1. ChatGPT is here to stay.  

2. ChatGPT will change the way people write and research software related development, architecture and implementation. 

3. ChatGPT will infringe on other companies’ and individuals’ IP. 

4. Companies will overreact to the infringement (real and potential) and life will become harder for the industry as a whole! 

I posit that all of the above is true.  

As company leaders, engineers and consumers we all have to be aware of the ramifications of using AI in the workplace. I’ll direct you to some recent headlines that touch upon this theme, each with their own distinct angle.  

• JPMorgan restricts employee use of ChatGPT – CNN, February 22, 2023 

• Companies are struggling to keep corporate secrets out of ChatGPT – Axios, March 10, 2023 

• Some companies are already replacing workers with ChatGPT, despite warnings it shouldn’t be relied on for ‘anything important’ – Fortune, February 25, 2023 

• Samsung accidentally leaked its secrets to ChatGPT — three times! – Toms Guide, April 5, 2023 

• ChatGPT Could Face Defamation Lawsuits for Making Up Facts About People – PC Magazine, April 6, 2023 

We have a couple of themes here that will surely underlie many of the future issues that come up as AI adoption increases. 

What Goes In Must Come Out 

The first theme, and likely the most important driver in this conversation, is the leaking of a company’s intellectual property into ChatGPT. The old adage of “What Goes In Must Come Out” applies here.  

Not everybody is aware of how these language models operate. To put it simply, these chatbots are trained on data and that data must come from somewhere. According to Tech Radar ChatGPT 3.5 was trained on 570GB of text data from the internet. That is a ton of data! 

However it doesn’t stop there. Anything a person enters into ChatGPT is likely to end up as a data point in some form or another. In fact the mindset should be that anything that is entered in will eventually come out. It is not a one way street. 

If It’s On The Internet, It Must Be True 

Sure it is. But here’s the rub. AI language models like ChatGPT have a couple of learning phases. The first one is the “generative pre-training” phase. What this means depends on which model is being used. I believe ChatGPT is currently at the GPT3.5 model. In this context the training is done in an unsupervised manner. Thus, data quality is bound to vary. Which is why OpenAI CEO Sam Altman said that it is a mistake to be relying on it for anything important right now.  

The data these chatbots are training themselves on comes in part from the internet because they are neural networks tied to the internet. That much isn’t rocket science. Thus, if we can all agree that the internet is chocked full of information that isn’t true then we can all agree that some of the data that becomes part of the AI’s knowledge base must also be viewed in that same vein.  

As ChatGPT continually adds to its knowledge base through “self learning” we must expect that it acquires knowledge that A) it doesn’t own, nor should have acquired and B.) it’s not all factual. 

I actually asked ChatGPT how it increases its knowledge garnered after the generative pre-training phase. The following image is a snapshot of the answer. 

Feedback from ChatGPT on an a Chatbots Leaning Techniques

The takeaway here is that there is a certain aspect of data acquisition that is culled by “self-leaning” techniques. This makes perfect sense. However, as the name implies, unsupervised is the key here. AI models such as ChatGPT don’t have the ability to independently discern fact from fiction. They aren’t sentient.  

Tie “What Goes In Must Come Out” with “If It’s On The Internet, It Must Be True” and you get “ChatGPT Could Face Defamation Lawsuits for Making Up Facts About People“. Making up facts is a misnomer here, it’s not making them up, at least not maliciously, or minimally not on its own. The data behind ChatGPT is most likely inherently flawed because the sourcing has not necessarily been vetted in the capacity to tag it as false. Anybody that works in data science knows the problems that arise with poor data quality. ChatGPT isn’t alone here.  

We are in uncharted legal territory here. It will need to shake out before companies jump in head first and I would not want to be the canary in the coal mine on that front.  

It’s My IP 

In doing research I had ChatGPT generate some code to see how well it could do the task. It turns out that it did quite well. When asking ChatGPT where the code came from it said that the code was open source licensed under Apache.  

This my friends is an issue. I would wager top dollar that a ton of code generated from ChatGPT is already finding its way into many companies’ applications. It is only a matter of time before the lawsuits start flying. Somebody needs to be the guard at the gate. 

Worse, the ramifications due to what I see as the potential over reaction to the above is that it won’t stop at ChatGPT. Information repositories such as Google, Stack Overflow, Medium and eventually chatbots like ChatGPT are the go to for software engineers. This doesn’t mean that engineers are copying all of the source code they find on the internet and putting it into the applications they write. But it sure as hell implies that people learn from others and that is the basis for much of the software written today. It will only become worse when companies start using AI’s to generate code.  

What Happens Next? 

All of the above begs the question, “what will happen to productivity when companies overreact and prevent their engineers from using the internet to learn how others came up with solutions for commonly occurring problems?”  

There are a ton of people that mentor others through blogs and tech articles, all of which end up on the internet. Much intentionally and free to use. How will this impact a company’s future initiatives If they have to crack down on what is considered acceptable use today? 

As company leaders we need to be prepared to take this on and answer these questions. The plan needs to be in place yesterday. Everybody needs to train their people on these issues as it is yet another in the continuing effort to protect intellectual property and prevent other’s IP from becoming part of someone else’s code base. Those that are ahead of the game will be well prepared to step into the vacuum that will be created by others that don’t. 

Written By Terry Trippany

The post AI, Chatbots and the Ramifications of Using them in the Workplace  appeared first on Creospan.

The post What’s Holding You Back from Unlocking AI-Powered Workforce Productivity? appeared first on Creospan.