It is already transforming our society in profound ways, both positively and negatively. On one hand, AI is dramatically accelerating new scientific discoveries and enhancing human productivity. But at the same time, it is making entire skill sets and professions increasingly obsolete and is allowing fraud and cyber attacks to be conducted with astonishing speed, scale, and effectiveness without requiring any technical skill.
Furthermore, AI tech leaders are sounding the alarm that if the pace of AI development isn’t dramatically slowed down and/or regulated, then AI could become uncontrollable and potentially cause catastrophic harm, or even the extinction of the human race.
The fact that anybody with a computer or smartphone can now communicate with modern cloud-based AI using ordinary spoken/written language, and that it can understand and respond to us in a natural, human-like way makes it incredibly compelling to use. Almost addictive.
But at its core, AI is software, which has vulnerabilities that can and are being exploited by criminals, hostile governments, and even other autonomous, malicious AIs looking to steal information or cause harm. So knowing what the risks of AI are, and how to use it safely, is crucial.
Disagree with it, and ask it to convince you otherwise… Most AI models are taught to (at least initially) provide answers it thinks you want to hear, which may not be accurate.
Below are some best practices for how to use popular cloud-based “consumer” AI models (e.g. ChatGPT, Google Gemini, Microsoft Copilot, Claude) safely, as a non-business user (business users have additional risks that are beyond the scope of this article):
- Don’t allow the AI model to use your interactions for training purposes (whenever possible), and disable saving of your conversation history. Most AI models will allow you to disable these “features,” which are often enabled by default. Otherwise, any information you give the AI could be accessible by others, and could even be used to train other AI models.
- Never give any personal information to an AI, such as your name, personal health information, your home address, phone number, or other contact info, credit card or bank account numbers, Social Insurance Numbers, or passwords. Also, ensure your login access to AI is secured by two-factor authentication where possible and practical.
- Always verify any answer you get from AI, even if it sounds very compelling or was exactly what you expected (or wanted) to hear, especially if it’s regarding an important topic or decision that could have significant consequences. Use multiple trusted, third-party sources to verify the accuracy of whatever AI tells you, and ask the AI to specify what its sources are.
- Interact with AI as a police interrogator would. Be as specific as you can about the information you’re looking for, and constantly question the AI: ask it to explain or defend its answers, and refine its responses. Disagree with it, and ask it to convince you otherwise. Keep it honest, by asking it how confident it is in its answers. Most AI models are taught to (at least initially) provide answers it thinks you want to hear, which may not be accurate.
The capabilities of modern AI models are evolving exponentially, literally changing by the day, like no technology before it ever has, with the potential to even improve itself without any human intervention. So using AI carefully, knowledgably, and responsibly is more important than ever.
