This is the third of a series of 3 posts (the first is here and the second here) I have been writing and publishing in my spare time, discussing the theme of the recent surge of interest in AI due to the release of ChatGPT by openAI. In the first article, I had ChatGPT write the blog post in my place; in the second, I discussed its statistical nature, which makes it perfectly suitable for repetitive tasks.
In this third post, I would like to rebuke the claims that we are close to the day when AI will be able to completely replace humans. I am an enthusiastic user of chatGPT since its inception and I find it absolutely mind-blowing, endowed with the potential to revolutionize what we mean by the words “working” and “learning”.
However, as I see it, such sensationalistic claims stem from the lack of understanding of what humanness means.
I am going to take a blue-collar tangent in this article and speak from a physical worker’s perspective.
What does knowledge mean without understanding?
Understanding is the interconnected layering of several pieces of knowledge. Such layering has to happen within a physical structure, to the best of my understanding. By “physical structure” I mean a living organism. To put it simply, a host body connected to the world through a system of inputs and outputs and endowed with feedback loop mechanisms.
Let me provide a real life example: I am next to a clerk trainee we are instructing during peak sales time on a spring Sunday morning. She is slowly leaning into the breakfast pastries window to reach for the croissant a customer asked her for after producing the withdrawal ticket issued by the teller less than a minute before. I am filling up a take-away tray with croissants with a given filling and only two are left and I need three to complete my order. Since I work very fast, they are gone before she can take hers because, as it turns out, the customer my young trainee is serving asks exactly for the same filling. Now, I needed three and there were only two left in the window, which means I have to wait for the next tray to come up from the lab anyway. She hesitates, seeing I emptied the tray in the window, but I know I have to wait in any case, so I encourage her to take what she needs from my order. I will keep in mind what I have to complete. Meanwhile, I move on to serving other customers.
The action I just performed is an example of understanding a situation on-the-go: I am connecting the visual information about the state of the pastries window with the present need of my employee and the customer she is serving; I know people are often in a hurry, so I am ranking the one-croissant customer’s request as more urgent w.r.t. the need of the customer I am serving, who has to wait longer anyway, all else being equal. Finally, I am using these infos to speed up the customer service following this pattern-matching my brain is performing and acting upon it. I am connecting several pieces of knowledge consisting of mental images drawn from my experience reservoir. The reason I am able to perform such connections is that my brain lives in a host body which stores information as a blend of physical sensations, emotions, visual images and the like…
Now, try to do this on-the-go adjustments by prompting a large language model, whose purpose is to find correlations among words, even one embedded in a robot’s armor and, moving from there, directing the robot to act!
Not only is that just impossible: even if we ever build steel and titanium robots able to perform a range of motions as close as possible to the capabilities of a human, we have no clue whatsoever as to how to embed it with emotions, which are such a crucial and automatic part in the selection of the content we give attention to. What’s even worse -to conclude- is that we have no clue as to how to define human attention.
Those may sound as trivial instances, but they are examples of the way manual work is hardly ever going to be totally replaced by machines. To further sharpen my point: the more human-to-human interaction the job implies, the less likely it is to be replaced by machines.
Of course, filling up Excel spreadsheets, recording invoices and perhaps coding itself are tasks that the human mind, tuned on narratives, story-telling and empathy, is less prone to accomplish than machines may be. Therefore, there is an actual possibility that such tasks will be, some day soon, almost completely automated. That is certainly a risk for the people making a living with such professions, but do we know how many opportunities it may open up? I claim we certainly do not.
Therefore, I am not scared about a future where I have to learn to use Artificial Intelligence to automate non creative tasks. If I have to worry about something, it’d rather be about AI replacing human creativity altogether. But this is a possibility I see as more remote: if, on one hand, I do see a plethora of AI-powered content being created in the future, at the same time I do not foresee the human need to express oneself in order to produce meaning waning in any case. And this is something which AI cannot replace. At best, it can boost it dramatically. As I have tried to argue several times, technology is a force to be reckoned with, not something to fight against.
Still, despite the fact that these lines of arguments sound so reasonable to me and to people I have discussed them with, I cannot help by feel that, in the general discourse, there is a looming fear of an AI-driven future. I would like to explore this just a little bit more in the next section.
Why do we find AI’s development scary?
The easy answer is: because we have always found contemporary developments in technology really scary for the future of work.
When spinning frames were introduced into the English textile industry in the mid 18th century, during the first industrial revolution, rebellions soon ensued among the textile industry’s workers, afraid that the new technology would take their jobs away. And it absolutely did. Frames took many a sewer’s job away. It’s just that new jobs emerged. It is a loss of opportunity for people being replaced while this happens. This is not questionable. It is a gain in terms of opportunity to employ people otherwise in the medium-to-long term. Technology is unstoppable, but it comes with trade-offs.
People afraid to lose their jobs are right. People afraid that no new jobs will appear in place of the old ones are wrong. The future is so hard to predict because we tend to imagine that the future will look like the past. This is part of being human, but we always fail to learn and persist in our attempts to predict, no matter what.
I also think that there is a deeper, even existential reason, which is specific to the case of Large Language Models.
First of all, AI has been around for way longer than LLMs. But is has gained the spotlight only after the release of ChatGPT. LLMs are, for most people in the mainstream who do not have any other acquaintance with computer science, a synonym for AI, despite the fact that they are only a tiny part of the field. Glooming predictions about the end of knowledge workers and -to a lesser extent- of work in general have started to circulate in the media after ChatGPT has gained traction among the public faster than any other technology ever released.
So the hype has appeared after LLMs have bridged the gap between human language and computer operations. Which they are impressive at bridging, to be sure.
What makes LLMs so specially scary, beside the speed of their spread?
Is it because LLMs are so general purpose? I don’t think so. The fact you can ask so much to an LLM just makes the kind of tasks you can work on by sitting in the pilot’s seat much broader. That creates more work and learning opportunities than one can imagine. So I do not think this is the answer.
Is it because we love filling up Excel spreadsheets and we are already jealous of the tech which will spare us this dull work? I don’t think this is the reason either. I do this myself and I find it pretty mechanical. My mind often wanders around while I fill spreadsheets up, it does not make me feel particularly human.
Is it because coders love looking for answers to the problems they try to solve by going to StackExchange or opening the online documentation of the language they are writing code in, instead of asking the LLM to provide a way out of their present conundrum? Well, GitHub Copilot has been around for longer than ChatGPT and it has not generated the hype the latter has. Enthusiastic reports from people using it in their organizations have been enthusiastic for longer than ChatGPT’s existence. So that must not be the answer either.
…
The list of questions could go on…
I have my tentative answer, for what it is worth: in the era of instant communication, we are overly identified with our language. More in general, we are identified with words rather than meaning. Also, we value so much the mechanical tasks that LLMs are so good at accomplishing in our place because we are used to ourselves being mechanical.
As I have tried to explain in the previous section, manual work is far from done with. And learning it, as I have very convincingly -I believe- argued, takes so much more than just following along with spoken instructions. There is an ocean of life in the psychic experience of the world which exists underneath the outmost layer which language is. Language is to the psyche what the foam on top of the waves is to the ocean. Yet that’s the thing we mostly identify with. This way, a technology which mimics human language scares us to death to the point we believe we are done.
Mirko Serino and Francesco Panarese
Accedi con Google