One study showed that human beings treat things with the computer very similar. The study stopped the test image recognition to people and ask them to think like robots.
Participants were presented blurred images, and that they require to choose between A or B image may be captured by artificial intelligence.
The study found that in 75% of cases, humans and machines will choose the same answer, suggesting that both may be the same fraud.
These findings show how advances in artificial intelligence continue to narrow the gap between human and machine vision capabilities.
Since temporary, artificial intelligence systems has been better than a lot of information in terms of human memory.
People used to think humans have an advantage in identifying everyday objects such as dogs, cats, table or chair.
This study shows that when people see something on the computer on autopilot, they will be deceived. They think that graffiti is to train, fences and school bus.
According to a study at Johns Hopkins University, the brain's neural network simulation has recently been close to the human ability to recognize objects.
These so-called "reverse" or "teasing" image is a big problem, because they can be *** use and pose a security risk, but they suggest that seen in humans and machines in practice is very different picture.
In some cases, as long as the reconfiguration of one or two pixels can let the computer to Apple as a car.
In other cases, the machine can see armadillos and bagels like a pointless TV screen.
Professor Firestone said: These machines seem to never be able to identify the way human error identified objects.
But what is surprising is that no one really tested this. You know how people could not see anything without a computer?
To verify this, Firestone and professor of cognitive science at Johns Hopkins University professor Zhou Zhenglong asking people to think like machines.
The machine requires only a relatively small vocabulary to name the image. As a result, Professor and Professor Zhou Firestone on a computer cheating showed dozens of pictures, and provides the computer with the same tag options for people.
People asked in detail about how to determine the target computer options given. A true conclusion of the computer, and the other is a random answer.
This is a donut or a windmill? Telephone robot really easy to use?
It turns out that people are very much agree with the conclusions computers.
In 75% of cases, people will choose the computer the same answer.
Next, the researchers made more stringent requirements, allow people to choose between the computer's favorite answers and the second best answer.
On this issue, it is a bagel or pretzel on the spot? People choose to re-examine a computer, 91% of respondents agreed that the first choice of the computer.
Even when the researchers asked people to guess the object 48 option, even when the picture looks like a static TV screen, the vast majority of subjects selected target device selection, much higher than the proportion of randomly selected.
In separate experiments, the 1800 test subjects.
Professor Firestone said: "I found that if I put a man on the same computer environment, people will suddenly objection to the machine.
AI remains problematic, but this does not mean that the computer can see and be seen humans completely different.