The last few years have seen a dramatic increase in conversation levels around the world regarding Artificial Intelligence (AI).
There’s been a good amount of dialogue covering a wide variety of topics: everything from Movies like Ex Machina exploring the first instances of true AI creativity, or Westworld exploring the consequences of AI becoming more self-aware and what kinds of humans we may become with AI around, or Sam Harris’ TED talk about the possibility of losing control of AI, intelligence explosion, etc. It’s a hot topic, and it deserves to be.
We’ve also recently seen major technology giants jumping on the AI bandwagon, with folks like Google and Microsoft essentially declaring their vision of “AI first” in dramatic ways.
But what does all of this mean to you and me? Especially, what about our fears that AI might be a significant disruptive factor in the job market? This article aims to reason about the future possibilities in as pragmatic a way as possible, based on my research into the topic.
Jobs and “value production”
To understand the implications of AI in the job market we first need to briefly discuss what “value production” is.
When anyone gets paid money, unless it’s a free gift, welfare or charitable donation of some kind, it is in exchange for “value” provided to the payer.
The reason why you have a job at the moment is because something about you (the skillset that you have in a particular domain) enables you to offer value to your employer (and either directly or indirectly to your employer’s clientele).
Now, as special as we would like to think we are, our employers hire us to add value to our company. What they’re after is not us per se, but our skillsets. They may like us, or even be close friends with us, but it costs them money to hire us, and so we need to produce value in return. This is the basic value for money transaction.
How do machines and AI factor into this equation? Let’s look at some history.
A brief history of AI and jobs
Before there was AI, there was computing. Before there was computing, there were machines. And before there were machines, there were tools.
- When you put together a bunch of tools and figure out a way to make them function together more meaningfully, you get a machine.
- When you put together a bunch of machines and figure out a way to make them function together more meaningfully, you get computing.
- And when you put together a bunch of computing and figure out a way to make it function together very meaningfully, you get artificial intelligence.
Ever since the advent of tools humans have been using them to augment the value that we can produce for one another. The person building walls built them quicker and cheaper with access to the right tools: chalk lines, circular saws, framing hammers, etc. — these all allowed wall-builders to compete more effectively with one another and deliver more value.
Tools have played a significant role in building the societies in which we now live. Most of us live in some form of a concrete jungle: moving from cars to houses to skyscrapers — none of that would be possible without tools.
The agricultural revolution: tools & processes scaled food production
The agricultural revolution is a good example of what tools allowed man to do. Thanks to tools and processes, we were able to create more value for each other in the form of greater food production capabilities. As a result of this, human populations skyrocketed and our societies changed forever.
Think about how the “job market” (or “proto job market”) changed with the advent of tools and processes that lead to the agricultural revolution. Hunter-gatherers had very broad skillsets that are different from farmers. When we became farmers the number of hunter-gatherers decreased substantially, replaced by the new skillset of farmers & tools which was able to deliver more value.
Yes, new jobs were created, but the old ones genuinely became redundant.
The industrial revolution: machines scaled production
Fast forward and we have the industrial revolution, an example of how machines took us to the next level. Machines could construct things like cars much more efficiently than humans can. This allowed businesses to scale better; offering products at cheaper prices (thus making them more accessible to the everyday person), and producing more, however there was an impact on shaping the job market. Many jobs that humans used to do were no longer required.
The information age: computing scaled information processing
Fast forward again and computing has done the same thing. Now NASA doesn’t have a job post open for any more “computers” like they did in the 1960s. Stock trading is done automatically through large and complex computer systems. These admin / computing things were once done by humans, but are no longer.
What do humans do when their jobs are replaced?
Of course, it’d be perfectly reasonable to say: “Yes, but at every point you’ve mentioned above, humans left their old jobs behind to do more sophisticated and higher value work. Hunter-gatherers became farmers and could produce more food. NASA scientists could get the computer to run the calculations thus being able to produce more “science”, Sure, machines produced cars in the place of humans but those humans gradually found other more interesting, more sophisticated and intelligent work to do, thus our entire society is trending “upwards”, to more valuable work.
This is, of course, true. In fact this is how we got here. If one looks at history one sees a decreasing number of people involved in agriculture / food production. Where once human society was a pyramid with a vast majority of people being peasants at the bottom producing food and only a small minority at the top living off that food, now the pyramid is reversed, with only a very small minority of our population creating food for the rest. This has been made possible by innovation: tools, machines and computers.
