If you talk to anyone who works in technology, they seem to know all there is to know about AI. As they should; it’s their full time job to know this stuff. But in talent acquisition, AI is just something we dabble with part time to make our lives a little better. Our priority is people. The problem is, AI isn’t exactly something that comes with a step-by-step instruction manual — this stuff is complex and confusing, and it’s pretty easy to feel dumb.
Guess what? We’ve all felt that way at some point. In fact, the only way to get smart about AI is to ask dumb questions until you have all the answers.
I’ve asked every dumb question possible over the years — and I’ve had tech experts who know way more about AI give me the answers.
This is what I found:
What is AI?
Let’s start with the basics, because if you don’t understand this, you won’t be able to understand anything. The tricky part is that “AI” isn’t just one thing, it’s lots of different things. AI can play chess, but it can also drive a car. It can also help schedule 1,000 candidate interviews instantly. Those things are all really different! But they’re still drawing from the same sort of technology. AI, at its core, is “technology that enables computers and machines to simulate human intelligence and problem-solving capabilities.”
What is generative AI?
ChatGPT, built by OpenAI, might be the most popular generative AI tool you would recognize. A user can ask GenAI a question (called a prompt) and the AI will almost instantly send back an answer that it generates (thus the “generative” part). The cool thing about GenAI is if you ask it another question, it treats it like you’re having a conversation, as it remembers not only the first question you asked, but also how it answered. So, your entire experience with the AI is contextual around the conversation you’re having with it.
What is a large language model (LLM)?
An LLM is just a big set of data. In ChatGPT’s case, OpenAI decided to use the entire internet as the data set. This creates a base for AI that allows it to have a conversation and predict what words come next in a sentence or complete a math problem. While that’s a huge data set to be trained on, it was also problematic as it turns out not everything on the internet is true! Who knew? There are only a few LLMs as they’re super expensive, and usually built by companies like Microsoft, Google, or Amazon.
What is “fine tuning”?
As generative AI continues to evolve, we’ll see many LLM models be “trained” on more private sets of data. The training is pretty simple — some geeks need to look at lots of data and decide what is “good” or “bad” and tell the computer that. But like thousands of times. Imagine your own company’s data having its very own generative AI model. You’ll see technology built-in HR around HR-specific data. The data is the real engine behind the power of AI.
Will AI take my job?
AI-empowered technology will assist in doing many things, especially administrative-type work and repetitive task work. But it will also assist in high-level data analysis, programming, etc. The reality is that it can make an average worker who embraces and uses the technology a 10x super worker. AI is like other technological advances in history on steroids. All the other technological advances in history eliminated certain jobs and created others. Yes, AI might take over your job if it can take over your job. This doesn’t happen like turning on and off a light switch. This happens slowly over time. Portions of jobs will go away with AI doing the work, and you’ll find higher-functioning work to do that AI can’t do.
Jobs and work shift with technology advances. Your worry shouldn’t be what work AI will take away from me. It should be hopeful about what work I can now do that I could never get to before. There were more bank tellers before the invention of the ATM and more retail workers before the advent of e-commerce. People adapt.
Is AI biased?
Yes. AI can be biased based on the data it was trained with, what it is asked to learn, and what it is asked to output. Are humans biased? Yes, at a rate that far exceeds anything we can control with AI. But here’s the difference: I can’t control the bias of humans, but I can control, measure, and change the bias of AI. AI might actually be the closest we’ll ever get to bias-free hiring and performance management.
Can AI just answer any question the way it wants to?
That depends on what kind of AI you are using to answer your questions. Generative AI, without controls, is simply predicting what the next word should be based on what it knows from its training set. It’s generating something brand new, so you literally don’t know what it could say until it says it. More traditional conversational AI that is based on “intents” tries to match what a user is saying and then get a pre-written answer from a system. But that only answers so many questions, so you’ll get some level of “does not compute” from it. The most modern systems use a technique called “retrieval augmented generation” to blend the best of both worlds — keeping the context of the whole conversation, while getting an answer that is approved and accurate.
How can I use AI in my everyday job right now?
In reality, you’ve used AI in your everyday job for most of the past decade. Do you have an iPhone or an Android? Do you use it for work? Do you use Microsoft Outlook or Google Office Suite? You’ve been using AI. This is why it’s funny that corporate legal teams are getting all uptight about AI today. We’ve been using AI embedded into our daily productivity tools for a decade or more but never questioned it! While many people are using ChatGPT, Gemini, and other generative AI tools to help them with many things, most of us will use AI tools that will be embedded in the systems we use every day for our work already.
Can AI take over the world?
Come on, you know you’ve asked this (at least to yourself). This is the biggest fear most people have: AI is going to take over! “One moment, I’m driving down the street, and the next moment, the AI takes control of my car and makes me go to the donut shop and buy all the donuts!” Yes, someone could (actually, they did) make an AI that can drive your car for you. But no, AI can’t force you to buy donuts! Like any software program, even one that is learning on its own, is still bound by the rules the software programmer has put in place. We saw this learning with early versions of generative AI that could be “prompted” by a person to say racist things or believe things that weren’t real. The AI was learning like a young child. Later versions continue to fix these early bugs.
Artificial General Intelligence (AGI) is what people are talking about when they say robots will take over the world. This is when robots can actually think for themselves, like humans. Make their own decisions without rules. While we can imagine AGI, science isn’t close to making this happen. We can theorize about how, but the technology to make this happen hasn’t even begun to be built, as the human mind is far too complex for any model we have that is remotely close at this point in time.
