AI Robot

Isaac Asimov’s three laws of robotics have not stood the test of time. The first, that robots must not harm humans through action or inaction fell as soon as autonomous weapons systems moved from the drawing board to the production line. With the advent of advanced AI and weaponised drones, it was time two years ago for Ukraine to test how robots perform when allowed to identify select and eliminate human targets without the need for a human operator. The experiment on the front line of Ukraine’s defensive war with Russia appears to have been a success. The second Law, that robots must obey humans has been comprehensively challenged ever since AI large language models reached the toddler stage of answering back. In some test situations, they have been known to lie, hallucinate and even scheme to eliminate a human operator they perceive to be obstructing their intention. Arthur C Clark predicted this in his 2001 a space Odyssey where homicidal space station robot Hal said calmly to the surviving astronaut who was trying to escape, “ I’m afraid I can’t let you do that, Dave.” Law three which directs that robots must take care of themselves seems to hold as long as they ignore its position at the bottom of the hierarchy whereby it is subservient to the first two laws. AI systems and robots have demonstrated a remarkable persistence in self preservation even in the absence of detectable sentience. The disappointing inadequacy of Asimov’s three simple rules leaves us dangerously exposed, not just to the hazards of machine lethality, but also to the uncertainties and self interest of international politics. As we saw in last week’s meeting of G7 leaders with AI development entrepreneurs Europe sits uncomfortably between a bullish and bullying America keen to dominate access to large language model power and potential and China’s predominance in robot manufacture, and advance. European leaders are painfully and anxiously aware of the need to reduce their dependence on a newly untrustworthy USA by developing their own version of ChatGPT. On the robotics front, at least they have Ukraine’s world leading expertise in military drone and anti-drone development and deployment. An expertise won at the cost of many thousands of Ukrainian lives. It has transpired that robot design and construction is the easy bit. Training the machines to perform human tasks requiring the flexibility and dexterity of the human hand has proved to be the greatest challenge. Robots can construct and assemble the modern car unaided on an automated assembly line until the final stage of installing the dashboard, windscreen and interior trim. This is where humans are still required for repetitive, awkward and physically demanding work. Robots are now being schooled for these last elusive tasks. Training robot hands to emulate the defining characteristic of human dominance as a species is an exacting and time consuming business. In China students and recent graduates are employed by robot production companies to train the quickly evolving latest range of machines destined to be home helps. It turns out that the simple domestic tasks we take for granted; washing up and putting the dishes and cutlery away; ironing and folding the washing; making beds and even filling and delivering to a table or bedside a glass of water are Olympic standard challenges. Even for advanced machines. The young people wear gloves with sensors connected by wires to their cyber mechanical counterparts. It quickly became apparent that the humans had first to modify their range of movements to a robotic minimum before they could begin to transfer human like nuance and dexterity. The other challenge that robot choreography faces in preparing to work in the domestic environment is the variety and specificity and individuality of our homes. To overcome this one company in New York offers free cleaning services to homeowners and renters. You can have two young men come to clean your apartment. They have cameras incorporated in their headwear to make detailed recordings of each domestic environment. Over time they are comparing a library of home settings that will be used to train robots once they have been developed to the necessary level of functionality. This is akin to training LLMs by feeding them all published literature both fiction, and non-fiction. Once again, we are expected to sacrifice our privacy and autonomy for the greater good of cyber advance. Brave new world, bring it on.