Why My Employer Banned ChatGPT?
What are they afraid of?
Unless you’ve been living under a rock, you’ve probably heard of ChatGPT. By this point you’re probably a bit sick and tired of hearing how revolutionary and game-changing ChatGPT is. Well, I’m going to look at this a little differently.
A message from legal…
Working for a tech company headquartered in California but being based in the UK. I often find myself checking emails late into the night. On the particular day I see an email with the title: “URGENT: Report to HR immediately”.
You can imagine my concern but upon opening the email this appeared to be a frantically written email from the legal department. But what did they want…
Generative AI models are being sued…
The legal team has caught wind of the latest events in the AI space. Getty Images are allegedly planning to take OpenAI’s Dalle to court, for copyright infringement. Why should this bother me, I don’t use Dalle and for that matter I don’t even use ChatGPT.
Well the concern is in the way these models are trained. There is no way to complete vet the training data until this is publicly made available. So there is the possibility the these generative models are learning from and copy copyrighted material, without seeking consent. This is where Getty Images suit derives from, they desire to protect their intellectual property and so you can see why my employer is worried.
ChatGPT is commonly being used by workers in tech to automate cumbersome tasks like writing short scripts. However, what if ChatGPT is learning from proprioty information?
Fair-use VS Copyright-infringement
Common knowledge dictates that providing work is transformative in nature, you should be able to draw inspiration from other art forms. This might work well for producing images using generative models but generating code is not as well received. You would often be expected to either cite or request permission for the use of another person code.
So this is what scares my companies legal team, what if someone decides to use ChatGPT to produce a model which is heavily derived from someone else’s work? Do we pay royalties? Do we need permission? Have we violated copyright protections?
This is not the only reason ChatGPT is being banned…
Generative models are flawed…
In their defence, OpenAI clearly state that ChatGPT is not 100% accurate and never claimed it to be. Now consider completely automating your tasks, without placing any check and balances into your workflow. Simply ask ChatGPT to write a model for you, copy-paste and you’re done.
You can see how it’s easy to adopt some of the errors, and considering the learning set is heavily weighted by internet forums and wikipedia, there is bound to be significant errors.
The potential to adopt these errors is worrying, could we unknowingly implement a change that has catastrophic behaviour? What would that do to the firm?
The bottom line is the company functioned efficiently and effectively before ChatGPT, so the merits of adopting experimental generative models currently do not outweigh the potential costs. It is rightfully so that ChatGPT is banned FOR NOW, but once the legal and accuracy issues can be navigated the future will most definitely be different.
Shrouded Science is an individual effort, if you enjoyed this consider following me. Consider supporting me by subscribing to my YouTube Channel: Shrouded Science.
