Practical AI

Anything You Can Do AI Can Do Better (?)

• By Howard M. Cohen
• 01/15/2025

Geoffrey Hinton, one of the "Godfathers of AI" who has won both the Nobel Prize and the Turing Award, has estimated that there’s a 10-20% chance AI could extinguish humankind within 30 years.

This is not an uninformed, unintelligent, alarmist person crying that the sky is falling. He actually knows what he’s talking about, better than maybe anyone else. Hinton’s observation suggests that AI will become more intelligent, more competent, and superior to us in so many ways that we may not stand a chance of surviving. This raises the question as to whether or not AI can make the same claim as AO—Annie Oakley—that is, who first claimed, "Anything you can do I can do better" (as composed by Irving Berlin in 1946 for the Broadway classic, "Annie Get Your Gun.")

Start With the Simple

You’re working in a retail setting, and your boss asks you, "What products has Joe Customer bought from us in the past year?" You turn to the company’s ERP system, look up Joe Customer’s account, find the list of his purchases and read them off to your boss. Now your boss asks you, "What do you think Joe Customer would want to buy next?" Looking at the list of products Joe has already purchased; you may feel that you can offer a relatively accurate prediction.

Now let's say, instead of asking you, your boss asks the company AI system, and it instantly returns the list of purchases. No big problem there. But it also answers the question about what Joe might like to buy next. To determine this answer, it goes beyond the list of past purchases and consults every last bit of data it has about Joe Customer, including extracts from his social media, transcripts of conversations with your Contact Center, and online profiles. It then offers is prediction. You are both likely to be equally correct in your answers to the first question. The list of purchases is factual.
Another thing you have in common is that you may both be wrong about your answer to the second question. You may not accurately predict what Joe will buy next.

Your process for coming to that answer was pretty simplistic and depended upon your necessarily  limited observations, your knowledge of the products involved, and your intuition. The AI assembled a set of relevant data and ran some algorithms against them to produce its recommendation. It still may be wrong. But when an AI is wrong, what many refer to it as a "hallucination," it is likely that the AI stood a better chance of being accurate than you did because it included far more data.

What AI Can Do Better

AI can clearly process larger data sets far faster than a human. AI can test multiple configurations or combinations of elements in an attempt to identify the most efficacious, as it is has used extensively in creating new molecules and proteins for medications. This is also how AIs play chess, and the challenging game called Go, by trying and testing various moves and sequences of moves to find the most effective move. Speed and volume are AI’s two greatest strengths.

Until now, humans have been better at intuiting things and making subjective assessments. However, recent improvements in the technology have narrowed this gap. AIs are now far more capable of making subjective "shade-of-meaning" decisions and show the promise of continuing to get better.

As we proceed into the "agentic" age of AI, the ability to set and fulfill goals is becoming more emphasized. AIs continue to be far more precise than humans, but are also becoming more and more capable of correctly interpreting human imprecision. The fundamental question we all need an answer is whether an AI can be smarter or more intelligent than a human. Can it learn to "want?" Can it learn to "wonder?" Or is it simply a digital device capable of finer shadings of meaning?

Put These Strengths to Work

Our ability to create truly practical applications with AI will increase as we come to more fully understand and appreciate the strengths and weaknesses of AI.