At each point when tools, machines and computers have “threatened” (and “replaced”) our jobs, we’ve had the opportunity to look upwards: to greater value work, to more interesting, creative and innovative areas, thus together we and machines have scaled productivity to the current levels we see in our “concrete jungle” society.
But some of us think this has a theoretical limit: enter AI.
Domain-specific artificial intelligence
Whenever machines have threatened our jobs humans have evolved. Case in point, I work as a SEO specialist — an occupation that didn’t exist 15 years go — in fact the entire digital marketing industry and niche is a massive industry that didn’t exist 20 years ago. This entire industry is enabled by computing. Business meets the internet and is “digitally transformed” and this has created a significant number of jobs — great!
But artificial intelligence may have the potential to change all of that.
Now, current artificial intelligence is doing well, producing well and that’s great. However, it is very domain specific. For example, IBM’s Watson was the first AI to win at the game of Jeopardy — it could beat the world’s best players at this one game. But it can’t beat them at everything. Similarly, Libratus beat some of the world’s best players at poker, but that’s all it can do.
Now with that being said, current levels of artificial intelligence are still dangerous to a number of occupations — To give an example, advances in voice recognition and problem solving within a very specific domain threaten to replace call-centre workers and are already becoming ubiquitous in the first few interactions with a call centres for certain big companies before you hit a human (like banks or utility companies).
Not all humans are equally skilled, and some skillsets are already in danger of being (or are already in the process of being) replaced by computer and AI systems.
General artificial intelligence
That said, the truest and greatest threat against human jobs is not a single AI that can outperform us in a single domain — because each time that happens we have the potential to evolve and find new domains in which to compete and produce new value. No, the true threat is the potential for general, human level artificial intelligence.
I recently caught up with a computer science academic in the university where I studied computer engineering. I asked him the question:
“Is there anything special or magical happening in this grey matter between our ears that would make artificial intelligence never reach our level of intelligence?”
His answer: “No. There’s nothing special happening up there (intelligence wise) that could not be simulated in principle”
Think about what human level artificial intelligence means. By definition it means computers that are as smart as humans. Imagine something as smart as you but it never sleeps and in certain domains it can process information much faster than you can. How do you compete with such a thing?
If general, human level artificial intelligence becomes a reality, I don’t know where humans will “retreat” to in order to keep “producing value” for one another.
Now, let’s talk about some of the common discussion points / areas about the potential of human level AI and common counter-arguments for the potential threat it may pose.
But machines will never have consciousness!
Consciousness is the subjective nature of experience. To be you in this very moment has a subjective character to it — you are conscious. To the best of our knowledge machines don’t have consciousness and never will. This is true.
However, consciousness is not a unique selling point. When you go to do your internet banking website to do some banking, you don’t care whether the website has subjective feelings — in fact, you probably have a better experience overall because it doesn’t!
Many things that add value to us on a daily basis don’t have consciousness and frankly don’t need it. I am typing this article on my laptop right now — my laptop doesn’t have consciousness and it doesn’t need to — it just needs to function well and “produce value” in its specific domain.
So much for the basic consciousness argument.
Consciousness, empathy, “bedside manner” and “uncanny valley”
Another argument is made around consciousness — machines will never have empathy for us and thus certain jobs are permanently “safe”, like a nurse or doctor who comforts someone when they’re vulnerable.
- Bedside manner can be programmed. Facial cues can be read and understood by AI. a “soft voice” and a “tender touch” can all be programmed. The practical elements of interacting with humans with empathy can be discovered / programmed into AI.
- Thus, the biggest challenge to overcome is not actually whether AI can learn to simulate empathy (it doesn’t need to really “feel” empathy, it just needs to interact with humans with simulated / realistic empathy) but whether the humans on the receiving end of AI empathy can surpass the “uncanny valley” phenomenon.
- In this context, “uncanny valley” means the propensity of humans to not really “feel” the empathy from the machine because they know it’s a machine and not a human.
- Similar to how when you call a call-centre and hear the words “your call is important to us, we will be with you shortly” you feel cynical about those words because you know they’re just a recording being played by a machine.
In reality, I think this problem is more complex than I am making out above. To illustrate with an example: people watch movies. In reality, when they are viewing a movie they are just seeing movements of electrons on a screen (everything involved in this interaction is just a machine playing an old recording) — and yet the “uncanny valley” is anything but a problem here. The feelings are still real, even if everything else is not “real” real.
As an afterthought, if human level AI with fully fledged “empathy” and “bedside manner” is created, I think it will only take a single generation of young people growing up with it for it to work. Just like when the first automatic elevators came out people “preferred” human-operated elevators, but eventually got over it, so I think AI powered “bedside manner” will just be a matter of time before it’s adopted widely.
