Skip to main content

Why a Machine Defeating a Human Does Not Signal the Apocalypse

The AI program AlphaGo's victory over go master Lee Se-dol was impressive, but it was a relatively narrow achievement

This article was published in Scientific American’s former blog network and reflects the views of the author, not necessarily those of Scientific American


Earlier this week, the AlphaGo program from Google DeepMind defeated one of the world's top players in the board game called Go, 4 games to 1. The achievement is a milestone in artificial intelligence and has led to intense debate about the long-term implications to society.

The fact is—and this may sound gloomy—machines will likely replace us in any well-defined task. When two otherwise equal devices compete to solve a problem, the winner will be the one that is more streamlined. Getting rid of overhead and task-irrelevant capabilities makes a machine more effective, and we humans are good at just too many things to be the best at any one of them. Thus, whenever the goal of solving a desirable subproblem can be clearly and precisely specified, special purpose hardware or programming will outpace more general human intellect. 

There are plenty of examples of this phenomenon throughout history. A handaxe can dig up roots better than the human hand, but it is useless for holding a spear the way a hand can. A hoe is better still at digging up roots, but it's not good at cutting wood compared to a handaxe. Specialization makes things more effective at one task but less effective at others. Similarly, robot arms that are fast and predictable but lack the sensitivity to change a diaper or give a friendly pat on the back are going to be better at building cars than people on an assembly line. Computers that do nothing but direct network traffic are going to be better at connecting phone calls than human operators. Indeed, we have already lost these battles.


On supporting science journalism

If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.


There's nothing for us to be ashamed of here. If you hold the goal constant but repeatedly broaden the space of competitors, performance improves. Consider the case of deep sea diving. The deepest a person has even swum is 331 feet down. With the help of tools like fins, weights, and inflatable bags, the depth can be more than doubled to 702 feet. But, with unrestricted use of machines, people have traveled to the very bottom of the sea—around 5.9 miles. We can achieve much more with an expansive design space than with one that is more limited.

The same holds true in the mental domain. Computers have memorized more digits of pi than people (13 trillion compared to the human record of 70 thousand) and multiplied 8-digit numbers faster (858 trillion in the same time the best human being can do 10). Computer programs have achieved significantly higher ratings in chess (3100 Elo versus around 2900 Elo, which means the best program would beat the best person roughly 76% of the time in a head-to-head match). AlphaGo's defeat of Lee Se-dol suggests that Go may be another realm where machines have outpaced us.

So, if anything we humans might be good at will ultimately be done better by specialized machine, what hope do we have in remaining valuable?

I think the key is that human intelligence is constantly redefining and extending itself, incorporating new facets built on top of everything that came before. As such, the rules of the game constantly change. The metrics for success change. Even the very purpose of existence changes.