Setting Realistic Expectations
Before we get into the nitty-gritty, here’s a little nugget of wisdom: setting realistic expectations is key when it comes to AI. It’s easy to get carried away with the hype, but AI isn’t a magical genie that grants your every wish. So, before you embark on an AI project, make sure it’s actually feasible.
The Inflated Expectation Syndrome
Ever heard of CEOs having sky-high expectations of what AI can achieve? It happens more often than you’d think. Sometimes, the media and academic literature paint such a rosy picture of AI successes that it creates unrealistic hopes. But let’s be clear, AI has its limits, and it’s not a cure-all.
What AI Can Handle
Now, let’s get to the good stuff. AI has come a long way, and there are tasks it can handle like a champ. One rule of thumb is that if a task takes you less than a second of thought, AI can probably automate it using supervised learning. Here are some examples:
- Determining Car Positions: You glance at the road, and in less than a second, you know where other cars are. AI can do that too.
- Spotting a Scratched Phone: A quick look, and you can tell if a phone is scratched. AI can handle that in a snap.
- Transcribing Speech: Listening to someone speak and jotting down what they’re saying doesn’t require much pondering. AI excels at transcription.
Where AI Falls Short
Now, let’s shine a light on where AI struggles. There are tasks that, as of today, AI simply can’t tackle effectively. One of these is generating complex, empathetic responses to customer queries. Here’s a real-world example to illustrate:
Imagine you run an e-commerce website, and a customer sends an email complaining that their toy arrived two days late for their niece’s birthday. They want to know if they can return it. If you want AI to determine this is a refund request and route it to the refund department, you’re in luck. AI can handle this kind of classification task, mapping the input (customer text) to one of several outcomes (refund request, shipping problem, etc.).
But, and it’s a big but, if you want AI to craft a heartfelt response like, “Oh, sorry to hear that. I hope your niece had a good birthday. Yes, we can help with that,” you’re in for a challenge. Generating such responses with empathy and appropriateness is a tough nut for today’s AI.
The Data Dilemma
One of the reasons AI can’t always deliver is the data dilemma. AI needs loads of data to learn effectively. For instance, in the customer support scenario, you’d need a significant dataset of customer emails paired with accurate labels (refund request, shipping query, etc.) to train an AI effectively. With a small dataset, AI often falters, generating generic or even gibberish responses.
Rules of Thumb for Feasibility
To help you gauge whether a project is feasible or not, here are a couple of handy rules of thumb:
- Simplicity Rules: If a concept takes less than a second of thought, AI might handle it. Spotting cars on the road? Yes. Writing empathetic responses to diverse customer emails? Not so much.
- Data Is Power: Having lots of data, both input (A) and output (B), improves AI’s chances. In our customer support example, thousands of emails paired with their respective labels (A to B mapping) is a recipe for success.
Wrapping It Up
AI is indeed the new electricity, transforming industries left and right. But remember, it’s not magic, and it has its limitations. I hope this video has shed some light on what AI can and cannot do, helping you make informed decisions when tackling AI projects.
