How to Work with Your AI: What AI Is and What It Does
Part of a series | AI Insights
Key takeaways:
Artificial intelligence (AI) is a tool that processes large amounts of information quickly but does not think or understand like humans.
Limited memory AI retains information temporarily to make predictions without learning from past experiences.
Generative AI retains vast amounts of data and can be programmed for tasks like creating content and analyzing data.
Accountability remains with humans — organizations and HR are responsible for the outcomes of AI-enabled decisions.
AI has been present in workplaces for a long time, often in the form of tools like autocorrect.
AI is highly effective in data processing and organizing but cannot fully replicate human communication and understanding.
As AI becomes part of our workplace, it's essential to understand what it is, how it generally works, and where it can be most effective. AI is different than many of the tools and processes we know and use. While AI can process libraries full of information much faster than we can, it doesn't actually think, understand, or know anything in the human sense. So, it's important to understand where AI is useful and to use it in ways that are both responsible and effective.
Accountability and responsibility in AI use
When we use any tool at work, we also have to make sure that outcomes are fair, consistent with legal requirements, and that we have the right information and insights to support what we're trying to do. While AI can support our judgment, it does not replace accountability. Especially with employment decisions, the organization and HR will own the results, whether or not AI is involved.
This is the first in a three-part article series to help you understand how AI works, what it does and can't do, how to effectively interact with AI and what to consider when choosing work tools that use AI.
Understanding the basics of AI
In simple terms, AI is a branch of computer science and engineering that uses vast amounts of data and processing power to perform tasks that would take humans much, much longer.
There's no way to list all the current and possible uses for AI and the list would change quickly. But one way of describing AI is to categorize it by how much data it stores and how it uses that data.
Types of AI: Limited memory vs. generative AI
Limited memory AI can temporarily store information to predict what to do next. Self-driving cars use limited-memory AI, along with large sensors that scan current conditions and determine what to do based on maps and other programming.
While there's a lot of processing of environmental data and sophisticated algorithms to direct movements, self-driving cars generally don't retain or learn from that information (at least for now). That means when a self-driving car goes through the same intersection many times, it doesn't recognize it or remember what it did the last time through. It just senses what's there and determines whether to go, stop or turn based on its destination. It's like a goldfish that swims through its tank and thinks, "Hey, it's a castle!" every time it goes by.
Limited-memory AI stores and processes information over a short time to handle the task at hand. Examples include screening tools that search for skills in a resume and performance management tools that provide help and feedback based on real-time employee actions. But none of these AI applications retain the information or use it in future tasks.
AI is a tool, not a person. But a tool in the hands of a person is powerful. A pen cannot write a poem, but someone with a pen can. Tools have a purpose, not a personality.
Othman Chhoul, Dir. of Product Management, Agentic & Generative AI Solutions, ADP
Generative AI, like large language models (LLM's) do retain information—gob smacking amounts of data. Some kinds of LLM's also add new information to stored memory and integrate it into their functioning. These systems can be designed and programmed for different functions from chatbots, to tools for automation, to systems that can answer complex questions and perform tasks.
Some examples of Generative AI include programs that can create job descriptions and job postings, chatbots that can answer employee questions about policies, and data analysis that can monitor and detect changes and patterns in real-time.
The evolution of AI tools in everyday work
AI can also be programmed to follow multi-step processes that can account for different variables throughout the process. Sometimes, it can scan systems and check its work. It can even find errors and fix them.
Limitations of AI
But AI systems don't know or understand information in the human sense. People still have to tell them what to do through the design, programming, data, and interactions. Fundamentally, AI systems are complex pattern-detection and prediction machines.
Learn more about how ADP can help deliver better outcomes in the AI era
Applications of AI in the workplace
We've been using some forms of AI at work for a long time, even before we called it AI. Autocorrect is one example. Beyond just spelling, autocorrect can scan for language patterns and predict the word the writer is most likely to want next. It can even recognize words and phrases you regularly use and offer more customized choices over time.
But we usually wouldn't let autocorrect write the whole note, mostly because it doesn't really know what we want to say. Getting meaning and communications right is important.
We continue to figure out new ways to use AI all the time. Some approaches hope to mimic human thought and intelligence one day. In the meantime, here are some things AI is effective at doing right now.
Data processing and organizing
AI can process vast amounts of data. Imagine taking billions of words and images, each with millions of parameters and instructions for what to do with them and doing it all in minutes. It's mind-boggling to try to make sense of. But AI systems work differently than brains—understanding and sense making is not what AI does. Instead, AI systems process the data they have to respond to the instructions or prompts they are given. They do amazing things, but they don’t think the way people do.
Anything you can do with logic, math, and data, you can do with AI. One classic example is categorizing and organizing information. On some level, this is what many of us do with all those emails, documents, and meetings we go to. Find the thing we need to understand to move us to the next step. Now, imagine taking everything you ever wrote in digital form and summarizing the things you've worked on over the last 10 years by topic and subtopic while you go get coffee.
Pattern recognition and prediction
If you've ever seen the memes of photos where corgis resemble bread loaves or chihuahua faces pass for blueberry muffins, you are watching pattern finding in action. While we're still working on the business case for puppies, what people do and the observable ways they do it can tell us a lot about how things really work, how to make our approaches more efficient or effective, and how we can understand our work and our organizations in new ways.
Predicting events or outcomes is a subset of pattern finding. At its most basic, AI prediction can tell you what word comes next, what step comes next, and the likelihood that something will happen. This is how AI systems write text. They predict the next word, instance by instance, over and over, only really fast.
Insights and analytics outputs of all this data and processing. AI can collect, synthesize, and report anything we can quantify, measure, or define. AI can also organize it, keep it all up to date, detect and measure changes and send alerts when significant changes happen.
Monitoring and anomaly detection
Checking for anomalies and correcting problems is another type of pattern recognition; here the system recognizes how things are supposed to work and flags anything that's out of line. Then the system can also be designed to fix the problem or guide you through the steps.
Adaptation over time
Adapting to new and more information makes AI systems dynamic and able to handle continuous change. Unlike humans, AI does not find this to be stressful; it's just how it works. But this ability allows nearly real-time detection of changes, updating of results, identification of new patterns, and making new predictions.
What can't AI do?
If it isn't based on math, logic and data about things that can be counted or measured, AI can't do it. AI also only knows the information and programming it has. It can't draw from outside context or make creative associations. It doesn't even understand what it does. It's not human.
Only humans can take information and apply wisdom, common sense, and emotions. Only humans can decide what's important and why it matters. And only humans can make decisions with compassion and based on fairness.
Especially in the field of HR, as the keepers of people data and those who deal with people's careers and lives, it's essential to understand how AI works and its limitations.
For HR, this means going beyond bias audits. It’s about assigning ownership for every AI decision, aligning policies across teams, and keeping open conversations with employees about what AI is doing and why.
- Naomi Lariviere, Chief Product Owner, Global Product and Innovation, ADP
While we can hand off some of the work to AI, we can't outsource responsibility for what happens when we use it. That's why we need to use AI with care.
Learn more about how ADP can help organizations manage workforce complexity, maintain compliance and deliver better outcomes in the AI era.
