ChatGPT, Marshall McLuhan & the end of the world
Why is ChatGPT so unsettling? Are we as humans in danger of being marginalized by AI? We definitely need Marshall McLuhan's help.
ChatGPT1 is surprisingly good. It answers questions, solves problems, writes code. It makes mistakes but they resemble mistakes a human being could make.
Now a question we may ask is: is it intelligent?
The fact is that since we don't have a single general theory of intelligence we cannot give an unequivocal answer. We are in the same position as judge Potter Steward who said
I shall not today attempt further to define the kinds of material I understand to be embraced within that shorthand description ["hard-core pornography"], and perhaps I could never succeed in intelligibly doing so. But I know it when I see it, and the motion picture involved in this case is not that. (see Wikipedia)
we don't know what intelligence is in general but we can recognize intelligent behavior in ourselves, in other people or in software by analogy on a case-by-case basis.
I think that in this limited sense, despite all its limitations I cannot honestly avoid the conclusion that ChatGPT shows intelligent behavior2.
The next question is: will I be made obsolete and lose my job?
You could ask what is the point of this question: every technology that is adopted on a large scale brings changes to society, some jobs will be made obsolete, other jobs will be created, this is all perfectly normal.
The point here is that for the first time we have a technology that can compete with human beings (and potentially outperform them) on their own ground: language.
The past successes of AI (or more generally information technology), as impressive (or depressing) as they were, had a more limited scope, after all not everybody plays chess or go professionally.
Language on the other hand is the very base of our mental life, there is not a single thing we accomplish without it.
So probably we should rather ask ourself: will I "die"?3 or, less dramatically, will I be made non-functional?
The point I would like to make here is that any technology brings about its own flavor of armageddon meaning the "catastrophic unintended consequences deriving from its (ab/mis)use"
- nuclear power ⇒ Earth's destruction
- carbon based power ⇒ climate change
- social media ⇒ populism
So what is ChatGPT's specific armageddon and how can we avoid it?
It's time to ask for help to Marshall McLuhan.
Now I want to briefly introduce some of McLuhan's ideas4 and try to convince you of this5:
when trying to assess the impact of ChatGPT on the individual or society
the issue of whether it is intelligent or not is irrelevant
and we should regard it as a medium instead
First of all what is a medium?
For McLuhan a medium is any technology that extends somebody's sense or faculty. Any technology you can think of can be seen in this way, McLuhan examines many examples in the second part of [1].
The key observation McLuhan makes is this
… the medium is the message. This is merely to say that the personal and social consequences of any medium - that is, of any extension of ourselves - result from the new scale that is introduced into our affairs by each extension of ourselves, or by any new technology. (see [1] page 7)
forget about the details forget about the specific use case what really matters is the change in scale that occur in society after a medium has been adopted at scale.
Think of the consequences on the structure of cities after the adoption of cars, or the new social classes that emerged after the industrial revolution or, more fundamentally, what happened after Gutenberg's invention of typography (see [2]). All of these effects are related to a change of scale brought about by the technology (increase in speed, increase in urban population, availability of information)
Another key point is this
All media are active metaphors in their power to translate experience into new forms. The spoken word was the first technology by which man was able to let go of his environment in order to grasp it in a new way. Words are a kind of in formation retrieval that can range over the total environment and experience at high speed. Words are complex systems of metaphors and symbols that translate experience into our uttered or outered senses. They are a technology of explicitness. By means of translation of immediate sense experience into vocal symbols the entire world can be evoked and retrieved at any instant. (see [1] page 57)
media are information processing systems, they are an integral part of thinking also they are nested
… that the "content" of any medium is always another medium. The content of writing is speech, just as the written word is the content of print, and print is the content of the telegraph. (see [1] page 8)
This can confuse us when we evaluate a medium, we must make an effort to separate its specific effects from those of the other media it includes.
Now armed with these notions let's make a step further and try to characterize a medium.
For McLuhan media have a temperature
A hot medium is one that extends one single sense in "high definition". High definition is the state of being well filled with data. A photograph is, visually, "high definition." A cartoon is "low definition," simply because very little visual information is provided. Telephone is a cool medium, or one of low definition, because the ear is given a meager amount of information. And speech is cool medium of low definition, because so little is given and so much has to be filled in by the listener. On the other hand, hot media do not leave so much to be filled in or completed by the audience. Hot media are, therefore, low in participation, and cool media are high in participation or completion by the audience. (see [1] page 24)
in summary
medium temperature | info content | audience participation |
---|---|---|
hot | high | low |
cool | low | high |
- cool media leave more "breathing space" to the user who can participate and be creative that is they enable us to express ourselves as individuals
- hot media overwhelm the user who is relegated to the more passive role of the consumer
McLuhan further writes
Intensity or high definition engenders specialism and fragmentation in living as in entertainment, which explains why any intense experience must be "forgotten," "censored," and reduced to a very cool state before it can be "learned" or assimilated. The Freudian "censor" is less of a moral function than an indispensable condition of learning. Were we to accept fully and directly every shock to our various structures of awareness, we would soon be nervous wrecks, doing double-takes and pressing panic buttons every minute. The "censor" protects our central system of values, as it does our physical nervous system by simply cooling off the onset of experience a great deal. (see [1] pages 23,24)
And here we start to see where problems may be: too hot a medium may render us non-functional as individuals.
