The ability of artificial intelligent cognition versus that of the human counterpart.

Abstract:

In this essay we will be discussing the idea of the ability of artificial intelligent cognition versus that of the human counterpart. We will look at the possible outcomes for replacing A.I for humans in the workplace and everyday life and if this is the correct direction to be moving in. The general rhetoric is mixed on the advancements of A.I and what type of A.I is employed. We will explore this below in more detail.

Introduction:

Cognition is the ‘mental action or process of acquiring knowledge and understanding through thought, experience, and the senses’ (Oxford Dictionary,2019). Human cognition has aided us in discovering fire, building pyramids and traveling to the moon whilst pushing humanity towards its most prosperous and most peaceful time to date. However, can, and should, human cognition be replaced by machine counter parts to do things that humans may not be doing to maximum efficiency. In the 1970’s, psychologists Daniel Khaneman and Amos Tversky ‘discovered cognitive biases, showing that humans systematically make choices that defy clear logic’ (Goldhill, 2017). If humans have been shown to defy clear logic, should we be allowed to engage in tasks that require clear logic as well as no lapses in concentration and attention. Some would argue that ‘we live in a world of deep uncertainty, in which neat logic simply isn’t a good guide’ (Goldhill, 2017). Maybe this is true for some aspects of our lives, but others may require exact precision in decision making, void of the cognitive biases that plague human beings. Air traffic control and driving vehicles are two tasks that are undertaken in a huge quantity every single day by humans, across the globe, which have dire consequences if performed inadequately and are two fields in particular that could greatly benefit from removing human cognition from entirely. Firstly, we need to understand why human cognition is flawed in the first place and why we should even consider replacing it using machines and intelligent systems at all. It is important to make the distinction between a weak A.I system and a strong A.I system. In this essay we will be looking at weak A.I’s potential in replacing humans in narrow form tasks such as driving and air traffic control management. A discussion about Strong A.I, or general artificial intelligence, and its possible societal upheaval is all theoretical at best given the state of the technology. Weak A.I, however, is very prevalent and can cause huge societal change in the short to medium term.

Human Irrationality:

From the 1970s through until the 1990's, Israeli psychologists Daniel Khaneman and Amos Tverksy published a few critical papers which helped change the idea of human rationality in the fields of psychology and, more importantly, in economics. Previously, the field of economics and the real world decisions based on this idea of economics assumed that the average man was a rational, unbiased decision making machine. However, Khaneman and Tversky's research showed otherwise (Lewis, 2016). In his 2011 best selling book ‘Thinking, Fast and Slow’ Khaneman describes the human decision making process in a two part, machine like, system. He says that we run on what he calls system 1 and system 2 styles of thought. System 1 is the quick thinking, instinctive, emotional part of our brain. We make snap decisions based on experience and biases. System 2, however, is the slower decision making process which thinks more logically and more deliberative about a problem. Our thinking, as outlined by Khaneman is his book, falls short to a huge amount of biases and heuristics such as anchoring, framing and his very famous idea of prospect theory (Khaneman, 2012). One such problem we commit as humans is the conjunction fallacy. We often hear or see information and make quick judgements on it when making a decision using system 1. However, we should stop to really consider the information and the probability associated with it by employing our system 2. Take the Linda experiment.

Linda is single, outspoken, and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in anti-nuclear demonstrations. Which is more probable?

1. Linda is a bank teller.

2. Linda is a bank teller and is active in the feminist movement.

More than eighty percent of participants in one such experiment chose option two. We chose the second option because of what Khaneman and Tversky call a representative heuristic. ‘Even though logically, we should not pick option 2, we consider option 2 more likely to be correlated with what Linda did in college’ (Brogaard, 2016). Not only do most of us not even stop to consider the question and the possibilities of it in its entirety but we also make a complete miss judgement in the probability of the outcomes. This mishap of probability calculation is highlighted in Kahneman and Tversky's prospect theory thought experiments also. Cognitive bias’ get worrying when they are applied in situations where an accident can be fatal. According to a study by Michael Roy of Elizabethtown College and Michael Liersch of New York University, most drivers would rate themselves as above average and that often traits like ‘their ability to text message while driving was what made them a unique and superior driver’ (Roy and Liersch, 2013). This kind of thinking is exactly what can cause people to believe they can make that one phone call or text or check their social media while driving. This cognitive bias also leads people to speeding, making reckless mistakes, or driving while not in the best condition to do so. Another important industry where these types of cognitive lapses can be fatal is in the air traffic control sector. ‘It has been estimated that between 60 and 90 percent of major incidents in complex systems such as aviation are caused by human error’. Air traffic control requires ‘a large number of cognitive skills by controllers including; perception, attention, memory, information processing, decision making and attention’ (de Reuck, 2014) All of these important factors are ones which humans laps and slip on all the time, according to Khaneman and Tverksys work. Humans, being humans, also need to eat, sleep and recuperate, leaving them susceptible to making vital mistakes due to lack of any of these inputs. This leaves us to wonder why we should be able to complete these tasks at all. Well, we are the best possible technological solution to perform these tasks given the alternatives at the moment. However, this is about to change.

