haunted-house-2024.qmd

---
title: "(Unspooky) AI Approaches to Power Your Research"
author: "Manu Alcalá Kovalski and Judah Axelrod"
date: "2024-10-31"
format: 
  revealjs:
    menu: false 
    embed-resources: true
    theme: urbn.scss    
    logo: images/urban-institute-logo-2.png
title-slide-attributes: 
    data-background-image: images/urban-institute-logo-1.png
    data-background-size: 25%
    data-background-position: 3% 85%
---

## Our (Supernatural) Vision

Leverage emerging AI-powered text analysis methods to:

- **<span style="color:#1696d2;">Enhance current research</span>**
- **<span style="color:#1696d2;">Unlock new research questions</span>**

Emerging NLP and AI tools  can **<span style="color:#1696d2;">power up existing research</span>** and **<span style="color:#1696d2;">unlock new research questions</span>**

## The Problems (Horrors) We Can Solve

- **<span style="color:#1696d2;">Overwhelming data:</span>** From websites, lengthy transcripts, case notes, etc.
- **<span style="color:#1696d2;">Time-consuming processing</span>**: Manual collection and sifting through information is often unfeasible
- Goal: **<span style="color:#1696d2;">Synthesize vast data into actionable insights</span>** without compromising time or feasibility.


## Tricks and Treats For This Bottleneck
::: {.panel-tabset}
### Basic Text Analysis
- Think of as "advanced CTRL+F"
- Flag key search terms and the windows of text around them
- Makes manual review more targeted and efficient.

### Advanced Text Analysis
- Semantic ("smart") search
- Name-entity tagging to pull out important entities (people, places, organizations)
- Topic modeling to find latent themes in text
:::

## Tricks and Treats For This Bottleneck
::: {.panel-tabset}

### Extractive AI
- Give Large Language Model (LLM) a prompt to find and pull out specific text or answers from body of documents
- Can be great for parsing through unstructured data, tables or cases where you may not know all the search terms

### Generative AI
- Ask AI to generate insights, summaries, or recommendations based on the text
- This is a place where we still have much to learn ourselves!
:::

## Your (Eerie) Enquiries
- How much do I need to know about AI?
- What kinds of data work well these tools? How about confidential data?
- I hear AI can hallucinate, so how can we trust and verify it?
- What if I am concerned about these tools replacing my expertise?

## What are *Our* Eeriest Enquiries?
- Which specific platforms, models, and cost structures should Urban use in AI for research?
- What use cases call for—or exclude—AI-based solutions?
- What are our blind spots?

## Demo Time!
<div>
```{=html}
<iframe data-external="1" width="100%" height="500" src="topic-browser.html" title="Webpage example"></iframe>
```
</div>