How can artificial intelligence harm humans and what is the solution?

When we talk about the dangers of artificial intelligence, the focus is usually on unintended side effects. We are worried that one day we will accidentally create a super-intelligent artificial intelligence and forget to give it a name; Or we use algorithms to convict criminals whose training data is loaded with racist biases.

But that’s only half the story.

What about those who intend to use artificial intelligence for non-criminal, criminal and harmful activities? Isn’t it more likely that danger will occur from them – and even sooner? The answer is yes, according to the opinion of more than ten experts from institutes such as “Future of Humanity”, “Existing Risks Research Center” and even the non-profit company “OpenAI” founded by Elon Musk, the answer is yes.

In a recent article titled “Malicious Use of Artificial Intelligence: Anticipation, Prevention and Deterrence » published, these researchers and academics have discussed some of the ways in which artificial intelligence may harm us in the next five years and what we can do about it.

Miles Brundage, one of the authors of this article from the Future of Human Institute, told The Verge website that because artificial intelligence can attack us with deadly attacks, we should definitely not panic or give up hope.

“I like to take the optimistic view that we can do more,” says Brundage. The point is not to paint a dark picture. There are many defenses we can develop and many things to learn. I don’t think it’s all despair, but I see this article as a call to action.”

The report by Brundage and the teams that collaborated with him is wide-ranging, but it focuses on a few key ways that AI can create physical and digital security risks, or create new risks altogether.

In this article, there are five suggestions to combat this problem, including informing artificial intelligence engineers about malicious uses in their research, as well as establishing a dialogue between academics and lawmakers so that there are no surprises when a crime occurs. >

Let’s start with the possible risks: one of the most important risks is that artificial intelligence will significantly reduce the cost of some attacks and replace human labor.

For example, consider phishing attacks, where people are sent messages to somehow obtain their account information. Like an email from shopping sites or an email from the bank where you have an account.

Artificial intelligence can do most of the work in these attacks and create a special message for each person based on the analysis of information and data. Or imagine chatbots becoming so advanced that they can play the role of your close friend and ask you for your email password.

This attack model seems complicated, but if this software is created once, it can be used again and again without additional cost. Phishing emails are already very harmful. These emails caused the publication of personal images of famous people in 2014 and the leaking of John Podesta’s emails from Hillary Clinton’s campaign in the 2016 US elections.

The leak of Podesta’s emails not only changed the outcome of the 2016 US election, but also led to conspiracy theories such as Pizzagate, which nearly killed people. Now imagine what an advanced artificial intelligence cannot do.

The next important point in this report refers to adding new dimensions to current risks. Consider the same example of phishing. AI already by listening to a few minutes of a person’s speech can imitate his voice. Or it can simultaneously use existing videos and transfer your facial movements. bush-video” target=”_blank” rel=”noopener”> replace the facial expressions of a known person. This report focuses on the risks that will emerge over the next five years and are quickly becoming a major issue.

And of course, there are a series of other uncomfortable actions that artificial intelligence can exacerbate. Political manipulation and propaganda (such as audio and video manipulation that can be a huge problem), as well as security surveillance, especially when it targets minorities.

A clear example would be China, a country with security cameras and facial recognition, a large area where the Muslim minority Uighurs live in Surveillance is.

Finally, the report points to the entirely new dangers that artificial intelligence can pose. The author mentions some possible scenarios, including placing a bomb in a servant robot and transferring it to a government home ministry. Using its special capabilities, the robot detects a certain politician and detonates the bomb when it gets close to him.

This example refers to two basic developments, one is the use of artificial intelligence in physical products and the second is the use of software technologies such as facial recognition. It may seem a little imaginary to refer to such stories, but the truth is that we have already seen innovative attacks by artificial intelligence.

The face swapping technology had made it possible for users to replace the faces of celebrities with the faces of immoral movie actors without any limits, which caused a lot of controversy in the past weeks. Although so far this story has not taken a particular victim, those who misuse technology will be able to blackmail or harass people they know by making such images.

The solutions presented in this article are simple and include:

• Artificial intelligence researchers should be aware of how their work can be misused by others
• Legislators should learn from experts about this issue
• People working on artificial intelligence should learn from cybersecurity experts to protect their systems
• Ethical frameworks for artificial intelligence should be created and followed
• More people should be involved in this story. Not just AI scientists and lawmakers. but scientists of ethics, businesses and the general public

Or in other words: a little more talk and a little more work is needed.