What do ChatGPT, JBIG2, a German computer scientist, #Xerox, and a construction project all have in common?

When Copying Goes Wrong: The Xerox Photocopier Incident and What it Teaches Us About Large-Language Models

The New Yorker connected Xerox to ChatGPT: 

"In 2013, workers at a German construction company noticed something odd about their Xerox photocopier: when they made a copy of the floor plan of a house, the copy differed from the original in a subtle but significant way. In the original floor plan, each of the house’s three rooms was accompanied by a rectangle specifying its area: the rooms were 14.13, 21.11, and 17.42 square metres, respectively. However, in the photocopy, all three rooms were labelled as being 14.13 square metres in size. 

The company contacted the computer scientist David Kriesel to investigate this seemingly inconceivable result."

This inspired me to call on my new assistant, GPT, and put this together.

__________

Have you heard about the incident at a German construction company involving a Xerox photocopier? If not, let me break it down for you.

A computer scientist was called in to investigate and found that the photocopier was using bad compression, specifically JBIG2, in its digital scanning and printing process. This type of compression saves space by identifying similar regions in an image and storing only one copy for all of them.

Unfortunately, in this case, the photocopier mistook the labels for the room areas as similar enough to store only one, causing the copy to have incorrect numbers.

So, what's the big deal? Why should you care about this photocopier mishap? Well, let me tell you.

The Xerox photocopier incident highlights:

• The importance of accuracy in technology
• The dangers of lossy compression, particularly when it comes to numbers and measurements
• The need to be mindful of the limitations and potential biases of technology and validate its outputs for accuracy

__________

Now, let's talk about how this incident is relevant to large language models like OpenAI's ChatGPT.

Just like the Xerox photocopier, large language models like ChatGPT can also produce results that are different from the original input and result in biases and inaccuracies. It's crucial to understand the limitations and biases of ChatGPT and validate its outputs before making decisions or taking actions based on them.

Think about it like this: if you were building a house and relied on inaccurate measurements, the whole structure could be thrown off. Similarly, if you make a decision based on incorrect information generated by ChatGPT, the consequences could be just as significant.

So, what can you do to avoid these types of situations? It's simple. Just keep these three things in mind:

• Understand the limitations and biases of technology
• Validate the outputs of technology for accuracy
• Be mindful of the potential dangers of lossy compression, especially when it comes to numbers and measurements

The Xerox incident serves as a reminder of the importance of accuracy in technology and the need to be mindful of the limitations and potential biases of technology and validate its outputs for accuracy. This holds true for large language models like ChatGPT as well as for other technologies like JBIG2. 

Just like a chef can only cook with the ingredients they have available and a sculptor can only work with the material they have, ChatGPT can only generate responses based on the data it was trained on. 

__________

This incident was caught by computer scientist David Kriesel, back in 2013.