Research Paper ‐ AI and its Effect on the Job Market and Social Economy - ConnorEast/Tech-Journal GitHub Wiki

- The Community College of Vermont -

AI and its Effect on the Job Market and Social Economy

In a world where AI has the capacity to overtake all aspects of our life can we know reality from fiction?- Connor East

Connor East 11/22/2023

Seminar In Educational Inquiry - (V23FA HUM-2010-VO02)

Table Of Contents

Chapter 1: What is artificial intelligence? ---------- 2/5

What is Artificial Intelligence? ----------------------- 2

Stages of AI ------------------------------------------- 3

Modern Day AI In the Workplace ------------------------- 3

AI in Social media --------------------------------------- 5

Chapter 2: Overarching Views ------------------------- 5/10

The Negative Effects of modernizing the workplace ------ 6

AI and its capacity for sentience ---------------------- 7

Chapter 3: AI Limitations and Biases ----------------- 7/11

Chapter 4: AI’s Effect on Culture and Society -------- 11/13

Social Media and AI Application ------------------------ 11

Cyber Warfare -------------------------------------------11/12

Chapter 5: Environmental Disaster -------------------- 12/13

Electrical and Environmental Costs of ASI -------------- 12/13

Potential for nuclear disaster ------------------------- 13

Chapter 6: Final Thoughts ---------------------------- 13/14

** # Chapter 1: What is Artificial Intelligence? **

Motivation, Perception, Law, Ethics. When you hear those words what do you think about? Do you think about how one can motivate their children to do better in school? Do you think about your perceived ethical compass? Maybe you are thinking about unjust laws and ethical practices. Now I will ask you a different question. What do those four words mean to you in regard to AI? Some will give AI positive attributes while others may postulate and state the negatives. My opinion on this topic is that AI has the ability for, but not the capacity for, all four of those concepts. Ability, in this context, is the physical power necessary to complete a task whereas capacity refers to situational intelligence and the capacity for un-nurtured growth. However, by applying AI to new fields and industries, we are intrinsically damaging the current-day job market as well as the social economy.

To start, we must define what Artificial Intelligence is. Artificial Intelligence, commonly referred to as AI, is typically defined as the ability of a system or algorithm to perform tasks that typically require the capabilities of human intelligence. When most people think of AI, they typically think of a device with superintelligence, a form of AI that surpasses human intelligence, rather than the many common tools we are exposed to on a daily basis. The devices in our pockets, be them phones, watches, or tablets, are all connected to thousands of algorithms that perform tasks for us that even fifty years ago we would need to do for ourselves. Some examples of this could be anything from ordering food, getting directions to a location, or even simply browsing social media (Mehan).

Stages Of AI:

In total, there are three types of AI’s. Those three types are known as artificial narrow intelligence (ANI), artificial general intelligence (AGI), and finally artificial super intelligence (ASI). ANI is short form consisting of only a small set of actions it can potentially take. You can think of it as analogous to a light switch. One action to turn on while the other turns it off. AGI has many more options as it is stated to be on par with human capabilities. Finally, we have ASI, ASI is the only form of AI we have yet to reach given that ASI has the capabilities to surpass human intellect and processing power (Mehan). However, as time progresses we get closer to the evolution of AGI into ASI. An example of evolution would be none other than BabyX, an AI capable of creating neural pathways similar to those found in humans (Originals, 5:09). The main difference between BabyX and other forms of AI comes down to its capacity to see and its visible rendering. BabyX takes the form of a small child who uses their owner's monitor camera as a set of eyes. From there, it guesses the item based on previously seen content. The answer BabyX gives is then confirmed or denied by the designer which may cause synthesized-emotional reactions. BabyX and all other AGI work using forms of machine learning. Machine learning is simply an algorithm that uses data to feed it in order to imitate human behavior. More specifically AI is typically trained using supervised learning, a form of training that requires a proctor to feed the machine data and to have them confirm the AI’s assumption of what said data is. The main issue that occurs as a result of this form of training can be that of bias. Due to the static nature of the information being fed to the AI, it will not be able to understand nuances. This will most likely lead to the AI having an unintentional and or intentional bias. We will discuss later on why said biases can be harmful, for now just know that bias is typically due to negligence more so than maliciousness on the part of the creator. (Mehan) Modern-Day AI In the Workplace:

Up to this point, workplaces only host a few select AI devices. This is due to the fact that we are still only at the beginning of AI development and change takes a significant amount of time. One major piece of AI intelligence in the workplace would be self-checkout machines which are a form of self-Service Technology (SST). The use of self-checkout has been linked to a few categories of usage. These categories consist of the following: Speed, ease of use, Reliability, Enjoyment, Control, Attitude, Hedonism, Responsiveness, Demographics, and Consumer Traits. From a corporate standpoint, the implementation of SST is a net benefit because it allows the company to reduce overhead costs while also using customers as a resource (Fernandes and Pedroso). Reducing overhead is majorly important for corporations due to rising employee costs, real estate costs, and the addition of e-commerce into the market. (Grand View Research).