The argument for A.I replacing human cognition:

As discussed above, we can see that humans do not always operate in the most logical, efficient way. This is okay when we choose what coffee or smartphone to buy, however, our biases and irrationality coupled with our ability to tire, get hunger and loss concentration adds up to create a machine, the human mind, that isn't the best equipped to deal with tasks like driving and directing air traffic to one hundred percent ability one hundred percent of the time. The possibility of having a car drive completely and fully autonomously on our busy streets, thanks to ever improving technology, could be a reality as soon as mid 2020 according to Tesla owner and co-founder Elon Musk. Even if Elon is out on his prediction, which he usually is, the probability of self driving vehicles is only a matter of a few short years away. This is an interesting prospect as it may just answer the questions about the ability of A.I in replacing humans in the literal and proverbial driver's seat. What is important to note here is that self-driving A.I is considered a weak A.I. The system is trained, via programming, to make decisions about what to do on the road. It is a weak A.I focused on completing one narrow task. This concept is critical in our discussion as it is often the case, as history shows, that computer systems beat humans in narrow focused tasks. For example, we would never allow ourselves to do a companies end of year accounts which have a large data set of numbers and huge amounts of computations to do. We instead aid a system like Sage to do all the mathematics for us. Driving a vehicle will be no different. We will likely input a destination, or multiple destinations, and allow the computer to complete the narrow task while we sit back and use our time to think and create. Strong A.I, on the other hand, is something which tries more so to replicate the abilities of humans. This includes reasoning, problem solving, judgement making and communicating. The implications of a system like this are quite far off according to many estimates where as weak A.I’s potential is here and is happening now.

Approximately there are 4 car deaths per hour in the United States alone. ‘For some, their breath is taken away at the notion that there are approximately 37,000 human beings killed as a result of car accidents each year’. Statistics and research suggest that just by the fact that A.I powered self driving technology won’t get drunk or speed over the speed limit, we could reduce road deaths by around two thirds (Eliot, 2019). This is both worrying in the sense that human error and cognitive biases can cause so many roadside deaths a year but also promising at the possibility that technology could relieve us of many unnecessary transportation deaths. Air traffic control is a different story. Due to the nature of the industry and safety regulations ‘Air traffic control (ATC) system errors rarely occur in today's system. ATC-related aircraft accidents are even rarer events’ (Danaher,1980). That being said, more than ninety percent of ATC errors are caused by humans when they do occur. This still has huge implications due to the large fatality possibility with aircrafts. Air traffic control company Nats are moving in the direction of alleviating the human error from air traffic control by using neural networks to analyse data and create an understanding of what normal operations look like for an A.I system. ‘our focus is on using AI and Machine Learning to support controller decision making. This might be by using it to simultaneously monitor multiple areas of interest across an airport, like runway exit points for example – something that humans aren’t capable of physically doing’ (Taylor, 2019). Although ATC is, for the moment, a joint task between humans and computers, we hope that soon we can remove humans and the errors they cause from the equation. This will align nicely with the direction planes are taking as well where much of the piloting of the aircraft are done by intelligent weak A.I systems. Moving in the direction of replacing human intelligence with that of artificial intelligence, for me at least, seems clear cut. The ability to process huge amounts of data at fast speeds gives A.I the edge over human cognition when performing weak A.I, narrow focused tasks.

Problems with A.I :

One of the main concerns with A.I stepping in to replace humans in the kind of tasks we have mentioned is that of humans themselves. Self-driving technology will, in time, work almost flawlessly with enough data and training for the systems to learn from. However, if we allow humans to stay in the equation they could be the ones to provide confusion to the A.I systems with our unpredictability and irrationality. Roads where some cars are driverless and others have human drivers could be problematic as human error could create unexpected problems with A.I systems as well as continue to cause crashes due to basic human inefficiencies such as lack of concentration, drunkenness etc. An autonomous driving system would work at its best when every vehicle is on the system and work in conjunction with each other, all making rational, data driven decisions. Implementing a system like this however will be time consuming and a challenge. ‘While self-driving AI has made progress by leaps and bounds in the past few years, a large part of the problem is that peak performance requires all cars on the road to be connected to each other, exchanging data’ (Rue, 2018).

