Humans beware.The machines are coming.

My view on the impact of Artificial Intelligence

The Matrix was a science fiction movie from 1999. It portrayed a world where machines were at war with human beings and even winning. Keanu Reeves who was the protagonist had to be rescued from the clutches of the machines to aid the war. The sheer boldness of the concept of the ‘matrix’ combined with eye catching special effects made it very popular. I loved it. Shortly after this there was another movie titled ‘I Robot’, which showed a futuristic world peppered with intelligent robots blithely helping human beings with their daily chores and activities. Will Smith, in this movie, plays the role of the paranoid cop who doesn’t trust these robots. His hunch turns out to be correct and the movie turns into a fight for supremacy of man over the machines.

Machines have been around since the invention of the wheel to the industrial revolution and beyond. They have helped with manual tasks and in many cases replaced humans. We can see this in farms, factories, construction sites but also in day to day situations like vending machines ,parking access control and more. The development of computers helped in performing some tasks of the human mind. Computers too are ubiquitous and can be seen in offices, academic institutions, retail stores and even homes. Machines and computers have been embraced by us even if it resulted in replacing human resource.

I first heard of Artificial Intelligence (AI) while in school. The concept appeared vague and my understanding of it vaguer. It did not have imminent application. In simple terms it was the ability of a machine to think and learn. AI was still not mainstream when the popular science fiction movies that I described above were produced. Robotics too was nascent and complex functional robots had not developed. There was Asimov, a robot developed by Honda but it was hardly jaw dropping.

My first brush with Robotics was in 2009 when I witnessed the DaVinci Surgical Robot in action. It was, and is, a marvel of technology. It was able to replicate fine finger movements that are essential for performing a surgery. Most people have a misconception that a robot is humanoid machine that works like a real human. The Da Vinci is nothing like that. It’s a giant block with arms. This is controlled remotely with delicate ‘joysticks’ attached to the surgeon’s fingers. If anything, it was like a giant video game console. Machines of the past did physically intensive or repetitive manual tasks but here was a machine that was capable of replicating non repetitive fine finger movements with precision. This, to me, was a paradigm jump in machine capability.

My interest in AI was rekindled in 2016. A machine called AlphaGo was in the news. It had successfully defeated the world champion in a game called Go using AI. The game of Go was considered the ultimate challenge for computers and a machine capable of defeating humans wasn’t expected to be developed for another decade at least. Future historians will look at the successful development of AlphaGo in 2016 as a watershed year in the evolution of AI.

I learnt that Go is a Chinese board game, played with back and white coins that lays great emphasis on human intuition. This is thought to be more complicated than chess. You may recall that a machine called DeepBlue beat Garry Kasparov, the world chess champion couple of decades ago. What is different you may ask ? This victory of AlphaGo was special not just because Go is more complicated than chess. It was also special because the way the machine achieved it. DeepBlue relied heavily on computing power (which today is available in most smart phones) but AlphaGo relied on AI – the concepts of ‘neural network’ and ‘machine learning’ were used successfully.

Rather than try to explain these terms which I can’t claim to understand, I will attempt to explain how AlphaGo achieved success. It was fed with millions of different Go games or board positions. Analysing this with its ‘neural networks’ it ‘learnt’ to predict human moves. This was refined further by making AlphaGo play against different versions of itself. With each game it played it ‘learned’ more and tweaked its strategy (neural networks) and bettered itself. The evolving product played almost like humans and surprisingly also came up with game moves not performed by humans before, resulting in game wins. The final product was formidable enough to beat the world champion 4-1. The victory itself was unexpected and the margin of defeat a complete shock.

Technology constantly gets better. In 2017 we had a better version of AlphaGo called AlphaGo Zero. This was a step ahead. Unlike the original AlphaGo it were not fed with millions of games or board positions. The neural networks were provided with broad rules and they relied heavily on machine learning. In simple words the machine taught itself the game from scratch. It made mistakes and learnt by trial and error and improved with each game. In the same year as AlphaGo Zero, another machine called Libratus that played poker was developed. It was also very successful using the same approach to AI. Both machines evolved to achieve victory against world champions. What was even more noteworthy was it took the new AlphaGo Zero just 3 days to reach the standard of the previous AlphaGo machine from 2016 that beat the world champion. This rate of learning a new and complicated task can never be matched by humans. It was increasingly clear to me AI was more efficient than humans minds. I had this uneasy thought they could replace us almost completely.

Many of us have benefitted from applications that use image and facial recognition based on AI. This could be using FaceID on your iPhone or by using electronic immigration gates eliminating the need for immigration officers. AI is already taking over human jobs and the technology is still nascent. In the future the applications for AI are endless. Even in 2017 it was predicted AI could be taught varied skills like stock market trading to cybersecurity to auctions and even political negotiations. If this seems far fetched I’d like to draw your attention to Google Duplex unveiled in 2018. The world was offered a glimpse of an AI enabled personal assistant in your phone. The demo showed the smartphone making phone calls on behalf of the owner and booking a hair cut appointment by having a normal conversation with a human being who is clueless that she is talking to a machine. It felt futuristic.

The scope of AI is ever widening . Apart from the earlier suggestions of being able to perform jobs in financial sector, cyber security and even dealing with people like customer service or political negotiation, AI can also be ‘taught’ creativity! AI can be trained to compose music, produce paintings or even write literature.