The main issue with SST tech is shown more so when you look at age demographics and recent events such as COVID-19. Elderly individuals have a tendency to use normal checkouts more often than self-checkouts due to a mixture of lack of self-perceived control (Fernandes, The effect of self Checkout), as well as loneliness. As a result of social distancing as well as the lack of individuals one has around them as they age, a large majority of elderly individuals are lonely (NIH, Understanding Loneliness and…). As such, the elderly have a higher need for social interaction and choose to go to non-self-checkout stations as a result. The younger generation also has issues with loneliness however we will discuss some of those issues later on.

Another example of AI in the workplace would be those of ceiling-based cameras. A lot of companies have begun to use cameras in place of security guards to catch shoplifters. The main issue is that those who steal less than 1,200 dollars worth of product can’t be federally prosecuted and no one is there to stop them from walking out the door. Overall it's a sad state of affairs seeing as customers and associates pay the price (Grand View Research).

For those who have an issue understanding the point of the previous paragraphs; let me give you a brief explanation. It is unlikely that human-based checkouts will be removed due to social isolation and loneliness in the eldest generations. This means AI won't completely destroy said checkout jobs. The next point is the potential positive of AI-based Surveillance in stores. By having AI-integrated camera systems we will be better equipped to handle theft. The second part is simply my perspective on theft.

AI in Social Media Applications: As of 2023, there are a multitude of different social media platforms. However in this case we will be looking at the four Major social media platforms: Instagram, Facebook, TikTok, and YouTube (Dixon). All four platforms allow individuals to post self-made content which is then scored on its engagement. The major error with this sort of system however is its overemphasis on ragebait content as well as unrealistic beauty/lifestyle content. This means the demographic for these programs, those 18-24 (Lin), is consistently given to those still within their development cycle. Overall this leads to a lowered sense of self-worth, depression, and anxiety (NIH, the Teen Brain). This point however is not the main focus of this paper and as such I will not discuss it too thoroughly throughout this paper; however, if you wish to read more look in the citations section for (East, The Effects of Social Media)

Chapter 2: Overarching View Points?

Social-Cultural Perspective on AI:

This sector of information is heavily biased in regard to the specific sector of the topic we are discussing. For instance, a large majority of individuals have skeptical and uncertain perspectives on whether AI should be used in regards to choosing resumes to show to the hiring manager (Rainie, Americans’ views on…). One reason why people worry about AI in this way is due to ingrained bias. An example of this would be how most facial recognition technologies are racially discriminatory. In a study called “Gender Shades” it was found that AI is 34% less accurate on African-American individuals when compared to their Caucasian counterparts. (Najibi). Just taking this one study into account it makes sense to think that there is a possibility that similar issues occur within the resume-checking AI that may cause highly skilled individuals to lose the opportunity for advancement. Instances of this result from males having larger resumes when compared to their female counterparts due to previous job inequalities; or it may be based on the name of the participant. For instance, someone named Jemalle may not get the same recognition as someone named Carter when put through AI. The AI may not have an image to go off of in a resume however it does have names, previous occupations, etc; which could be misused by the AI. Other content that affects our perspective on AI is the media we watch and given AI is almost always an enemy in Sci-fi shows and movies, it makes sense people would be wary.

The Negative Effects of modernizing the workplace:

This sector of information is heavily biased due to the fact this section is mostly focused on the future modernization of AI in the workplace and as such, is simply speculation based on current trends. According to a paper by Absuelidze, it is likely that 48% of both blue and white-collar jobs will be automated. One reason this is likely to occur is due to the fact that automation delivers better customer service 27% of the time when compared to its human counterpart. In total 54% of jobs are at risk and it is believed that 9% of new US jobs will have been created by AI. This will only further be heightened by the fact that by 2025 AI has been estimated to be a $118.6 billion dollar industry (Abuselidze). The main issue with this relates to a theory by John Maynard Keynes called the “Technical Unemployment Theory”. This theory postulates that the induction of AI into the workforce will lead to low-skilled laborers not being able to attain jobs. This will further divide the economic class into the categories of the Proletariat and the Bourgeoisie. Overall, Economic growth will springboard when AI takes over those jobs; however, a large number of unskilled laborers will be out of a job and the government likely won’t send them to college for free (Abuselidze). For reference, the terminology low-skill and or unskilled refers to entry-level jobs and or blue-collar jobs.