Another issue which could prove major issues is that not having an actual driver requires technology and technology is vulnerable to hacking. ‘Hackers have already remotely hacked internet-enabled cars, starting engines and even driving them around. It’s not out of the realm of possibility that a hacker could gain access to a remote taxi service, locking the customer in the car’ (Rue, 2018). This is seriously worrying as the possibilities associated with this could prove major security as well as health and safety issues on our roads. The idea that our vehicles could be hacked into will also be a major deterrent to users considering adopting the technology. The same applies here with air traffic control and is in fact amplified as air traffic control hacking could have huge national security issues. Weak A.I systems like those powering self-driving technology also work on a heuristic programming model. Heuristics are mental shortcuts that help us make decisions and judgments quickly without having to spend a lot of time researching and analyzing information. Computer programs often do the same thing when processing a task which is necessary as there often isn’t a set algorithm, for example, for “driving a car”. Heuristic programs do the job more times than not, often rarely, if ever, making a mistake. However, this is not guaranteed and an A.I system is bound to make a mistake. When this happens, where then does the blame lie. The answer to this is difficult and is a seperate essay entirely.

Conclusion:

Systems, invented by humans, have been replacing humans since the beginning of time. The wheelbarrow alleviated us from lifting multiple objects as it was more superior at this job than we were. Printing presses could produce books more accurately and more efficiently than we could and therefore replaced us. Artificial intelligence has begun to replace the store cashier in your local supermarket and soon it will replace us behind the wheel, in front of air traffic control management systems and in nearly all processes that can be carried out using weak A.I systems. Robotics and artificial intelligence is just better than we are at doing these tasks and that isn’t necessarily a bad thing. It saves us from our own egos, cognitive biases and unnecessary mistakes every day as well as making both our professional and personal lives easier. The story thus far is one of progress. Replacing humans within agriculture lead us to the cities to create the mass production of goods such as books and automobiles. Replacing humans again in the manufacturing of goods lead to the invention of the services industry. When A.I and robotics inevitably drives our taxis and answers our customer care calls we will no doubt innovate and create which ultimately has always pushed humanity forward. ‘Since the dawn of the industrial age, a recurrent fear has been that technological change will spawn mass unemployment. Neoclassical economists predicted that this would not happen, because people would find other jobs, albeit possibly after a long period of painful adjustment. By and large, that prediction has proven to be correct’ (VARDI, 2017).

I have no doubt that replacing flawed human cognitive powers with those of weak A.I systems is not only the best thing to do but also the right thing to do. Thousands of lives could be saved by removing human drivers off the road alone without considering the possible impact A.I could have across the board in other industries. These systems, created by us, can free humans to do what we do best. Think and create.

References:

1. Brogaard, B. (2016). Linda The Bank Teller Case Revisited. [online] Psychology Today. Available at: https://www.psychologytoday.com/us/blog/the-superhuman-mind/201611/linda-the-bank-teller-case-revisited [Accessed 17 Oct. 2019].

2. Danaher, J. W. (1980) ‘Human Error in ATC System Operations’, Human Factors, 22(5), pp. 535–545. doi: 10.1177/001872088002200503.

3. de Reuck, S. (2014). Factors Underlying Human Errors in Air Traffic Control. University of the Witwatersrand, Johannesburg, [online] p.17. Available at: http://wiredspace.wits.ac.za/bitstream/handle/10539/16984/de%20Reuck%20312568%20Research%2020141003.pdf?sequence=2&isAllowed=y [Accessed 20 Oct. 2019].

4. Eliot, L. (2019). Essential Stats For Justifying And Comparing Self-Driving Cars To Humans At The Wheel. [online] Forbes.com. Available at: https://www.forbes.com/sites/lanceeliot/2019/05/30/essential-stats-for-justifying-and-comparing-self-driving-cars-to-humans-at-the-wheel/#4e80e7a846ed [Accessed 22 Oct. 2019].

5. Goldhill, O. (2017). Humans are born irrational, and that has made us better decision-makers. [online] Quartz. Available at: https://qz.com/922924/humans-werent-designed-to-be-rational-and-we-are-better-thinkers-for-it/ [Accessed 15 Oct. 2019].

6. Khaneman, D. (2012). Thinking, Fast and Slow. London: Penguin.

7. Lewis, M. (2016). The Undoing Project. [Place of publication not identified]: W.W. Norton & Company.

8. Roy, M. and Liersch, M. (2013). I am a better driver than you think: examining self-enhancement for driving ability. Journal of Applied Social Psychology, 43(8).

9. Rue, N. (2018). AI, Self-Driving Cars, and Lyft/Uber: Helpful or Harmful?. [online] Medium. Available at: https://becominghuman.ai/ai-self-driving-cars-and-lyft-uber-helpful-or-harmful-a85b90c81ac0 [Accessed 24 Oct. 2019].

10. Taylor, A. (2019). Embracing Artificial Intelligence in aviation and air traffic management - NATS Blog. [online] NATS Blog. Available at: https://nats.aero/blog/2019/06/embracing-artificial-intelligence-in-aviation-and-air-traffic-management/ [Accessed 22 Oct. 2019].

11. VARDI, M. (2017). No Need To Worry About Robots Replacing Human Workers? Look At History. [online] Fast Company. Available at: https://www.fastcompany.com/40462489/robots-automation-labor-look-at-history-industrial-revolution-moshe-vardi [Accessed 24 Oct. 2019]