The medical profession is often seen as needing complex decision making based on knowledge and experience. It isn’t easy to replace a trained and experienced doctor with machines easily. However it would be naive to think doctors are AI proof. It is suggested that radiology and pathology can be aided with AI as their work involves interpreting complex images. AI has already proven to be successful in basic image recognition. AI enabled machines can ‘learn’ to interpret these complex images and eventually produce accurate radiology and pathology reports. There is more, soon your family physician could be an AI enabled chat bot. This suggestion might be offensive to doctors who spend years training in medical school. Early steps in this direction have already commenced. A company called Babylon Health is using AI to create a doctor chat bot. The early versions may not be perfect but AI by its very nature learns and it won’t be long before we have a reliable AI enabled doctor who is not human. To prove their point Babylon health put their AI enabled chat bot through an ‘exam’ in 2018. They picked up questions used for the MRCGP ( UK general practitioner qualifying exam) which are publicly available for trainees and got their ‘AI doctor’ to answer them. The machine after just 2 years of experience as a ‘doctor’ got 81% of the questions right in its ‘first attempt’. They argued this result was better than many humans sitting the real exam that too after years of training.

Convinced that AI will take over significant functions of a physician I wondered about the future of surgeons. You will recall from earlier, we already have human controlled surgical robots. Would it be a flight of fantasy to think that it would be possible to integrate AI into them, teach them surgery to produce an AI enabled autonomous surgical robot in the distant future ? It is likely this will be achieved one day. This concept of autonomous robots is not completely uncharted territory . Boston Dynamics, a robotics company, caught everyones eye with their agile Robot dog which exhibited some autonomy. When one sees the video of this dog finding its way around, it is easy to envisage a more intelligent and autonomous robot dog powered by AI.

All this narrative about the power of AI could be disconcerting to some but make many others sceptical. Are machines that infallible ? I too had some doubts until it was pointed out that machines don’t have to be infallible, they just have to be better than humans. I will take the example of self driving cars. This technology has been around for some time. Initially most of us didn’t think it was possible but now that early versions have proved to be successful we raise questions about its safety. We completely overlook the fact human drivers are not perfect and we also cause road traffic accidents. The machine doesn’t have to error free it should just cause less accidents than humans to replace manual driving. This is not impossible. Machines won’t need a sip of water while driving, won’t change the radio station, check messages on the phone, over speed or ever be found drunk driving. It will just focus on the task at hand and eventually become safer than humans and replace them.

The impact of AI on Humanity

Machines have consistently replaced blue collar workers. With advances in robotics that trend will not only continue but its scope will widen. AI will have a similar impact on white collar jobs. I hope by now you are convinced that this is no longer fiction but an imminent reality. Many desk jobs will be done by AI enabled machines that have been taught the job. The range of jobs that will be at risk will go beyond the traditional white collar and could include so called ‘gold collar’ jobs like- financial accountants, doctors, lawyers and more! It might start with an AI assistant to a human employee to reduce man power but evolutionary trends in AI show that they will one day be capable of replacing humans. If one were to take a long term view, this will result in very few employment opportunities. As unemployment rises this will lead to ‘under employment’ i.e due to abundant labour supply it will result in low wages for unskilled work that still need human input . The sports and creative fields will perhaps be valued like they are today. This is not because a robot isn’t more athletic or AI can’t be creative, but it is likely human performance and creativity will be desired. This is much like a hand crafted item being more valuable even if products similar to that can be mass produced in a factory. Not all of us will possess the talent to be a sports person or a musician or an artist. What if you aren’t among the lucky few to have a job in a world dominated by AI machines and robots? Professional video gaming maybe the popular job of the future!

I have often pondered how to tackle the problem of significant unemployment in an AI enabled era combined with an ever rising population. States will be compelled to come up with solutions. The arrows in the quiver could include Universal Basic Income, Population control (a forced one child policy and maybe even some incentives for having no children), making euthanasia legal and less regulated, restricting heath care for deserving candidates and no resuscitations beyond a certain age by law. Some of these are controversial but I am sure we cannot escape debating them in the future.

The other question I have considered is whether humans will be comfortable taking instructions from a machine ? I suppose most of us cannot entertain the thought of being told what to do by a machine. The reality I think might be entirely different. I will start with a ‘dumb’ machine like a speed camera/gun. If it says you were over speeding, chances are that you will accept the machines recording rather than contest it. Also you do not contest the math results of a calculator. It is not just about contesting a machine it is also about trust. If you wanted directions, you are more likely to trust Google maps over a human who points you in a different direction. We also allow machines to make decisions for us, like the electronic immigration gates which check our identity before letting us pass through. It appears we are at ease having machines control us. Soon we could see robots in law enforcement. An early glimpse of this was seen in May 2020 in Singapore, where a robot dog was used to monitor a park and advise people to maintain social distancing in view of the pandemic. In future this robot dog could be fully autonomous. The changes will be so gradual that one day we will be taking instructions from a fully autonomous bot with no human oversight.

The capability of AI enabled machines will keep improving. There is already research into AI enabled autonomous weapons and drones. The result of this will be a machine that will decide which human is dangerous and who is not. It will also make a decision to kill or not. This is controversial and worrying. A future world where most of the work is done by machines, where rules are enforced by machines and where warfare is performed by machines with no human micro management is not an unthinkable scenario.

If our laptop malfunctions we use the oldest trick in the book, we switch off and reboot. If the same thing happens to an AI enabled machine I suppose we can turn the power off and regain control. What if these intelligent machines learn to power themselves back on or learn to ‘disobey’ a shut down command ? What if a machine trained to enforce discipline or even kill refuses to turn off? What if they hack power grids to remain powered constantly ? What if they are all networked and act as a swarm ? Could we find ourselves in a battle to control these machines ? The movies from two decades ago don’t seem like fiction any more. By successfully developing AI, humanity has set the ball rolling to create its nemesis.