The Chinese room: Machines don’t “understand” what they are doing
When Deep Blue first beat Garry Kasparov (the world chess champion) at chess in 1997 — did it know it was playing chess? No. It was just a program.
This is called the Chinese room — so named because a hypothetical AI that translates languages into Chinese (like Google Translate does today!) and sends the responses back in Chinese doesn’t necessarily understand that what it is speaking is Chinese or the meaning of the Chinese words themselves (or that what it is currently doing is translation).
So the argument is made: AI will never be as intelligent as we are, because it doesn’t truly “understand” what it’s doing.
- That’s why it’s called “artificial” intelligence. 🙂
- What is “understand” (in this context) but to grasp the relationship of a specific activity and other possible activities? So I understand that what I am doing is playing chess — what does that really mean? Is it not merely grasping that one is in engaged in the activity of playing the game of chess? I don’t think there’s a rule that will prohibit machines from ever gaining this level of “understanding” (or, simulating it, which is the important point, again “artificial”).
- (most importantly) for most cases it doesn’t actually matter — my laptop doesn’t “understand” that there is a human typing on it right now, but it can still produce value. Deep Blue didn’t understand that it was playing chess, but it still beat Garry.
- Machines don’t need to “understand” what they are doing in the same way that we understand. They just need to have intelligence in a particular domain to compete effectively and deliver value.
- Let’s say I understand the meaning of the English word “cat”. What is this “understanding” but knowledge of an association between this sound of the mouth and the physical animal? — AI can have that, and it already does.
But humans will just keep climbing the value ladder — creativity!
If true human-level AI is developed, (AND it is cheaper to acquire than human skillsets / time) then there will be nowhere on the “skill ladder” for us to go — as anything we are intelligent enough to do machines will be also.
Will they ever develop creativity like us? There are several signs that this is possible, for example AI has already composed music (Jukedeck is a commercial AI music creating product that already exists), and it has created novel cryptographic algorithms, given that all of so-called “creativity” is really just reconfiguring pre-existing elements (that’s what the human brain does, sorry guys, nothing we create is absolutely original) there’s no reason machines can’t do this, and they already do.
Furthermore, it’s been speculated that increasingly intelligent AI could lead to an intelligence explosion in which case not only our jobs but also our existence could be threatened. Such an event is highly speculative and depends on a number of factors but if it were to occur the interests of Homo sapiens could fall down the wayside, along with our jobs.
A final anecdote: the momentum of machines (Centrelink story)
Recently I went with my wife to Centrelink — the place where Australians go to discuss their government benefits.
When we met the girl at the counter she said to us that her and her colleagues were striking just over Christmas because they hadn’t received a raise in too long a time.
When I went to sit back down I looked at the scene surrounding me. Human employees were using computers — machines — to look up client information, process client claims, input data, and have conversations with the incoming clients. Humans working with machines.
I noticed how there were no librarians, or archivists like there might have been many years ago with a paper system. Those jobs were gone.
I thought about how in ten, or twenty, or thirty years time, when I go to Centrelink there may be no humans left — I may just interact with machines, because they are already a significant part of the system, we just need something to understand human speech and be intelligent enough to process it before we can “get rid” of the human operants. Already much of the processing work that is done at Centrelink can be done online through websites. This is just the natural progression.
Machines don’t go on strike. They don’t take annual leave. They don’t demand raises. They don’t take days off sick. They don’t have emotional problems at home due to family circumstances. They don’t complain. They don’t sleep. They don’t eat. They don’t take coffee or smoking breaks.
They just perform. They just produce value. And they can be more professional than the professionals.
We live in a culture where, instead of being an insult, it’s actually a compliment to call someone “a machine”.
Machines already outperform us in many value production domains. What makes us think there is something uniquely special about a Homo sapiens brain that a machine cannot out-produce?
True, machines may never have subjective experiences. True, the machine that beat us in chess didn’t truly “understand” that it was playing chess, but that doesn’t stop them from producing value in those domains. Clearly, consciousness is not a unique selling point, and is intelligence (however artificial) is the is most valuable for value production.
When you go to do your internet banking website to do some banking, you don’t care whether the website has subjective feelings. You don’t care whether the website “understands” that it’s doing banking for you. In fact, you’d far rather use the website then get onto the phone with a human. Why? Because the machine is able to produce value in this domain that far surpasses the value a human can produce.
I don’t know what the future holds. This is not a prophecy. But I do think that if AI is able to go “general” and approach human level intelligence, “this time is going to be different”.