By Spike Narayan-
(Spike Narayan is a seasoned hi-tech executive, managing exploratory research in science and technology at IBM)
Is AI disrupting the technology job industry? Just scanning the newsfeeds every day, one would be led to believe that it is true. When we begin to peel the onion, two facts are obvious. One, many tech companies are trimming their workforce; and two, the tech companies that are downsizing are in the AI business for sure. The question before us then is, “Is there a causal relationship between these two facts?”
Let’s examine this hypothesis.
Over the past two decades there’s a fear that AI will take away jobs. This is true of any new technology. The first fear is always loss of jobs. For the most part, though, this is unfounded.
The advent of computers is a classic example. The initial fear and the consequent impact of computers are diametrically opposite. Not only did computers help humans do their jobs better, but the technology spawned dozens of new companies creating millions of new jobs. The question then is, why AI, as a technology, will not have the same long-term effect of job creation. It probably will, but in the near term there is an undeniable level of discomfort.
A lot may have to do with the word “Intelligence”. It connotes competition with the human brain. ‘Artificial Intelligence’ (AI) was coined in 1956 at the world’s first conference on the subject organized by John McCarthy, then a young assistant professor at Dartmouth College. McCarthy, one of the founders of the discipline of artificial intelligence, later went to Massachusetts Institute of Technology (MIT) and then spent the rest of his teaching career at Stanford University, where he developed new forms of mathematics and computation science.
During that 1956 conference, Marvin Minsky, the computer scientist who later co-founded MIT’s AI laboratory, defined AI as “the construction of computer programs that engage in tasks that are currently more satisfactorily performed by human beings because they require high-level mental processes such as perceptual learning, memory organization and critical reasoning.”
Decades later, after the so-called ‘AI Winter’, when compute power, data sources and algorithms were sufficiently advanced, AI saw a resurgence. However, when such AI programs were introduced, the technology was ironically neither artificial nor intelligent. It was simply a fancy calculator with statistical analyses capabilities that helped visualize data and trends. Over time, it developed the ability to crunch and analyze larger amounts of data, helping us visualize data that has more than three dimensions.
In the last decade, AI engines have surpassed human capability in several tasks, leading to a renewed debate on job losses. Much of the gain can be attributed to deep neural networks (DNNs). Many neural network variations like convolutional neural nets (CNN), recurrent neural nets (RNN) and long-short term memory (LSTM) and dozens more have been developed and optimized to suit different use cases.
Many of the more recent advances are impressive in the domains they were fine tuned for. Surprisingly, CNNs and their variations were starting to be viewed as augmenting human capability rather than as threats to our jobs.
This shift in perception could be because many of these AI applications are actually quite useful and good enough to warrant mass adoption. For example, movie suggestions from streaming services, vacation planning websites, dating apps, shopping for goods and services… the list goes on. None of these AI-aided services are viewed as a threat to jobs.
So, what has changed in the last six months?
If I were to summarize in one phrase, it would be Large Language Models (LLMs), a technology sweeping the tech industry like no other. Many companies have been working on it for the last three to four years, but it was not until OpenAI’s ChatGPT (Large Language Model) was made public in late 2022 that people saw its awesome capabilities.
You can learn more about LLMs here.
What, then, is the main difference between DNNs and LLMs. In DNNs there is significant amount of training needed to infer (predict) accurately in a chosen field of use, thus making a particular DNN useful only in the area it is specifically trained for.
In contrast, LLMs are trained more generally from extremely large data sets so that one can attain high inference accuracies in a chosen field of use with very little application specific training. The result of this is that LLMs can be used in orthogonal fields easily with relatively small tweaks, making this technology quickly scalable.
The claim, therefore, is that AI can now be available for mass adoption more readily. The potential for pervasive use is easy to see. All this still only points to the fact that there will be more AI applications that we can use to do our jobs better.
What is the connection to job cuts? The causal relationship is nuanced. Many of the layoffs that have made news this year appear to be connected more to over-hiring than to AI. Companies have publicly stated that even after the resource actions, they will have more employees on their payrolls compared to 2019.
Companies saw revenues skyrocket during Covid-19, leading to rapid payroll bloating. As Covid-19 waned, revenues followed suit, leading to job losses to maintain corporate profitability. But for the Covid related revenue and hiring bumps, we would not have seen job losses of this size. Unfortunately, they happened during the time ChatGPT made its appearance. It was coincidental, in my opinion.
In summary, AI, like any new technology, will lead to some temporary job losses but will lead to significant job gains over time. As with any technology adoption there will be new education and skills training needs for young and old alike. LLMs will change the landscape of AI applications making them much more useful, will and help us do our jobs better.
The recent job losses, in my opinion, cannot be convincingly linked to AI adoption.
