Globant Sessions | IA: Finding the right question | English

Summary of Technological Concepts & Product/Usage Analysis

• AI is already embedded in daily decision-making: Recommendation systems (e.g., streaming/search), GPS routing, and e-commerce notifications personalize outcomes using data people provide.

• When AI should be used: The speaker frames strong AI use cases as those where:

  1. Decision complexity is high, and
  2. Decision frequency is high.

Examples include modern search ranking (more complex than ~15 years ago), real-time navigation, and rapid order updates.

A Three-Part Framework for AI Complexity

1. Technical aspect Model design choices such as neural network architectures, mathematical modeling, and defining data collection.

2. Business aspect Improving decisions people make by identifying/creating applications that optimize decision-making.

3. “Philosophical” aspect Whether to implement AI at all—questions like which data to collect, whether to process it, and whether to take a different approach.

Core Principle: Define the Right Problem

• If you adequately define the problem, you’re “halfway” to the solution.
• Trying to perfect an answer to a poorly formulated problem creates more downstream issues.

Recommendation/Engagement Systems Can Narrow Outcomes

• Social media models optimized for engagement metrics (likes/comments/time) can cause content variety to shrink, reinforcing user preferences.
• The speaker notes it’s difficult to balance keeping users interested with exposing them to diverse or new content.

Data Limitations with Human Decision-Making

• Training data often reflects what people already viewed or liked, so it doesn’t fully capture openness to new ideas.
• It’s hard to obtain data that both challenges beliefs and maintains engagement, because people may reject content that feels opposing (example discussed around health/vaccine misinformation).

Key Challenges for Applying AI (As Described)

• Capturing business opportunities: Knowing when to reach out and build AI applications that are technically viable while also meeting the right complexity/frequency and problem framing.

• Learning to adapt (“learn how to learn”): Memorizing facts is less valuable because technology and social context change quickly.

• Asking the right questions: Even if AI produces a strong answer, it won’t be accurate if the question/problem is wrong.

• Contextual data & changing conditions: AI needs stable references, but real-world conditions change (example: transport algorithms shifting abruptly when mobility stopped). Generalization becomes possible only when things stabilize.

Leadership Implications

• In domains where AI automates tasks, leaders must align the team around defining the right problems.
• The leadership challenge is about creativity and decision-making for how and when to use AI—not replacing human thinking.

Bias Example in Recruiting AI

• A corporation used AI to filter candidates by training on profiles of current leaders, described as biased toward white American men.
• Because the training/comparison basis was already biased, the model excluded women and other groups.
• The speaker also mentioned using AI more effectively to find candidates outside existing user channels—people who match what the company needs even if they don’t resemble historical patterns.

Closing Takeaway

• AI can strengthen data-driven decision-making in an increasingly digital world.
• The guiding question isn’t “can we solve this with AI?” but:

  “Does it make sense to solve this problem with AI?”

Main Speakers / Sources

• No specific named person appears in the provided subtitles.
• The talk is presented as a Globant “Sessions” session, with the speaker addressing the audience.