The Inconvenient Truth About AI: Why Artificial Intelligence Won't Surpass Humans Anytime Soon

Many people believe that now is the time to look at today's success of artificial intelligence from a historical perspective. MIT robotics professor and co-founder of iRobot and Rethink Robotics Rodney Brooks told IEEE Spectrum what he thinks about the future of artificial intelligence. However, before talking about the future, it is necessary to recall the history of artificial intelligence. 

In the 1960s, early AI researchers often predicted, with bated breath, that human-level intelligent machines were only 10 years away. This form of AI was based on logical reasoning using symbols and was implemented on digital computers that today seem laughably slow. The same researchers considered and rejected neural networks. 

In the 1980s, the second era of artificial intelligence was based on two technologies: rule-based expert systems, a more heuristic form of symbol-based logical reasoning, and a revival of neural networks, driven by the emergence of new learning algorithms. Again, there were breathless predictions about the end of human dominance in intelligence. 

The third and current era of AI emerged in the early 2000s with new systems of symbolic reasoning based on algorithms capable of solving a class of problems called 3SAT, and with another achievement called simultaneous localization and mapping. SLAM is a method of gradually building maps as a robot moves through the world. 

In the early 2010s, this wave gained powerful momentum with the emergence of neural networks trained on massive datasets. Soon it turned into a tsunami of promises, hype and profitable applications. Regardless of what you specifically think about artificial intelligence, the reality is that virtually every successful deployment has one of two traits: there is a human somewhere "in the loop," or the cost of a system error is very low. 

In 2002, iRobot introduced the first mass-market autonomous vacuum robot, the Roomba, at a price that severely limited how much artificial intelligence it could be equipped with. However, limited AI was not a problem. In the worst failure scenarios, the Roomba simply failed to pick up a layer of dust or missed debris on the floor. 

That same year, the company began deploying the first of thousands of robots in Afghanistan and then in Iraq, used to help troops disarm improvised explosive devices. Failures there could kill someone, so there was always a human in the loop giving control commands to the AI systems on the robot. 

Today, AI systems decide on their own which ads to show us on web pages. A foolishly chosen ad is not a problem; in fact, there are many of them. Similarly, AI-powered search engines show us a list of options so we can choose what we need at a glance. On dating sites, AI systems choose whom we see, but fortunately these sites do not arrange people's marriages without their participation. 

So far, the only self-driving systems installed on production cars, whatever marketers may say, are all Level 2 systems. These systems require a human driver to keep their hands on the wheel and remain attentive, ready to immediately take control of the car if the system makes a mistake. There have already been fatal consequences when people were not paying attention. 

These are not the only terrible failures of AI systems that went unnoticed. For example, people have been wrongly arrested based on facial recognition technology that performs poorly on racial minorities, making errors that no attentive human would make. 

Sometimes we are trapped even when the consequences of failure are not dire. AI systems provide speech and language understanding for our smart speakers, as well as entertainment and navigation systems in our cars. We consumers will soon adapt our language to every such AI agent, quickly learning what they can and cannot understand, much as we do with our children and elderly parents. 

AI voice assistants are designed to provide us with as much information as they deem necessary without wearing us out, while reporting everything important that may need fixing. The question is not what your AI system can do for you, but what it has made you do for it. 

Artificial intelligence has deeply penetrated our lives. Among countless other tasks, neural networks power our virtual assistants, give online shopping recommendations, recognize people in our photos, check our bank transactions for signs of fraud, transcribe our voice messages, and filter out hateful messages on social networks. 

To take the next step, researchers are trying to combine AI and robotics to create intelligence that can make decisions and control a physical body in a messy and unpredictable real world. This is a potentially revolutionary goal that has attracted the attention of some of the most influential research organizations on the planet.