The over-hyped world of AI.
Artificial
intelligence won’t change as much as the promoters claim. But, for some, its
impacts will be disruptive – and sometimes disastrous.
There is something about artificial intelligence in
computers that induces a special kind of madness in humans. The considered,
insightful thinking that’s so urgently needed in this new era of technological
disruption has been all but swamped by hype, financial speculation and blather.
The latest burst of effervesce began at the moment ChatGPT
was launched on 30 November 2022. Suddenly, you could ask a computer to write
an essay for you and it would. It might not be the best possible essay but, as
Samuel Johnson said in another context, “is like a dog walking on its hind
legs. It is not done well; but you are surprised to find it done at all”.
The capability of artificial intelligence is improving so
quickly that its inherent limitations are being overlooked. These supercharged
computers will change the world, creating problems for many and wealth for
some, but computers will never think. Only people can do that.
The bubble
With a degree of dignity long familiar from the gold rushes
of previous centuries, the lords of Wall Street and Silicon Valley, closely followed
by legions of get-rich-quick speculators around the globe, have pushed up tech
stock prices to unrealistic and unsustainable levels. Since the ChatGPT launch
in 2022, a small number of favoured companies have so outperformed the “old
economy” that they have profoundly distorted the entire US sharemarket. In
three years the Dow Jones Industrial Average, which tracks 30 major industrial
firms (only two of which are major tech stocks) rose by 34%. In contrast, an
index of 41 IT stocks rose by 98% and Google, one of the leaders, went up by
143%.
In a note to investors Michael Cembalist, an analyst at J.P.
Morgan, wrote: “AI related stocks have accounted for 75% of S&P 500
returns, 80% of earnings and 90% of capital spending growth since ChatGPT
launched in November 2022.”
When the bubble bursts, as bubbles always will, a great deal
of money will have been made and lost, not usually by the same people. As with
the 2000 tech-wreck, smaller outfits can be expected to be decimated but most
of the major companies – for the time being, anyway –have solid fundamentals.
But the bubble is not confined to share prices.
“AI is showing up in other places as well. Data centres are
eclipsing office construction spending and are coming under increased scrutiny
for their impact on power grids and rising electricity prices,” Cembalist
continued. “In the PJM region (of 15 midwestern states) 70% of last year’s
increased electricity cost was the result of data centre demand.”
Oracle and Open AI (which owns ChatGPT) look increasingly
shaky. “Oracle’s stock jumped by $60 billion a year,” Cembalist wrote, “an
amount of money Open AI doesn’t earn yet, to provide cloud computing facilities
that Oracle hasn’t built yet, and which will require 4.5 gigawatts of power
(the equivalent of 2.25 Hoover dams or four nuclear power plants.
Oracle hopes its bet will pay off, allowing it to secure
first-mover advantage and taking market share away from its competitors. But
those competitors – Amazon, Alphabet, Meta and many others – cannot afford to
let that happen. The race is now on.
This sector, though, is already in hock. At the time of
writing, AI companies had $US1.5
trillion of debt. For every dollar Apple has in equity, it owes $155. IBM’s
ratio is 245% and Broadcom is 166%. Other big companies have, for now, less
debt. But as they too join the infrastructure race, and as their share prices
plunge when the bubble bursts, those figures will skyrocket.
The AI revolution depends on those data centres being built,
not only in America but everywhere. If they turn out to be impractical, so will
the AI revolution.
The limits of AI
People think. Computers calculate. There’s a difference.
The mobile phone in your pocket has vastly more computing
power than the system that guided men to the moon in 1968. But even today’s
massive digital capabilities cannot yet produce a self-driving car that can
take you with an acceptable level of safety to work and home again.
Here’s the difference. The trajectories of the moon and
Earth, the thrust of the rocket engines and the effects of gravity were all
known and predictable. NASA engineers could be confident those critical
parameters would not change. The mathematical jargon for this is
“stationarity”.
