Part 1-The Security and Ethics of AI When Working with Data
Whether you think AI is going to take your job soon or it’s the best thing since sliced bread. One thing everyone can admit is that AI is going to become an integral part of working with data. Learning when it’s appropriate to use language models Like Chat GPT and Gemini AI is extremely important. When it comes to data security and integrity, safety and security is the top priority.
Data Security
The first rule of working with data is NEVER share sensitive information online. This includes names and company specific terminology. The only exception is organisational level platforms like internal communication channels or internal cloud servers. Every person working in data should know this. Well, the same logic applies to language models. While a majority of major language models claim a ‘high’ level of security, it’s important to remember that nothing on the internet is 100% secure. If you can avoid being the cause of a data breach, it’s best practice not to be.
The second rule of working with data is to always ask for permission, not forgiveness. When working with sensitive information, checking that you are handling said data with the owner-of-said-data’s best interest in mind, is the key to ensuring good data security practices.
It doesn’t matter which AI model you’re using, if its available to the public they’re trained using user data.
If for example, you’re using non-anonymised data. You run the risk of the AI model training and using said data at a later date when the model is prompted by another user. One way you can get around this is to create ‘dummy data’ with a similar schema. You can use tools like mockdatagen.com and Mockaroo.com to create ‘dummy data’. This way, if you need something specific from an AI model. You can use the ‘dummy data’ with a similar structure in the language model. Only do this when necessary, because in most cases as a trained data analyst you can often do it yourself much faster.
An additional recommendation when using AI models, is to check for settings which opt-out of chats being using for training or chat data being stored. Getting into the habit of turning these on as an extra safeguard isn’t going to hurt. Do not rely on or trust these options 100%, because in many countries, laws around data use and retention are still being amended to include companies which develop generative AI. What that means is, data storage and retention is up to their own discretion in this moment.
I treat these opt-out options like a padlock on the door, someone can still break into your house, but it adds a little bit of extra security.
Utilising all those these data security practices is the best way to ensure maximum security when combining AI and Data.
Data Ethics
When working with data and AI models, ensuring that are using them ethically is paramount when it comes to integrity as a data analyst. Using AI to complete tasks or analysis without appropriate communicating, can affect how you are viewed as an analyst and even return unfavourable outcomes. There is absolutely nothing wrong with using tools which are at your disposal to increase productivity. As long as you’re following the guidelines set out in regard to data security. You just need to be mindful of everything you’re putting online whether it’s in a language model or not. An example, if I were working with a company’s data, I would ask what their guidelines and polices are regarding using AI language models. If they don’t want AI used at all, then that’s what must be done. Regardless of whether or not it’s not as efficient, you need to follow their guidelines.
The CEO of OpenAI (The company responsible for ChatGPT), Sam Altman mentioned how surprised he was to learn how trusting people are of ChatGPT. He elaborated that although ChatGPT is one of the most advanced models on the market, It’s still very often wrong. Language Models are a very powerful tool. If you don’t know how to use a powerful tool, then there’s a strong chance it will be misused. So as long you’re prioritising the security and ethics of data when working with AI then you shouldn’t run into any issues.
Check out part 2 here where I will be going over how to prompt AI models to get the most out generative AI.
