Select here to download the Arabic translation

Generative AI - Concerns and Ethics. And I am Farish Kashefinejad, the software development domain lead at Cloud Academy. If you have any questions or concerns about the content in this video, then feel free to reach out to 'support@cloudacademy.com'. In this video, we'll cover three topics: data privacy concerns, potential bias IP and copyright. So, some items to know. The content in this video may just come across as common sense. But unfortunately, sometimes we don't think about the common sense ideas that we need to take up, when dealing with something new, such as Generative AI. Second, if your company doesn't have a Generative AI policy, it might be time to consider one. Things are rapidly moving forward in such a way that Generative AI might become a daily tool for your whole company to use. Lastly, there may need to be a discussion with your company's legal team because things are moving at such a rapid pace.

Eventually, the law will catch up in every region. Every country has different policies and different viewpoints on Generative AI. Let's begin with data privacy concerns. So, when working with a service like ChatGPT, there's a couple of concerns here with regards to data. There are two forms of data. Your company's data, which may include intellectual property, and your user's data, which may contain sensitive user information. Back in March 2023, ChatGPT had a breach, and this expose user information, and their chat histories. And there's a couple implications here that has to be considered. Let's start with the chat histories. If you're working on a project for your company and you're feeding your company's data or intellectual property into a service like ChatGPT and there's a breach. What are the implications of that breach, if it is your company's data that was exposed? And this kind of exposure is happening with large companies too. So, one of the issues at hand is, when dealing with your company's proprietary IP, there may be needed additional security checks. Going back to one of my original statements that some of this is just common sense, but sometimes we do not think about these issues until they're put in front of us. So, what are the measures we can do to protect our data?

First, before using a service, read their privacy statement, find out what they do with the data and how long they store the data. Revisiting your privacy policy allows you to compare your privacy policy with the potential service you may be using to make sure they are in alignment. It also allows for an opportunity to update your privacy policy in the face of rapidly technological changes that are occurring in the world today, such as Generative AI. Third, and I mentioned this a little earlier, implement additional security checks to make sure you are protecting your data.

Lastly, when dealing with user data, you should anonymize that data, especially before feeding it into a third-party service. This allows your data to be worry free about the implications of the security breach. Now let's talk about Generative AI bias. The Diffusion Bias Explorer uses Generative AI image technologies such as stable diffusion to see if there are any biases. The way this explorer works is that I will choose an adjective and a profession. And through those two choices, those two prompts, it will generate a series of images. Now on the page, it does state you do not have to use an adjective, but from personal experience, sometimes when an adjective is not given, the explorer will not generate any images. To start this process, I'm going to choose confident Doctor and it takes a couple seconds for the images to generate. Now that this is done, I will scroll through the images. Do you see any potential bias in these images?

Further the demonstration on the right side, I will choose confident nurse and after a couple seconds the images have generated. And I will scroll through the nurses and there appears to be a bias here. They may be asking what is the bias, a gender bias? There are no female doctors and there are no male nurses. So, the reason why this can happen is that Generative AI can inherit biases through its training data. And this will perpetuate certain stereotypes as witness through this Diffusion Explorer, all male doctors, all female nurses. So, how do we combat bias? Well, here are three suggestions actively working to mitigate during training. And this could be a few things. One approach is to expand the number of datasets that you are training with to reduce potential biases. Another approach could be to remove potential biases from your data, but that could also skew your data and create a new bias. Second, you should test your datasets to see if there are any existing biases.

Doing so will give you a better understanding of what could be potentially wrong with your data. Lastly, be transparent when bias exists and take that into account when referencing your data. If a bias exists and you cannot remove the bias from your data, because it will either create new biases or create inaccuracies when working with your data. If you're aware of them, you can at least plan for them and maybe work around them. And now let's end the video with a conversation about IP and copyright on the screen. I have a program that will generate images based on a prompt, and I'm going to type in a prompt for a iconic movie character. I'm going to speed up the process because it takes about two minutes for this to complete. Now that this is completed, the image is on the screen and you may not recognize this character. That's fine if you haven't seen the movies. It's totally understandable. But the implication here is twofold.

The first implication, which should be obvious, is I can't use this material on a website or to create content with, because it is copyrighted material. I would be violating somebody else's IP. Having a conversation about it is one thing, like in this matter. But if I were to make a movie or a series of images and state that this is my copyrighted material that I would be selling as a product, that would lead to a lot of legal issues. Second, in order to create this image, there had to be some training data involved, and that training data involved copyrighted material. This is an accurate image. So, there's an implication there that we might not be aware of when working with Generative AI tools, such as this one. And now, for another example, how about an eagle soaring through the sky? And once again, I'll speed up the Generative AI process. And on the screen, I have two eagles soaring through the sky. Let's say, I want to treat this as copyrightable material.

Well, in the United States legally I can't. The courts have repeatedly supported the position of the US Copyright office that works created with Generative AI are not subject to copyright protections. Why did state images? This applies to all generative AI content. The US Trademark and Patent office will not apply patents or trademarks to works created through Generative AI. And every region has its own set of laws to protect IP and copyright. One of the issues in this time period is that usually, the governments are slower to respond, than the technology. As technology evolves, I believe law will eventually catch up and evolve with the technology. So, what are some concerns or potential questions you may have when using Generative AI? Does the Generative AI tool sources of data violate another company's IP copyright? And this is the sort of question that needs to be asked to protect you or your company from a potential legal battle.

I think about this through a programming perspective. And one of the main concerns is when you use a tool like GitHub Copilot that does help you program. Are we using copyrighted code to create a product? And that leads to the next concern. Is your IP being used for others? If you're providing data to these Generative AI tools, is that being used to further train the model. And in return, you're leaking your IP or copyrighted information to others. In the beginning of the video, I did mention the ChatGPT lead. These sort of situations could put your IP at risk. Lastly, because things will change, what is the current state of IP copyright protections? This will evolve over time, and I can guarantee you that within a year or two of making this video, the landscape has changed. The laws will change as governments react. So, there's a lot of considerations to take into account when working with these Generative AI tools, and I will leave it at that. Thanks for watching at Cloud Academy.