AI and its Capacity for Sentience

The question of whether AI will ever have the capabilities for sentience has been a hot debate topic since the evolution of AI crept into the public eye in the early 1980s. The main issue with assessing consciousness is that there is no one set expression of it. In other words, in order to assess consciousness one must use a rubric that allows for flexibility. The main thing of note however is that AI does not currently have the capabilities for consciousness however, it is possible for an AI to achieve consciousness. In fact, it may be possible to create a sentient AI with our current level of technology depending on whether or not we currently have the hardware capabilities (Butlin, Consciousness in Artificial Intelligence). Proper teachings on the do least harm principle could have major effects on the AI psyche. Here is a theoretical example: let's say you tell an AI “All humans will be dead by the end of the century if nothing is done about climate change. Given what you know, how should we go about fixing the issue?” The AI could respond in a number of ways. However, if the least harm principle was misunderstood it is highly probable the AI would say killing off a large majority of the population will curb climate change.

** # Chapter 3: AI Limitations and Biases**

In this paper we have already discussed some biases however here is an overview of some of the problems with biased AI.

Race: As we have previously discussed, AI has difficulties understanding darker skin pigmentation. This is either due to mistrained data or it is the result of darker pigmentation making it more difficult to find defining features. (Najibe) Gender: AIs make out-of-date comparisons between gender and the work said gender does. An example would be assuming women are nurses while men are doctors. Another example would be the fear of AI using biased data for accepting specific applicants due to gender rather than company viability. (International Women's Day) Police use: Law enforcement has begun using AI for predictive policing. This is bad because of AI's limitations in regard to race. (Buolamwini)

Some of the Limitations include:

Electrical cost of Machine Learning: One thing most people don’t realize about AI is the quantity of electricity needed to generate answers based on the program's parameters as well as the quantity of data learned from it. Here is a shocking statistic, GPT-3 an AI with 175 billion parameters consumed 1,287 megawatts per hour which equivalently releases 522 tons of carbon emissions into the atmosphere. In total, that would be the equivalent of 123 gas-powered vehicles being used for an entire year. (Jason)

Device Hardware: The main sectors of hardware that need reinforcement are the CPU, GPU, Memory, Network, and Storage IOPS. Below is an image from TechTarget.com showing different companies' system parameters in regard to AI integration. (Marko, Infrastructure for machine learning)

Given the length of the paper, as well as the audience's ability to focus on the content would be negatively impacted if I discussed each sector of hardware; we will instead look at the systems themselves. CISCO UCS: This device uses 4, 7 inch chassis accompanied by the NVIDIA Tesla V100 which is considered to be the most advanced data center GPU ever built (NVIDIA). This type of system relies upon redundancy to improve system survivability and decrease the likelihood of full-scale system failure. Another benefit of this system would be how its integrated management processor allows for the system admin to monitor not only the servers' health but also its inventory and event logs. In terms of environmental requirements for this system, it must be kept in an area that is maintained between 50 and 70 degrees Fahrenheit when in operation. If the system is non-operational the temperature must be between -40 and 158 degrees Fahrenheit. Humidity and altitude also need to be monitored however that is not a major concern for this paper. In total, the maximum energy supply available to this system is 1600 watts however it is difficult to actively measure the exact wattage one system would use without specific parameters such as what programs they are running.

Dell DSS: This device uses 4, 7 inch chassis and supports two separate Intel Xeon Scalable processors. Said processors may have up to 24 cores. These processors improve device performance, 10x higher performance in AI deep learning-based training, Packet processing is also said to be quite improved as well (Intel). It supports a wide range of GPU specifications as well as has the capacity for a maximum RAM of 768 gigabytes. The coldest this system can stay in is 50 degrees Fahrenheit to its maximum of 95 degrees Fahrenheit. When not in operation its range widens from -40 degrees Fahrenheit to 149 degrees Fahrenheit (Dell).

