AI Can Be Empowering
From tool to partner.
Background
Last weekend, I gave a presentation to a ground-up community consisting of members who were at least 50 years of age (eligibility criteria). The founder, a friend, reached out to see if I would be interested to give a talk on how the more elderly members of society could navigate the Age of AI. I researched the following questions:
- ”Will AI kill you?”
- ”Will AI displace you?”
- ”Will AI empower you?”
My 109th article is an expanded section of “Will AI empower you?”. Exactly “how” will AI empower you? What would “empowerment” look like?
(I write a weekly series of articles where I call out bad thinking and bad practices in data analytics / data science which you can find here.)
AI as a Person
Despite our best efforts, we can’t help but anthropomorphise user-facing AI solutions. AI has been quietly embedded into our lives for a sufficiently long time — e.g. in every smartphone, there is an AI dedicated to managing the battery health and battery life. These embedded forms of AI have radically and significantly enabled our lives, removing unwanted operating frictions to create more seamless and joyful experiences. I don’t believe any of us are losing sleep and feeling threatened by these kinds of AI. We would be loathed to have them removed from our lives.
But the arrival of Generative AI (Gen AI for short) in the last 3 years has uprooted all of that. Gen AI is a user-facing solution built on Large Language Models. Because it’s designed with a user interface (UI) and an intentional user experience (UX), we can’t help but see this form of AI as an entity. The ability to communicatively interact with Gen AI makes humans want to develop a relationship with it; it makes us inadvertently anthropomorphise it.
It is this psychological shift in the relationship between man and AI that has resulted in new anxieties. Will AI disrupt our livelihoods? Will AI displace our jobs? But it is precisely this evolving relationship between man and AI that will empower us in the coming decade. But only if we understand what and how it might empower us.
Information Alignment
I was made aware, through a conversation with a client, about this amazing article on Medium — “The 4 Quadrants of Prompting: A Cognitive Framework for Working with AI”. It is the basis of the exposition that follows. A lot of the “Man vs AI” anxiety is psychological in nature, stemming from both a poor understanding and poor visibility of the boundaries between the 2 entities. Enter “Information Alignment”. I’ve created the diagram below to summarise the key take-aways from the article. The construct is simple: on one axis is what you know or don’t know, and on the other axis, is what Gen AI knows or don’t know. The intersections of the 2 axes create 4 quadrants, much like the classic SWOT analysis. (It is important to call out that what Gen AI knows is simply the collective human thought and knowledge that has been fed into the large language models.)
I found this perspective of Gen AI particularly illuminating for the following reasons:
- The upper quadrants are all about using Gen AI for productivity gains. It’s an efficiency play. Gen AI is enabling. In both these upper quadrants, you can get the job done if you needed to.
- The lower quadrants are all about using Gen AI for insights. It’s about improving effectiveness. It is here that Gen AI is empowering. In both these lower quadrants, you probably can’t get the job done on your own.
- The way we interface with AI should take into account which quadrant we are operating in. At present, the UI / UX design for Gen AI is still fairly unimaginative.
AI as Enabler
Let’s look at the top 2 quadrants. We start with the one on the right, where both you and AI are knowledgeable about the task at hand. Here, AI is nothing more than a tool, albeit a pretty impressive one. As with any tool, you need to need perfect clarity on what you are solving for, and your prompts (i.e. the instructions you give to the Gen AI interface) must be precise. It must provide you a single output that meets your specifications. For example, if you are intending to use Gen AI to summarise your meeting (assuming you have an audio recording of it), you need to tell the Gen AI to “create a 1-page summary of this meeting, using the attached format, highlighting key themes and take-aways, including follow-up actions if any.”
In the left-upper quadrant, where you are knowledgeable about the task at hand, but AI isn’t, you need to treat AI as an assistant. In that quadrant, AI lacks situational awareness, constraints, preferences, and an understanding of what success should look like. You shouldn’t be telling Gen AI to simply “plan a vacation for me this summer”. You need to specify the time period, the possible locations, the presence or absence of travelling companions, the preferred travel mode, the budget, the preferred activities, the preferred food. For example, your Gen AI prompt should be: “Plan a Malaysian vacation for me and my family, consisting of my wife and 2 pre-teen children, from Jul 1–10. Budget is S$1000. We love eating local food and visiting quaint small towns. Can’t drive more than 4 hours. Need recommendations on destination, accommodations, activities, and restaurants.”
The tasks in these 2 upper quadrants require instructions that are single-threaded and uni-directional. They benefit from the current session-based UI / UX design, which is a nod to the Age of Computer Programming.
AI as Empower-er
In Jun of this year, I wrote an article arguing that AI can’t cover up your incompetence. The gist of that article was that AI cannot imbue you with competencies that you don’t already have. I continue to stand by that claim. However, AI can empower you. And what I mean by “empower” is that it can amplify your strengths and abilities, making you even more effective.
Let’s look at the bottom right quadrant, where AI is knowledgeable about the task at hand, but you are not. The interaction with AI in this quadrant is a souped-up version of Google Search. Rather than the trial-and-error searches based on keywords, Gen AI plays the role of an advisor that curates and synthesises the knowledge you are seeking. It doesn’t give you piecemeal information but connects it based on your explicit context. You need to provide it with enough details on why you need the information, and what you intend to do with it. For example, if you need information to decide how best to pivot your career, you can write the following prompt: “I am a 40-year-old software engineer looking to pivot into my next and final career. I am unsure what pivots are going to be more successful? Should I pursue a similar role but in a different industry, or should I stay in the same industry but take on a different role? Can you interview me to find out more about my specific challenges?”
Finally, we look at the bottom left quadrant. This is an interesting space because both you and the AI are not knowledgeable about the task at hand. These are the kinds of tasks that you would brainstorm or co-create with a trusted friend; AI plays the role of that trusted partner. A good example would be if you are considering leaving your well-paid corporate job to go launch a start-up. How would you go about thinking of the pros and cons? What research would you need to do? Whose opinions would you need to seek? You could write a prompt like such: “I want to explore laughing a start-up business. I have 10 years of data science experience with a retail bank and have reached a senior level in my career. I am able to lead complex projects and influence key stakeholders to achieve successful outcomes. I feel I should continue to leverage my data science skills and reputation for my start-up business. Should I? If so, how would I go about exploring possibilities systematically? Can you assume the role of a consultant so that I can brainstorm with you on this?”
The tasks in these 2 lower quadrants are multi-threaded, bi-directional, conversation-based. The current session-based UI / UX design is poorly equipped for these kinds of information-rich interactions. We have yet to crack it. But I’m optimistic that we will. But despite the wait, we can still empower ourselves meanwhile by recognising the quadrants we are in, and crafting more appropriate session-based prompts.
Conclusion
I have found the 4-quadrant information alignment approach to interacting with Gen AI to be a particularly useful starting point to explore a whole range of topics like job displacement / disruptions, role augmentation, prompt structure, interface design. Before this, I didn’t think about which quadrant I was operating in, and the therefore, missed the opportunity to better define successful outcomes. Even the simple recognition of AI as a tool vs as an assistant can give you a significant leg-up in productivity. It can take away so much mechanical-cognitive toil, and give you back precious time. But learning how to operate with AI as an advisor and a partner is going to be the sweet spot that empowers and transforms the workforce.
