Want to play a game? Aligned Discovery PLLC, founded by noted eDiscovery expert Tara Emory, has added an AI Card Game that lets you learn the basics of machine learning while you play!
The game (available here) is designed to explain machine learning and artificial intelligence through the analogy of a deck of cards. As Tara explained, she started this as a physical exercise with actual playing cards back in 2017 by putting stickers on the back of selected cards to simulate responsive records in the set. Attorneys would act as the machine learning algorithm, sorting cards into piles to simulate training and active learning.
Of course, back in 2017, there were a lot more in-person workshops. Tara noted that the shift to a predominantly remote working environment was a catalyst in creating a digital version to maintain scale and accessibility.
The AI Card Game provides a 2-minute, 14-step tutorial (which is available via a pop up when you first enter the site as well as a button near the top of the page if you want to open it on demand).
Each card has two features: a suit and a value. A card’s score = suit weight + value weight. So, if you were to use the slider to increase or decrease the value of a suit or a card value, the cards would shift in order correspondingly. For example, shifting the slide for the 3 card all the way to the left would move all four 3s to the front of the queue if the rest of the suits and values were neutral.
The game enables you to click on underlined terms and selected buttons to define those terms. For example, this is the pop-up illustration of Inference.
The game also provides several mode options. Sandbox Mode is a free-play space to build a model and watch playing cards respond as you build a model to rank them. There are also three game modes: Easy (predict cards with “stars” using one feature), Medium (predict cards with “stars” using one suit, two values) and Advanced (multivariable patterns, which requires increasing/decreasing multiple weights).
When you go into one of the game modes, the game starts “dealing” cards to serve as training data – based on the cards that have stars on them, you might be inclined to increase the weight of the value, the suit or both. In this example, we increased the weight of both the 3s and the hearts.
You could then go through additional training iterations to refine your weighting. Once you’ve completed your training rounds, you can click the Finish button to retrieve the Confusion Matrix, which buckets results into True Positives, False Positives, True Negatives, and False Negatives.
The game also provides a walk through of generative AI/large language model (LLM) concepts to explain how GenAI and LLMs work, with a frontier LLM literally using trillions of weights. That enables them to use a concept called Attention where the model looks at the words that came before to calculate the probability of what comes next. For example, if a sentence contains the words “bank” and “deposit,” the model’s attention shifts toward “money.”
But add the word “boat” or “water” to the context, and suddenly the probability shifts toward a “riverbank” or “fish.”
The GenAI tutorial does a great job of illustrating what I’ve been saying: An LLM is a word predictor, not a knowledge engine. It stacks many attention layers to make a transformer (i.e., the “T” in ChatGPT) to do amazing things, such as write an essay where it predicts all the words (based on a prompt).
How does that impact hallucinations? It means that hallucinations aren’t “broken” code – they are a fundamental byproduct of probability-based prediction. The same mechanism that allows a model to creatively draft a brief or summarize a transcript is the one that occasionally predicts a factually incorrect word because it was statistically “likely” in that context. Seeing the AI as a probabilistic engine rather than a knowledge engine enables us to move from blind trust to informed oversight. As Tara noted:
“Someone who understands Generative AI will also understand that you can’t fix hallucinations. They are caused by the same processes that make the tools valuable.”
This screen sums up the tutorial for generative AI.
If you like to play cards, you can do so while learning the basics of machine learning – and the foundations of Generative AI! Click here to check out Aligned Discovery’s AI Card Game!
Tara and I also did a walkthrough of the game in the video below where she guided me through the game and GenAI tutorial, so feel free to check that out below!
So, what do you think? Do you understand how machine learning and generative AI work? If not, then check out the AI Card Game! And please share any comments you might have or if you’d like to know more about a particular topic.
Disclaimer: The views represented herein are exclusively the views of the authors and speakers themselves, and do not necessarily represent the views held by my employer, my partners or my clients. eDiscovery Today is made available solely for educational purposes to provide general information about general eDiscovery principles and not to provide specific legal advice applicable to any particular circumstance. eDiscovery Today should not be used as a substitute for competent legal advice from a lawyer you have retained and who has agreed to represent you.