NVIDIA DGX-1: This device uses 3, 7 inch cassis and supports Tensor Core Technology. Using the DGX-1 with Tesla V1100 is 96X Faster in regards to AI dataset training when compared to a CPU-only server setup. The maximum power draw of this device appears to be 1,500 Watts; however, that would depend upon what the hardware is being used for at any given moment (DGX-1).

NVIDIA DGX-A100: This device uses a 6U chassis which equates to 10.5 inches and supports 2 terabytes per second memory. This system is one of if not the most powerful end-to-end AI/High-performance-computing application datacenter. Its maximum electrical draw power is capped at 6.5 Kilowatts or 65000 Watts (Nvidia).

_SuperMicro 1029: _This device is the smallest on our list, only using a 1U chassis. It contains up to 28 cores and has a total output power of between 700 and 750 watts. When in operation it must be kept between 50 degrees and 95 degrees Fahrenheit. When non-operational the range increases to between -40 and 140 degrees Fahrenheit. It uses Intel technology as well (SuperMicro).

Chapter 4: AI’s Effect on Culture and Society

Social Media and AI Application:

In order to discuss social media we should first reiterate the point that the population affected by social media the most are those who fall between the ages of 18-24 (Oberlo) or younger given that individuals may lie about their age when signing up for social media. Given the fact that the 18-24 age bracket has the highest usage of social media when correlated with the fact that brain development and maturing ends somewhere in your mid-to-late 20s (NIH) means that a large quantity of social media users' brain development is impacted by the rage bait and unrealistic standards algorithmically brought to the forefront of media. For reference, Instagram's age statistics are as follows: Age Range: 18-24, 31%. Age Range: 25-34, 29.5%. Age Range 35-44, 15.3. With the remaining 24.2% being split between the ages of 13-17 and 45-65+ (Oberlo). This has led to an increase in loneliness as well as other mental disorders among Gen Z (Haltigan). On top of this, we are constantly confronted with advertisements. Eventually, people buy the items leading to rampant consumerism which leads to lowered self-esteem and further environmental degradation (Stanovic).

Cyber Warfare:

This section falls into two categories. One category is military might, while the other is in regards to cybersecurity. In the case of the military AI is being used and or adapted to provide optimal automated targeting. This results in the loss of less civilian life (Mehan). In regard to cybersecurity, malicious actors have been using AIs in order to increase the scope and or span of their attacks. They have also been using AI to find loopholes and or as a cyber-lockpick. Picking the lock [Password] of people’s accounts. Cybersecurity specialists have begun to use AI to help protect against said attacks however it isn’t always effective (Mehan).

Chapter 5: Environmental Disaster

Electrical and Environmental Costs of ASI:

As we have already discussed we are currently at the beginning of the age of AGI. Yet even still an AI like GPT-3 needed 1,287 megawatts per hour in order to assess the data fed to it. In total, it used 1,287,000 kilowatts of energy over its lifecycle. For reference, this quantity of electricity could have been used to power 100 households for an entire year (Ludvigsen). The newest GPT model known as GPT-4 has been leaked to have used somewhere between 51,772,500 and 62,318,750 KWh. Now that you have that basic knowledge, let's assume an ASI is exponentially larger reaching and has the ability to supersede the intellect of a large portion of the population. If that were the case then it would require magnitudes more than what is available in an industrialized society. (Stiefel). If theoretically an ASI was made it would likely end up releasing an equivalent of 2.3 trillion tons of carbon dioxide into the atmosphere; the same amount as the US since the beginning of the Industrial Revolution (Hone). This amount would lead to further environmental destabilization and likely mass-scale deaths.

Potential for nuclear disaster:

Given how much electricity would be needed in order to power an ASI, I find it highly likely we would need to use nuclear reactors as a power source. While it is unlikely for a US nuclear reactor to have an uncontrolled nuclear reaction, it is possible. An uncontrolled nuclear reaction could result in widespread air and water contamination due to the radioactive products created during a nuclear reaction (EIA).

Chapter 6: Final Thoughts

AI is a powerful tool at our disposal. We have the ability to enlarge the job market as well as make others' lives more comfortable. The main downside is that those who lose their blue-collar jobs due to AI implementation may not have the intellect to obtain a job that requires higher intellect as well as the environmental impact of AI. AI also negatively impacts the social lives of thousands of teenagers and young adults leading to a heightened need for approval and a lack of self-worth. Overall AI may simply be best as a tool and maybe we should not attempt to create something sentient without first finding ways to mitigate or nullify its environmental impact.

** Work Cited **