# How to Lie with AI: Understanding Bias, Ethics, and the Hidden Risks in ML - Clarissa Rodrigues

## Executive summary

The presentation explores how Machine Learning models can exhibit bias and 'lie' unintentionally due to flawed or non-representative training data. While AI is rapidly integrating into daily life (e.g., pricing, search, criminal justice), the speaker emphasizes that developers must maintain vigilance, prioritize explainable models, and ensure that model complexity aligns with problem complexity to mitigate ethical risks and unintended bias.

## Key takeaways

- ML Models are not inherently predictable like humans.: Unlike traditional algorithms where input/output is predictable, ML models can produce varied outputs based on internal weights and parameters. This lack of inherent transparency requires developers to be critically aware of model decisions.
- Bias originates from data, not the algorithm itself (Garbage In, Garbage Out).: To build a robust model, it is crucial that the training data is representative and free from historical or aggregation biases. Simply having more data does not guarantee accuracy; representativeness is key.
- The importance of Explainable AI (XAI).: Developers must strive for transparency by using explainable models to understand what the system is doing behind the scenes, rather than relying solely on complex black-box architectures.

## Technical details

- ML Model Architectures: Various models exist for different objectives: Regression, Clustering, and Classification. Advanced deep learning structures include Support Vector Machines (SVM), Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), and general Deep Neural Networks.
- Bias Mitigation: Bias can manifest in various ways, including gender bias (e.g., translation defaulting to male pronouns) or historical bias (e.g., COMPAS system). Techniques exist to reduce these biases during model optimization.
- Generative AI and Hallucination: Modern Generative AI models are trained by humans, which allows them to sound highly natural. However, they can 'hallucinate' or provide incorrect data (as demonstrated with cloud services), requiring careful validation of training sources.

## Practical implications

- Implement rigorous data auditing processes to identify and correct historical or demographic biases in training datasets.
- Prioritize the use of explainable model components over overly complex 'black box' models when the problem complexity does not warrant it.
- Treat ML outputs as hypotheses, requiring human oversight (Human-in-the-Loop) rather than accepting them as definitive truth, especially in high-stakes applications like finance or justice.
- Be critical of model claims and understand that AI is a tool reflecting human decisions, not possessing inherent intelligence or conscience.

## Topics

AI Ethics, Machine Learning Bias, Data Governance, Deep Learning Architectures, Explainable AI (XAI), COMPAS, Uber AI Solutions

Source: https://www.youtube.com/watch?v=ooaSTz6zL4w