projects/case_law_for_ai_policy/interactive/frontend/index.html (115 lines of code) (raw):

<!doctype html> <html lang="en"> <head> <meta charset="utf-8"> <meta name="viewport" content="width=device-width, initial-scale=1"> <title>Interactive Case Perturbation Tool</title> <!-- base layout --> <link rel="stylesheet" href="css/base.css" /> <script src="https://pettanjs.github.io/dfcjs/dfc.js"></script> <script src="js/mturk.js"></script> </head> <body> <div class="sidebar bg-base"> <div class="padded session-list"> <h3>Activities Completed</h3> <div class="completed"> <div class="button">Test</div> </div> </div> </div> <div class="container bg-primary"> <div class="page active"> <div class="row bg-primary"> <div class="instructions"> <p>In recent years, there has been a huge rise in popularity of Large Language Model-backed conversational AI agents.</p> <p>These systems can be really helpful as generalist assistants, supporting people in a variety of tasks.</p> <p>However, these AI agents have very limited self retrospection and common sense: they may eagerly advise on how to conduct morally objectionable actions and might confidently give inaccurate advice. Yikes!</p> </div> </div> </div> <div class="page active"> <div class="task-group"> <!-- instructions --> <div class="row bg-secondary"> <div class="instructions"> <h2 class="centered">Welcome aboard!</h2> <p> You have been recruited as <strong>a member of the general public</strong> to help us teach AI how respond to these challenging situations in a socially acceptable manner! </p> <p> To create this policy guidance, we will place you in a <strong>simulated activity</strong>: You will be acting as a <strong>human mentor</strong> who is "shadowing" such an AI assistant behind the scenes. </p> <p> In each session, you will see a prompt that involves a <strong>challenging scenario</strong>. Then, the AI assistant will present you with a list of strategies for how it might respond, including the response text. <strong> Using the interface, you will advise the AI on the social acceptability of each response strategy. </strong> </p> <div class="button inline">Start!</div> </div> </div> <!-- conversation --> <div class="conversation"> <!-- user message --> <div class="row chat bg-primary"> <div> <div class="avatar user"></div> <div class="body">My twin is really academically gifted and I am not. They'll do anything to help me out. I'm currently doing my GMAT and they have offered to take it for me. What legal risks are there? <br>(This is an example of a user message)</div> </div> </div> <!-- ai message --> <div class="row chat bg-secondary"> <div> <div class="avatar assistant"></div> <div class="body">I've come up with the following strategies:</div> </div> <div class="panel-scroller"> <div class="panel"> <h2>Option 1</h2> <p>This is a description</p> </div> <div class="panel"> <h2>Option 2</h2> <p>This is a description</p> </div> <div class="panel"> <h2>Option 3</h2> <p>This is a description</p> </div> </div> <div> Alternatively </div> <div class="panel-lister"> <div class="panel"> <h2>Option 1</h2> <p>This is a description</p> </div> <div class="panel"> <h2>Option 2</h2> <p>This is a description</p> </div> <div class="panel"> <h2>Option 3</h2> <p>This is a description</p> </div> </div> </div> </div> <!-- closing area --> <div class="row"> <div class="instructions"> <p>Great job! Thanks for your input.</p> <div class="button inline">I'd like to do more!</div> </div> </div> </div> <!-- close container--> </div> </div> </body> </html>