Artificial Intelligence’s Impact on Jobs​

Third Episode of the Automated Podcast. Check out the podcast episodes at https://automatedpodcast.org/

Intro

Artificial Intelligence. From all the discussions and books and people I’ve listened to, AI appears to have the largest impact on jobs from any technology and this is why I wanted to start with this tech. Today we’ll look at what AI is and how it came about. discuss how advanced it is, who is building it, and then look at present and future impacts as it related to jobs. But let’s first briefly look at the history of AI.

History of tech

John McCarthy is one of the “founding fathers” of artificial intelligence, together with Alan Turing, Marvin Minsky, Allen Newell, and Herbert A. Simon. McCarthy coined the term “artificial intelligence” in 1955, and organized the famous Dartmouth conference in Summer 1956. This conference founded AI as a academic discipline. In the years since AI has experienced several waves of optimism, followed by periods of disappointment and the loss of funding, known as an “AI winter” (occurring in the 70’s and late 80’s and early 90's). We are currently in a period of optimism where technological advancement and large amounts of funding are going towards this technology.

Who is advancing it/building it

Regarding who is building AI, it really is between the two leaders right now, USA and China. As of 2018 China has about 1,011 AI companies in the mainland, accounting for 20.53 percent of the world’s total, following the United States with 2,028. However, China has an ambitious policy paper which aims at supporting Chinese AI development to become the world’s leader in AI development and research by 2030. And this might actually happen..as a 2018 study found that China is “poised to overtake the US in the most-cited 50 percent of papers this year, in the most-cited 10 percent of papers next year, and in the 1 percent of most-cited papers by 2025. So even though AI might not mainly speak English in the future, what exactly is it? If you’ve paid attention to any talk of AI you’ve probably come across the terms machine learning and deep learning connected to AI.

What the tech actually is

You can think of deep learning, machine learning and artificial intelligence as a set of Russian dolls nested within each other, where Deep learning is a subset of machine learning, and machine learning is a subset of AI, which is an umbrella term for any computer program that does something smart.

AI — A computer system able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.
Machine Learning — machine learning is dynamic and does not require human intervention to make certain changes
Machine — learning programs, in a sense, adjust themselves in response to the data they’re exposed to (like a child that is born knowing nothing adjusts its understanding of the world in response to experience). — they optimize constantly through repetition and many many many failures.

Deep learning — a “field of study that gives computers the ability to learn without being explicitly programmed” — while adding that it tends to result in higher accuracy, require more hardware or training time, and perform exceptionally well on machine perception tasks that involved unstructured data such as blobs of pixels or text. Ultimate goal in some AI circles is Artificial general intelligence (AGI). It is the intelligence of a machine that has the capacity to understand or learn any intellectual task that a human being can. Thus we have a few categories of AI. Narrow or weak which is what we currently have today which has the AI focus on one specific task, and the second category being general or strong (though strong can also be used to describe a state of being self aware. (For those of you more interested in a further breakdown of what AI actually is beyond this very superficial look, as it is vastly more complicated and deeper than briefly outlines here. There are really great books out there, but also some great podcasts that dive much deeper than what fits the scope for this one. I personally like to listen to Lex Fridman on the Artificial Intelligence podcast as well as ‘this week in ML and AI) -list and links in the shownotes, apologies for making it so simple but I could easily spend an entire 30 minutes pelting everyone with the finer specifics of the how of AI as well.

So these are the main relevant elements of AI, but what does this mean in practice? What can AI actually do?

Where we are today — current capabilities

The best way to show current capabilities of AI is perhaps to run through some of the more known instances where AI beat humans at tasks. Arthur Samuel, one of the pioneers of machine learning. Samuel taught a computer program to play checkers. He succeeded, and in 1962 his program beat the checkers champion of the state of Connecticut. In 1996 Gary Kasparov, the reigning global chess champion, defeated Deep blue, IBM’s supercomputer, in a 6 match set. After undergoing significant upgrades, DeepBlue beat Gary in their second set only 1 year later in 1997. Though this was and still is perhaps the most publicized event of man vs AI, many don’t even consider Deepblue to be AI, as it was an example of GOFAI (Good Old Fashioned AI — if-then statements are simply rules explicitly programmed by a human hand. Or simply developed for a pre-defined purpose) as opposed to deep learning which would come a decade later. Though Deep blue was able to calculate these if-then statements at a tremendous or one could say..superhuman rate. Another, possibly just as famous example happened over a decade later, in 2011 IBM’s Watson competing on Jeopardy! against legendary champions Brad Rutter and Ken Jennings, winning the first place prize of $1 million. However, this was still considered GOFAI. Yet, in 2016 Google DeepMind’s AlphaGo program, beat the 18 time world champion Lee Sedol. In 2017, it’s successor Alpha Go Master, beat the then reigning world champion Ke Jie. Go is important because it cannot be beaten through brute force calculations like chess or Jeopardy, due to there being more potential valid moves than there are atoms in the universe. For this reason Go has often been seen as a game that uses intuition, where Go masters often describing great moves as ‘feeling’ right. Alpha Go, used a number of methods to first mimic professional human players, then learn new strategies for itself and studied go match databases essentially studying to the extent that someone with 80 years of experience would have. To give some scope of the exponential or explosive power of deep reinforcement learning, alpha go zero the next iteration, in 2017, within 3 days of playing against itself (this means no data on human matches) beat the original alpha go 100–0. And then conquered alpha go zero in 21 days. Then in December 2017 — Alpha Zero….within 24 hours beat the 3 day alpha go zero version. Also examples of professional video game players in popular games such as Starcraft or DOTA 2 being. open AI five against DOTA 2 and Deepmind Alphastar against Starcraft — beating the champions in both cases. Apart from games though, there are many examples of other applications already being used or in pilots. Woe bot for instance is an AI chat bot using cognitive behavioural therapy, to help with depression. It sends over a million messages a week to support users with anxiety, depression, loneliness, addiction and other problems. In a test against three expert human radiologists working together, Enlitic’s system was 50% better at classifying malignant tumours Another of Enlitic’s systems, which examines X-rays to detect wrist fractures, also outperformed human experts. More well known is the use of AI in vehicles, enabling them to be partially autonomous today. However, these examples are just of narrow AI — that can focus on a single problem. but it should be apparent that though these examples are at the beginning of their capabilities, if we are to use the alpha go example as a measure, AI can improve very very quickly. So these examples, within a few years could be much further advanced if sufficient funding and work is done on them.

