In this solo presentation, former Google employee and AI thought leader Toju Duke recounts his seven‑year journey into artificial intelligence, emphasizing both its productivity benefits and its potential harms. He introduces a concise, three‑step framework for responsibly using AI—Ask, Assess, and Analyze. The "Ask" step urges users to clearly define the task they want AI to perform and to consider whether delegating the task might erode cognition, citing an MIT study linking heavy ChatGPT use to a 45% drop in cognitive abilities. The "Assess" step focuses on selecting the appropriate AI tool or agent for the defined task, whether a large language model, multimodal system, or a suite of agents, and stresses thoughtful research. The "Analyze" step reminds users that AI outputs are error‑prone and must be continuously fact‑checked and reviewed. Throughout, Duke uses email management as a concrete example to illustrate both the power and the pitfalls of over‑automation. The talk concludes with a call for human oversight to mitigate risks while leveraging AI’s efficiency gains.
Chapters
I remember making an intentional move into the field of AI close to 7 years ago, as I recognized the capabilities of AI way back in 2019. I started my journey into the field of AI from the day I attended a machine learning training when I was working at Google, and I haven't looked back ever since. Since then, I've worked on ensuring AI doesn't only drive progress across businesses and individuals, but also towards reducing the harms and risks it projects. And I strongly believe that everyone should work with AI because it does help save time and drive efficiency. But since it's still in the very early stages of development, you need to understand how to work with AI effectively. Doing this will greatly reduce the associated risks to you, the user. And that's why I recommend using the simple framework when using AI, which I call the 3 A's. These are ask, assess, and analyze. One, ask yourself what you need AI to do. Is there a task or goal you're trying to achieve which you think the AI tool could help you with? This could be an LLM, a multimodal system, or a series of AI agents. For example, you may have so many emails that need your attention and you'd like to know which emails to prioritize. Yes, AI could definitely do this for you. But on the flip side, you may have so many emails that need an urgent response and you'd like to respond to all of these emails immediately. Yes, AI could also do this for you, but at this stage you need to stop and pause. If your AI agent sends all of these emails on your behalf without you having to write, read, or analyse them, what would that do to your cognition? And also to your sense of worth and value? Did you know that a recent study from MIT showed that heavy ChatGPT users saw a 45% decline in cognitive abilities? Of course, this isn't surprising, as the less time you spend exercising a part of your brain, the more dormant it becomes. 2. Assess. The next thing you'll need to do is understand which AI tool or agent is best to help you execute your given tasks. Again, this requires some thorough thinking, analysis, and research, and it doesn't matter if these tasks are for work or for personal purposes. 3. Analyze. It is a well-known fact that no matter how good an AI tool is, it's still prone to make mistakes, huge mistakes in fact. So don't allow your AI agent or tool to complete all of your work for you without fact-checking and carrying out reviews on an ongoing basis.
