Artificial Intelligence’s Promise and Peril

Source details

Generative AI ("GenAI") is described as the most impressive advance in machine-learning technologies yet, marking a significant leap in AI’s ability to understand and interact with complex data patterns.
GenAI is poised to "unleash a new wave of creativity and productivity" while raising "important questions for humanity."

1950: Alan Turing first envisioned machines reaching levels of mastery in a 1950 paper.
1960s: ELIZA generated human-like responses as a precursor to "chatbots."
"Two decades later": artificial neural networks appeared, advancing language understanding and image recognition but constrained by limited training data and computing power.
2000s: Deep learning emerged as the third wave of AI.
2014: Generative adversarial networks (GANs) introduced two competing neural networks—"generator" and "discriminator"—to sharpen imitation of data, text, or images.
2017: The paper "Attention Is All You Need" introduced attention mechanisms that enabled machines to focus on relevant input parts.
November 2022: ChatGPT was launched by OpenAI, followed by other GenAI chatbots from big-tech firms.

Deep learning enabled practical applications such as Google Translate, digital assistants (Alexa and Siri), and self-driving cars.
GANs use a generator to create imitation content and a discriminator to distinguish real from simulated content.
Attention mechanisms allow models to pay attention to relevant input segments, improving content generation quality.

Traditional AI (advanced analytics, machine learning, predictive deep learning) has long been used for number-crunching, market trend analysis, and product customization.
GenAI differentiator: ability to "delve deeper and interpret complex data in a more creative manner," producing forecasts, alternate scenarios, charts, and code.
Government use cases: improve citizen services and address workforce shortages.
Central banks: enhanced capacity to sift through vast banking data to refine economic forecasts and better monitor risks, including fraud.
Investment firms: detect subtle shifts in stock prices and market sentiment to propose more creative investment options.
Insurance companies: explore generative models to create personalized policies aligned with individual needs and preferences.

Hallucination: GenAI can produce nonsensical and untrue facts; characterized as a "stochastic parrot."
Knowledge cutoff: ChatGPT’s knowledge is limited to its latest training date.
Bias and transparency: amplification of existing biases in training data and lack of decision transparency remain urgent concerns.
Weaponization and misinformation:
- GenAI can craft stories that reinforce echo chambers and ideological silos.
- March 2022: An AI-generated video purportedly showed Ukrainian President Volodymyr Zelenskyy surrendering to Russian forces, demonstrating potential for political and market manipulation.
- Fabricated stories, doctored images, and synthetic videos can spread misinformation, incite panic, and destabilize economic or financial systems.
Job displacement: advancing GenAI may automate tasks previously performed by humans, "leading to many job losses" and necessitating employment and retraining strategies.
Existential risk warnings: leading AI experts, including ChatGPT’s creator, cosigned a letter asserting that "mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."

Vigilant oversight is imperative as GenAI cannot be uninvented.
Development of new regulatory frameworks is necessary to govern GenAI deployment and risks.
Emphasize ethical, transparent, controllable innovations that "harmonize with human values."
Prioritize strategies for employment and retraining to address potential job displacement.

Source: "Artificial Intelligence’s Promise and Peril," HERVE TOURPE, F&D Magazine, December 2023.

F&D Back to Basics: Artificial Intelligence’s Promise and Peril
- F&D Back to Basics: Artificial Intelligence’s Promise and Peril (Markdown version)
- F&D Back to Basics: Artificial Intelligence’s Promise and Peril (PDF)