ChatGPT can usually turn out a passable undergraduate
history essay, where all the information sources are known. But can you trust
it to rank the credibility and relevance of competing sources, as it would in a
more advanced and complex task, where values and nuance are involved? Can it
judge whether the Glorious Revolution of 1688 was actually glorious, or just a
Dutch invasion?
The thought processes which most clearly give humans the
edge over computers – and always will – are the most elusive: instinct and
intuition.
Millions of years of natural selection have given us
genetically programmed behaviours which have helped Homo sapiens not
only to survive but to become the dominant species. Some are automatic actions:
you withdraw your hand from a hot cooktop even before the pain registers in
your brain. Others frame our most fundamental judgements: we evolved as a
social species, so we cooperate with others. In the workplace, that has allowed
us to produce the complex teams and organisations of which the modern world is
made: governments and corporations.
Computers, no matter how much they “learn”, will never form
a corporation or run a government. Only people can do that.
The eminent American psychologist, Gary Klein, has described
intuition this way:
“Intuition depends on the use of experience to recognise key
patterns that indicate the dynamics of the situation. Because patterns can be
subtle, people often cannot describe what they noticed, or how they judged a
situation as typical or atypical. Therefore, intuition has a strange
reputation.”
“It is a simple fire in a one-storey house in a residential
neighbourhood,” Klein wrote. “The fire is in the back, in the kitchen area. The
lieutenant leads his hose crew into the building, to the back, to spray water
on the fire, but the fire just roars back at them …
“Then the lieutenant starts to feel as if something is not
right. He doesn’t have any clues; he just doesn’t feel right about being in the
house, so he orders his men out of the building – a perfectly standard building
with nothing out of the ordinary.
“As soon as his men leave the building, the floor where they
had been standing collapses. Had they still been inside, they would have
plunged into the fire below.”
What the fire chief didn’t know, but sensed, was that there
was an unseen basement below, where the real fire was.
“Many people think of intuition as an inborn trait –
something we are born with. I am not aware of any evidence showing that some
people are blessed with intuition, and others are not. My claim … is that
intuition grows out of experience.”
Intuition, then, can be seen as a kind of learning. But it’s
very different from machine learning. It is inconceivable that an artificial
intelligence computer system could make the decision that the fire chief did on
the overt evidence that was available. AI works in a defined, stable world. The
real world in which humans live is often unknown and messy, needing a very
different kind of perception and thinking.
Artificial intelligence has an important role but its
potential has been massively exaggerated. It is perhaps best seen not as a
revolution in itself but as the latest chapter in the story of technological
change that began 250 years ago. Nevertheless it, like its predecessors, will
change the world: and not always for the better.
Which jobs will go?
Despite many confident predictions, no two of which proffer
the same result, nobody knows how many jobs will be replaced by artificial
intelligence. Some clearly will, and already are. But most – probably a large
majority – seem relatively safe. Nevertheless, enough people are plausibly
likely to be replaced by machines to produce massive disruptions to societies
and economies, most often in the developed world. Along with climate change,
this will be a potent potential threat to the western democracies and the way
they are governed.
A huge number of jobs have already gone, casualties of
technology and globalisation. Take car manufacturing.
When Australia’s last car factory closed in 2017, at least 40,000
jobs in the sector had gone. Many never worked again. Around three-quarters
of the 900 people thrown out of work by the closure of the General
Motors-Holden plant in Elizabeth, a working-class suburb of Adelaide, found
jobs within two years. But two-thirds of these were in precarious,
lower-paid employment: casual, contract or part-time. Less than 5% found new
work that had better conditions or higher pay. Younger people were more likely
to be employed; those over 45 mostly stayed on the dole.
Other functions are only now being replaced. Ordinary
wage-earners routinely paid tax accountancy businesses to fill out their
relatively simple returns. The bulk of that work was not done by the qualified
accountant but by lower-paid staff, often casual or temporary and employed only
at tax time. The accountant would take a look at the finished product, sign it,
send it off and take the fee.
