GeeCON 2023: Tannaz N. Roshandel - Machine Intelligence

### Overview In this talk, the speaker discusses the pervasive role of artificial intelligence and machine intelligence across modern life, ranging from entertainment recommendations and chatbots to healthcare, transportation, and education. Despite the substantial benefits in automation and decision-making, widespread AI deployment introduces significant ethical risks and real-world harms. The presentation explores the nature of ethics, the risks inherent to algorithmic black boxes and data aggregation, and the social responsibility developers and individuals carry to build fair, transparent, and accountable technology.

### Key Topics - **Ethical Challenges and Real-World Harms**: Highlighting algorithmic failures, including self-driving car accidents, HR systems that screen out qualified women, misidentifications by facial/security systems, unfair mortgage denials for minorities, and AI avatars generating misleading diversity representations. - **The Nature of Ethics vs. Algorithms**: Defining ethics as a human process of discovering, questioning, and defending core values. While technology operates through mathematical models without innate morality, human creators must take responsibility for ethical outcomes. - **Data Tracking and Filter Bubbles**: How smartphones and wearable devices continuously collect telemetry—noting that individuals generate roughly two megabytes of data per second—which services combine to exploit vulnerabilities, predetermine choices, and isolate users within narrow filter bubbles that impair critical thinking. - **Individual and Corporate Responsibility**: Referencing Milton Friedman's 1970 observation that businesses cannot have responsibilities independent of the individuals operating them, emphasizing that engineers, users, and diverse organizations must actively shape ethical norms.

### Technical Details - **Black Box Processing**: Modern AI functions mimic neural structures via mathematical models trained on vast historic datasets. Because algorithms learn and iterate autonomously, the underlying operations frequently function as opaque black boxes, making results difficult to anticipate or explain without explicit explainability features. - **Explainable, Trusted, and Responsible AI**: The speaker underscores the necessity for frameworks such as Explainable AI (XAI) to inspect why systems make particular decisions (e.g., auditing HR models to identify which specific input features disqualified candidates). - **Data Aggregation Risks**: The core privacy concern is often not a single service capturing data in isolation, but the cross-service compilation and correlation of multiple datasets to model user behaviors.

### Takeaways - AI is an assistive tool to augment productivity and decision-making, not an autonomous replacement for human judgment, empathy, or moral reasoning. - Mitigating bias requires building diverse engineering and leadership teams that incorporate a broad spectrum of perspectives, backgrounds, and life experiences. - Individuals and technologists must continuously practice critical thinking—observing, analyzing, and evaluating consequences—while advocating for transparency, open communication, and fair AI practices.