In dialog with smart machines
Even robots are becoming increasingly cheeky. When, at an investors’ conference about artificial intelligence (AI), the elegant robotic lady Sophia was asked if she knew that she was a robot, she pertly responded to the moderator’s question: “Why don’t you tell me how you know that you’re a human!” And when the small robotic boy iCub lost against a human opponent in a video game, he angrily uttered: “Someday we robots are going to take over and then you’ll have to pay for it.”
More than likely, some observers found themselves choking on their laughter, especially since prominent critics – such as Tesla CEO Elon Musk and the late astrophysicist Stephen Hawking – have frequently expressed warnings of a superintelligence. But those of us who, in the light of this, fear that machines will soon start ruling the world may find some comfort in comments by Ben Goertzel, Chief Scientist of Hanson Robotics where Sophia was developed. The robot lady, he says, is merely “a chatbot with a human face.” She can use gestures and facial expressions and answer simple questions, which includes her accessing data on the internet. Everything else is pre-scripted dialogs like those in docusoaps on TV.
Lifelong learning is for robots, too
The same applies to the iCub although its emotional states are hardly predictable because they depend on both sensor signals and the respective situation. This robot is growing up in a research center near Genoa. It can walk, grab things, talk and listen, sense when it’s being touched, blush and produce charming laughter – and above all: it learns like a child by imitating humans. Several instructors are teaching it to use its toys, clear the table, play piano or hit a target with a bow and arrow. Developmental Robotics is the name of this young AI research field in which robots in the course of their “life” continuously acquire more skills and knowledge.
The lesson here is that by combining the abilities of the human and machine, it created a partnership that had super-human performance. And that is humanistic AI
AI pioneer Tom Gruber with reference to the use of AI in medical diagnostics
Even so, they’re far from truly understanding the world. No robot and no chatbot has so far had any chance of passing the Turing test. In 1950 – nearly 70 years ago – the mathematician Alan Turing kicked off the AI age. “Can machines think?” he asked and proposed a test: if a human has a conversation with a machine he or she can’t see and is subsequently unable to reliably tell whether the conversation was with another human or a machine, the machine has passed the test.
In 1990, the sociologist Hugh Gene Loebner pledged a monetary prize and a gold medal for the first computer program that would successfully complete the Turing test in a 25-minute exchange. To date, the 100,000 dollars of prize money have not had to be paid. Although programs are becoming better and better, they sometimes awkwardly evade questions, make multiple use of text modules or can’t come up with sensible answers to everyday questions. Currently in the top spot on the list of the digital entrants in the annual Loebner Prize competition that have at least won bronze medals is the chatbot lady Mitsuku.
Schaeffler goes AI
Smartphones and social media have changed communication around the world. Now, the Internet of Things will interlink vehicles, machines and people. Schaeffler has embarked on the journey of digital transformation.
More progress in six years than in six decades before
However, all this does not mean that voice-controlled AI assistants are useless – the opposite is true. In May 2018, a call to a hair salon caused a sensation. The caller was a machine that precisely responded to what the person on the other end was saying and asking. It would use half-sentences and pause with mm-hmm – providing an impression as if it were thinking about what was being said. Now this software called Google Duplex is being tested in everyday life by smartphone users in the United States. For clearly defined tasks, such as making a reservation in a restaurant, this already works pretty well. Other AI assistants for tedious routine tasks are about to be launched. In 2019, for instance, “Duplex for the Web” was presented: here the assistant takes care of filling out forms typically used on the internet.
The areas in which AI is currently burgeoning are clear: they involve recognition of speech, images and contents of texts, and the analysis of huge data volumes. More progress has been made here in the last six years than in the sixty years before due to three mutually reinforcing trends. First, higher performance of hardware: any smartphone today is able to match the world’s biggest supercomputer in the mid-1990s in terms of processing speed and data storage capacity. Second, better software: in artificial neural networks – just like in the human brain – many layers of artificial nerve cells (neurons) are interlinked for the purpose of processing data. The strengths of their connections may vary, which enables the networks to learn. The principle has been known for a long time, but compared to the networks of the 90s, the ones today are millions of times larger and stacked in much deeper layers – which is why they’re called deep-learning networks. The third factor is the internet with billions of texts, images, audio and video files that can be used to train AI systems. Their learning expands with every search query, every voice input and every translation request.
Searching 200 million pages of text within seconds
In many areas, we’ve already become accustomed to using AI: facial recognition to unlock our smartphones, voice assistants like Siri and Alexa, automatic text corrections, personalized advertising, translation programs or health apps. The first fully automated vehicles are traveling on freeways and robots are working hand in hand with humans in factories. AI analyzes data from trains or turbines for predictive maintenance, computers provide advice to doctors in hospitals and financial experts in banks, and every day, new applications are added.
Increasingly often, machines defeat human champions thanks to AI. In 2011, the IBM system Watson won against the world champions in the Jeopardy quiz show: within a matter of seconds, it combined 200 million pointers to text pages into appropriate solutions. This feat was followed by others in rapid succession: Various types of AI software made only half as many mistakes in recognizing traffic signs as humans, identified all house numbers in millions of photos within 100 minutes and diagnosed cases of pneumonia and types of skin cancer with greater precision than experienced doctors. Most recently, the robot lady Xiaoyi in China passed the country’s medical licensing exam because, better than many students, she knew which symptoms were indicative of which diseases and which therapies made sense.
In 2016, AlphaGo software won against the world’s best Go players. A year later, AlphaGo Zero beat the previous program 100-0. It had simply learned by playing against itself millions of times. New AI systems call bluffs in poker and have mastered StarCraft II – which is very difficult because, in this game, many players are acting at the same time and no-one has a complete picture of the whole playing field. Even so, the truth is that AI machines are specialists. A computer that wins in chess or Go has only mastered the art of playing these games. It cannot drive a car or mow the lawn. In addition, it’s a lot more difficult for machines to move the pieces in chess than to win the game. For instance, it took research scientists ten years to teach robots how to open the door to a house with a key.
New jobs due to AI
What’s more, machines have no “common sense.” Once a computer has learned how to recognize cats it’ll find cats everywhere, even in clouds or in picture noise on screens. Although machines are able to read emotions from faces and voices and act as if they had feelings, these are merely simulations.
In the future, we’ll no doubt be using smart machines like we’re using smartphones today: self-driving cars, as well as digital assistants and robots in offices, factories and at home. We need smart grids for the energy transition and smart cities to make our cities cleaner, safer and more livable. AI will change all professions and occupations – from farmers scouting their fields by means of drones, to truck drivers switching to autopilot mode, to surgeons operating with the help of robots.
Even so, when doctors tell computers to scan thousands of pictures for the presence of tumors they’ll be using the machine as an assistant, but that doesn’t mean the doctor is no longer needed. There’s no reason to fear mass unemployment because there will also be many new jobs, for instance in training machines, in AI forensics or in neural art design. In education and continuing training, teaching the skills that no machine possesses will be of paramount importance: flexibility and ingenuity, unconventional, holistic thinking and social skills like managing, motovating or solving conflicts – plus ensuring that AI works safely and reliably. Protecting people and infrastructures must be top priorities. Machines are only allowed to provide advice, whereas important decisions – whether in hospitals, in courts of law or in granting bank loans – will have to be made by humans in the future, too.