Now, the Australian Taxation Office and other agencies
around the world get data automatically on taxpayers’ wages and investments,
and are able to use artificial intelligence to pre-fill those returns. All the
taxpayer usually has to do it fill out any extra deductions and click the link.
The accountants are still likely to have enough more complex
tasks to keep them in business – but without those extra employees who once did
the donkey-work.
This is typical of the vulnerable jobs which are likely to
go as AI becomes more capable, refined and reliable: routine tasks with a
well-defined, finite set of inputs. But that does not describe the majority of
work now done by humans.
There are two categories of job replacement. The first, like
the car factory experience, is when a job once done by a person is now done
entirely by a machine. It has already replaced most bank tellers and has begun
to replace train drivers. Depending on how AI develops, taxi and truck drivers, many clerks, postal
workers, receptionists and parking inspectors could find themselves out of
work.
The second category is where a function is made more
efficient, only partly replacing human input. Tax accountancy is an example of
that. Another is radiology: AI can take over some routine tasks but its
capacity to replace specialist radiologists has been seriously overhyped. In
many cases, it has made
things worse. And its application in medical imaging has been poorly
researched. Radiologists will still have jobs, but some of their assistants
may not.
Between them, these two levels of job replacement are
already having a profound effect on employment. Nobody can yet know how many
people will be put out of work, but the economic and social impact will depend
not only on how many people are involved but on how quickly those changes
occur. If AI development and implementation happens as quickly as the industry
expects, the impact could be devastating – even if only a minority of jobs are
affected.
How would we live?
A couple of decades ago, The Economist devoted most
of an issue to what would happen when computers could do almost all the jobs
currently in existence. The cover showed smiling, relaxed people reclining on sunny
lawns in blissful leisure.
There were a couple of things wrong with this vision of
Utopia. The first was what would happen to all those people now without work?
What would they do? Who would pay for them to survive?
The most common answer is some sort of universal basic
income: an amount paid to everyone, regardless of how much they earn and
whether they work or not. But if AI replaces only – say – 25% of jobs, a UBI
would be irrelevant. Most people would not need it, and it might not be enough
for the 25% to live decent lives. Australia’s Jobseeker unemployment benefit,
for instance, leaves its recipients well below the national poverty line.
 |
| Queuing for the dole: the emerging underclass |
The cost of paying that to the 25% of employed Australians
who lose their jobs would be $220 billion a year. That would increase this
year’s federal government expenditure by 28%, bringing it up to $1 trillion. It
would replace the Jobseeker program, but that costs only around $16 billion.
How would the government pay that enormous, ongoing bill?.
Borrowing would be impractical: gross debt already stands at over a trillion
dollars. Taxes would have to be raised by 32%.
A far more likely outcome would be for people losing their
jobs to technology to have to rely, as in the past, on normal unemployment
benefits. As such, they would live around or below the poverty line, massively
increasing economic disadvantage and social dislocation. Over the past 45 years
of the neoliberal era, the bottom 50% of Australians have gone steadily
backwards while those at the top have prospered.
The scale of disruption in employment, as computers replace
a large minority of the workforce, is likely to be unprecedented since the
Great Depression of the 1930s and represents a massive shift of wealth from
labour to capital. The promise of AI is in fact a promise of intractable
inequality.
Potentially, artificial intelligence is second only to
climate change as a disruptor of human society. Democratic government is
already under siege because of the destruction caused by decades of
globalisation and neoliberal economics. Right-wing populists, promising to take
the world back to the way it was, are flourishing.
At no time since the technological revolutions began two and
a half centuries ago have the opponents of change prevailed. They will not this
time either: artificial intelligence will do as much, or as little, as
technology and economics allow. If any solutions are possible, they will be in
dealing with change rather than preventing it.
But where are the plans?