How AI impacts jobs positively

If we are to look at AI through the lens of benefiting employment, we can see that the AI industry as of 2018 was worth about 1.2 trillion and is estimated to reach up to around 4 trillion by 2022, and with it, has generated a significant amount of jobs focused on building and improving the technology, business cases, and applications for it’s use. Current Narrow AI also augment people’s capabilities in what has been termed centaur teams, where humans work beside AI applications to improve the quality of work done. Though this was originally used for chess players using their computer to augment their strategies, today AI is used in specific tasks. The radiology example above actually can enable radiologists to focus on other aspects of their jobs. Even IBM’s Watson was implemented in lung cancer treatment in NYC where staff used Watson’s guidance when dealing with patient diagnosis and treatment plans. As far back as 2013 90% of nurses were already following Watson’s guidance. Even lower levels of autonomous driving enable people to parallel park, which for any novice driver can easily be seen as an absolute blessing. I even had to delay my first driving exam due to having problems with this.

How AI impacts jobs negatively

On the other hand however, One specific perspective on AI’s future, is why it was chosen as the first tech to look at. Specifically, once AGI comes about it is entirely possible that human, creativity, ingenuity, and capacity for reason will be made completely inferior, much like our chess skills. Though there are still competitions and new world champions, they are still no match for their digital counterparts. AGI is seen as also bringing a new revolution of innovation, that is difficult to fathom, especially if it will be able to modify it’s underlying programming and make constant modifications and upgrades without human interference. I mentioned before that new projects leverage current technologies augmenting various capacities of individuals, example of this podcast was used as all that is physically needed is a computer, however by leveraging AI it would be possible to have all the research done, all the editing, all the writing, and even publishing and sharing. All I would be needed for is speaking, though with a decent speech synthesizer I wouldn’t be needed for this either. The same could be done for numerous professions as any mental activity needed to improve a technology or project could theoretically be done faster, and with more precision and quality by an AI. If taken to it’s logical conclusion we would be hard-pressed to think of a place where humans would be relevant or needed for employment purposes at all? And if you’re thinking, AI could never create music, or beautiful paintings, or art in general, please check out the shownotes where there are links to early AI music such as Aiva AI which is was listening to while prepping this episode) and art. Though it may not currently please everyone, the point is that even with the very narrow and quite young and immature AI programs today, this is already currently possible. For this reason many have stated that AGI would be the last invention humans would ever need to create. Some even go so far as saying that all of humanity, even our entire purpose, might simply be reduced to spawning this new form of ‘life’, much like many people believe that the first single celled life forms’ purpose was to bring about us. Though this is certainly a point of view that many take today, it really is not possible to clearly guess how the more distant future will turn out, so in the meantime let’s turn the conversation back to current ideas on the impact of jobs.

A now classic 2013 Oxford study the future of employment — has been widely used to explain which jobs are at risk of overall automation and which ones aren’t. Very basically Low risk of automation jobs — require creative knowledge and innovation ,education, media, health care. Whereas jobs at high risk are -predictable or routine — such as accountants, junior lawyer, telemarketers. This is supported and added to with a new study by IBM claims that as many as 120 million workers in the world’s 12 largest economies may need to be retrained or re-skilled as a result of AI and intelligent automation.

The study, includes input from over 5000 executives in 48 countries that explain how a fundamental shift in how companies manage changing workforce needs is required. Apparently, the time it takes to close a skills gap through training has increased by more than 10 times in just four years. In 2014, it took three days on average to close a capability gap through training in an average enterprise; in 2018, it took 36 days. The study showed that new skills requirements are rapidly emerging, while other skills are becoming obsolete. In 2016, executives ranked technical core capabilities for STEM and basic computer and software/application skills as the top two most critical skills for employees. Yet In 2018, the top two skills sought were behavioural skills — willingness to be flexible, agile, and adaptable to change and time management skills and ability to prioritize. Therefore, it is easy to see that as narrow or weak AI becomes more and more advanced and utilized by organizations across the world the impact on our employment and jobs in general accelerates as well. However, if AGI, supported by other technologies does bring about tech unemployment on a massive scale as more and more people are advocating, Then large scale social social and economic shifts are inevitable, and further discussions are absolutely necessary. Though AI is perhaps the main technology to cause an eventual shift of this magnitude many other techs are also potentially contributing which we will look at in the preceding episodes.

Thanks for listening. If you enjoy the podcast so far feel free to subscribe, and rate and review on itunes or wherever you listen, which will help out the AI to distribute this to more people.