Generative AI can generate game levels, characters, and game assets,.

Personalized Recommendations: Generative AI can analyze user data, preferences, and behavior to generate personalized recommendations for products, services, movies, music, and more. This can be used in recommendation systems for e-commerce, streaming platforms, and online advertising.

Virtual Assistants and Chatbots: Generative AI can power virtual assistants and chatbots to generate human-like responses to user queries, engage in conversations, and provide information or assistance in various domains, such as customer support, healthcare, and education.

Data Augmentation: Generative AI can generate synthetic data to augment limited or insufficient data for training machine learning models. This can help improve the performance and robustness of AI models, especially in scenarios where obtaining large amounts of real-world data is challenging.

Drug Discovery: Generative AI can assist in designing and generating novel drug molecules with desired properties, potentially speeding up the drug discovery process and aiding in the development of new medicines.

What is the Impact of Generative AI on Financial Services?

• Generative AI is already being used in financial services, such as fraud detection and credit scoring
• The benefits of using generative AI in financial services, are an increased efficiency and accuracy in decision-making
• Potential drawbacks and risks associated with using generative AI in financial services, are bias and privacy concerns
• Many manual jobs stand to be lost such as: Data entry, Manual Trading/Manual Execution, Risk assessment & monitoring, Market Research & analysis, financial reporting & compliance
• It’s important to note that while generative AI has the potential to automate certain tasks in finance, it also has the potential to create new job opportunities. As technology evolves, there may be a shift in the demand for different skill sets, with a greater emphasis on skills such as data science, machine learning, algorithm development, and strategic decision-making.

Generative AI and Fraud Detection

Anomaly Detection: Generative AI models can be trained on normal or legitimate transaction data to learn the patterns of normal behavior. Once trained, these models can identify anomalies or deviations from the learned patterns, which may indicate fraudulent activities.

For example, a generative AI model can learn the patterns of legitimate credit card transactions and flag transactions that deviate significantly from those patterns, such as unusually large transactions, transactions from unknown or suspicious locations, or transactions that occur at abnormal times.

Fraud Alert Generation: Generative AI can be used to generate fraud alerts or notifications that provide real-time feedback to users or fraud detection teams. These alerts can be generated based on the analysis of multiple data sources, such as transaction data, user behavior data, device information, and more, to identify suspicious activities and trigger alerts for further investigation.

Adaptive Fraud Detection: Generative AI can continuously learn and adapt to changing fraud patterns and techniques by analyzing new data and updating the fraud detection models in real-time. This allows for proactive detection and mitigation of emerging fraud patterns, making fraud detection systems more resilient and effective.

When we look at Fraud detection in Trading we may imply Trading Surveillance, where data led models are already employed by nearly all major banks and exchanges

Generative AI and Credit Scoring

Data Augmentation: Generative AI can generate synthetic data to augment limited or incomplete credit data for training credit scoring models. This can help banks overcome the challenges of limited data availability, especially for borrowers with thin credit files or in emerging markets, and improve the accuracy and robustness of credit scoring models.

Feature Engineering: Generative AI can automatically generate relevant features or variables that can be used in credit scoring models. For example, generative AI can generate synthetic features that capture borrower behavior, transaction patterns, or other relevant information that can enhance the predictive power of credit scoring models.

Fraud Detection: Generative AI can be used to detect fraud or misrepresentation in credit applications. By generating synthetic data that represents potential fraud patterns, generative AI can help banks identify suspicious applications that may have been crafted to deceive credit scoring models and mitigate the risk of granting credit to fraudulent borrowers.

Credit Risk Assessment: Generative AI can be used to simulate different scenarios and assess credit risk. For example, generative AI can generate synthetic scenarios that represent economic downturns, changes in interest rates, or other macroeconomic factors to assess the resilience of credit portfolios to different risk scenarios and make informed credit decisions.

Personalized Credit Offers: Generative AI can be used to generate personalized credit offers for individual borrowers. By analyzing borrower data, financial history, and credit preferences, generative AI can generate tailored credit offers that align with borrowers’ needs, risk profiles, and creditworthiness, improving customer experience and increasing credit approval rates.

Explainable Credit Decisions: Generative AI can generate explanations or justifications for credit decisions, making credit scoring models more transparent and interpretable. This can help banks comply with regulatory requirements and provide borrowers with understandable explanations for credit approval or denial, increasing transparency and trust in the credit scoring process.

Credit Portfolio Optimization: Generative AI can be used to optimize credit portfolio management by generating synthetic scenarios and simulating portfolio performance under different conditions. This can help banks optimize credit risk, diversify their portfolios, and make data-driven decisions for credit risk mitigation and portfolio management strategies.

Generative AI can enhance the accuracy, transparency, and efficiency of credit scoring processes for banks, enabling them to make more informed credit decisions, mitigate credit risk, and provide personalized credit offers to borrowers. However, it’s important to note that the use of generative AI in credit scoring should comply with regulatory requirements, ensure data privacy and security, and be thoroughly validated to ensure the reliability and fairness of credit decisions.

Generative AI and Trading and  the era of the automated Quant

Market Data Generation: Generative AI can generate synthetic market data, such as synthetic price and volume data, order book data, or trade data.

Trading Signal Generation: Generative AI can analyze vast amounts of market data, news, social media, and other relevant information to generate trading signals.

Algorithmic Trading: Generative AI can be used to develop and optimize algorithmic trading strategies. By analyzing historical market data, generative AI models can learn patterns and relationships in the data to create trading algorithms that automatically execute trades based on predefined rules or conditions.

Risk Management: Generative AI can be used to assess and mitigate trading risks. For example, generative AI models can simulate various risk scenarios, such as market crashes, extreme price movements, or liquidity shocks.

Portfolio Optimization: Generative AI can be used to optimize trading portfolios by generating synthetic scenarios and simulating portfolio performance under different market conditions.

High-Frequency Trading: Generative AI can be used in high-frequency trading (HFT) to analyze vast amounts of market data in real-time and generate trading signals or execute trades with low latency

Sentiment Analysis: Generative AI can be used to analyze market sentiment from various sources, such as news, social media, or financial reports, to generate trading signals based on the sentiment analysis.

The risks are also there such as over fitting, models need to be retrained on new market dynamics, potential for herd mentality and super crashes. In the end all automated models need human oversight.

Bloomberg Generative AI

Bloomberg has created a new finance chatbot called BloombergGPT, which utilizes its vast proprietary and curated financial datasets to provide unprecedented financial research and analysis capabilities.

It’s the next step in Corporate AI, where you take generic Generative AI and layer in your own proprietary data. Think all of the data Pharmaceuticals, Governments, Armed forces etc

The training of the BloombergGPT model required approximately 53 days of computations run on 64 servers with NVIDIA GPUs, and the cost estimation to produce the model alone is estimated to be over $2.7 million.

The source datasets used for training include financial news, company financial filings, press releases, Bloomberg News content, as well as general and common datasets like The Pile, The Colossal Clean Crawled Corpus, and Wikipedia.

BloombergGPT has led to create some unique outputs, such as generating initial drafts of Securities and Exchange Commission filings, summarizing financial content into headlines, and reducing the cost of filing.

Ethical issues

In conclusion, while Generative AI has the potential to bring significant benefits to the financial services industry, it also raises ethical concerns related to bias, transparency, data privacy, fraud, job displacement, regulatory compliance, and ethical decision-making. It is crucial to adopt responsible and ethical practices in the development, deployment, and use of Generative AI in finance to ensure that it is used in a fair, transparent, and accountable manner, and to mitigate potential risks and challenges.