As if 2020 wasn't catastrophic enough, a team of academics, policy experts and AI insiders are warning us of trouble on the horizon. 18, to be precise. These are the top AI threats we should worry about in the next 15 years.
Science fiction and popular culture want us to believe that our undoing would come from riots of intelligent robots. Actually one recent study in Crime Science reveals that the worst threat will have more to do with us than with the artificial intelligence itself.
By evaluating threats based on potential harm, profitability, viability and defeat, the group identified that the deepfake, a technology that already exists and is spreading, represents the highest threat level.
Deepfake, the number one enemy
There is a fundamental difference between a robot uprising and deepfake technology. The first could cause material damage. The second is systemic: it could cause the erosion of trust in people and in society itself.
The worst AI threat can always seem like something far in the future. After all, how can AI hurt us when Alexa can't even properly set a pasta timer today? Shane Johnson, Director of the Dawes Center for Future Crimes at UCL which funded the study, explains that these threats will become increasingly sophisticated and intertwined with our daily lives.
“We live in an ever-changing world that creates new opportunities, good and bad,” warns Johnson. “This makes it a priority to anticipate future crime threats so that policymakers and others with the expertise to act can do so before serious trouble occurs.”
The authors admit that the judgments expressed in this study are inherently speculative in nature and influenced by our current political and technical landscape, but the future of these technologies cannot be ignored.
How the "playlist" of the worst threats was born
To formulate their predictions for the future, the researchers brought together a team of 14 academics in related fields, seven private sector experts and 10 public sector experts.
These experts were then divided into groups of four to six people and given a list of potential problems that the AI could create. Scenarios ranging from physical threats (autonomous drone attacks) to digital forms of threats such as phishing.
To make their judgment, the team considered four main characteristics of the attacks:
Potential damage
Profitability
feasibility
Vulnerability
For experts, the damage of deepfake would be material, mental and even social. They could also harm other artificial intelligences (for example by evading facial recognition) or be indirect fruits (for example blackmailing people using a deepfake video).
While these factors cannot truly be separated from each other (e.g. the harm of an attack might actually depend on its feasibility), experts were asked to consider the impact of these criteria separately.
The scores from the various expert groups were then ranked to determine the overall most damaging attacks that we may see through AI (or by an AI) in the next 15 years.
Spoiled for choice
By comparing 18 different types of AI threats, as mentioned, the group determined that video and audio manipulations in the form of deepfakes are the worst threats ever.
“Human beings have a strong tendency to believe their own eyes and ears. This is why audio and video evidence has traditionally been given much credence (and often legal value), despite a long history of photographic deception,” the authors explain. “But recent developments in deep learning (and deepfakes) have greatly increased the chances of generating fake content.”
The potential impact of these manipulations?
From individuals scamming the elderly by impersonating a family member to videos designed to sow distrust of public and government figures. Attacks that are difficult for individuals to detect (and even experts in some cases) and difficult to stop.
Other major threats include autonomous vehicles used as remote weapons. Interestingly, the group judged robotic robots (small robots that can climb through gutters to steal keys or open doors to human thieves) as one of the least dangerous threats.
The worst AI threats. So are we done for?
No, but we have to work on it. In our optimistic imagery there is some kind of button to stop all the robots and all the evil computers.
In reality, the threat is not so much in the robots themselves. It's in the way we use them to manipulate and harm each other.