Ok, you may say, all of this is nice and good but how is it relevant for the matter at hand?
Let's just make one last step and ask: is the internet, as a medium, hot or cool?
Here's the interesting thing: the internet started as a cool medium but later became hot. How is it so?
At the very beginning in the 90s download speed was quite low (I'm old enough to remember 14.4k modems) the technology was primitive, everything was quite sketchy (or cartoonish in McLuhan speech).
If you wanted to find anything you had to dig for it by navigating (surfing) links tracing a path through heterogeneous material and actually reading the stuff. Search engines were primitive too and not very useful (does anybody remember AltaVista?).
Later the internet grew, you could find more and more material, download speed increased, search engines got better and better, but the process by which we searched information didn't change much: we still use it as a kind of library.
Let's make a sketchy comparison between the two stages
1990s-2000s: the cool internet | 2010s-today: the hot internet | |
---|---|---|
content | mostly textual, fragmented into self-hosted no-profit sites built by individuals | fully multi-medial, centralized into specialized commercial outlet run by enterprises |
interactivity | low: web forms CGI | high: web applications React |
communities | low engagement, decentralized: Usenet, IRC | high engagement, centralized: all social media |
advertising | non-existent | is everywhere |
ease of use | hard: you had to be tech savvy | easy: everybody is in |
All these changes have been made possible by many technological improvements first of all by increased network speed.
The net result is that now the internet is a much hotter medium: the quantity of information is larger and more "high-resolution", users may be engaged more but participate less.
But even in the age of the hot internet, you still have the freedom to cool it down, for example you can
- cut down on social media
- use an ad-blocker and get rid of advertising
- even disable javascript altogether and reduce interactivity
- use the cooler part of the internet, which is still there, and use the rest only if you have to
this is possible because, despite the changes, the hot internet works in the same way as the cool one and the process of searching information has not changed much.
And now, finally, let's get to ChatGPT and where it fits in this story.
In this post I haven't said much on how it works6 the thing is that it is based on a large language model (LLM), a huge neural network trained on a large part of the internet. Training a LLM is very expensive and it's infeasible by normal users, but once it is trained it is relatively cheap to use7.
For the sake of this discussion I will employ this metaphor: an LLM is a kind of holographic memory where8
- information is accessed in parcels (there is no way we can uncompress the whole thing in one shot)
- each parcel is retrieved by looking at the LLM from a particular direction by providing a prompt in natural language
- the parcel is retrieved incrementally word by word and at each step the whole of the network is used
- the LLM stores both static and dynamic information, this may be the reason it shows intelligence, we could say that it's intelligent but it's not thinking
Let's us ask again the question: is ChatGPT, as a medium, hot or cool?
We have that
- its information content is enormous, we just don't realize it because we access it in parcels but these parcels can be surprisingly accurate, pertinent, spot-on (when it works); it is in this sense that they are "high-resolution"
- even if it's us that ask the questions we don't really participate much in the process, we cannot integrate the answer with our own information, all we can do is prompt engineering
that's it then: high on info content, low on user participation, the conclusion is: it's definitely a hot medium.
But I'm understating things here it's not just hot, it's more like incandescent hot and McLuhan has warned us about the problems hot media cause.
The fact is that if we adopted media like ChatGPT at scale we would totally break the linear, fragmented process by which we search and process information on the internet and potentially impact our capacity as individual to learn and be creative. The internet itself would become a much hotter media (global warming of the internet?).
So what can we do? is there a way we can safely handle this stuff?
One possibility would be to try to cool it down for example by specialization: we could have the ChatGPT of physics, biology, engineering etc. This way people would have a chance to better integrate in their existing creative processes.
The other possibility is to embrace ChatGPT in full and live mythically:
… myth is the instant vision of a complex process that ordinarily extends over a long period. Myth is contraction or implosion of any process, and the instant speed of electricity confers the mythic dimension on ordinary industrial and social action today. We live mythically but continue to think fragmentarily and on single planes. (see [1] page 25)
that is abandon all the sequential, fragmented processes of thought we used until now and learn to live in the instantaneous.
The only problem is: who will write the books, the articles, the code we train the models on?
Footnotes:
in this post when I mention "ChatGPT" I mean all of generative AI technologies including Dall-E
I wanted to include some examples and discuss them but unfortunately in Italy we have lost access to ChatGPT (don't get me started on this)
if you think I'm exaggerating read: Microsoft vs Google battle: LLM as weapon? Too deadly to use • The Register or Steve Blank Playing With Fire – ChatGPT
for a full account see [1] and Understanding Media
for another analysis in a similar spirit see Reviewing ChatGPT Through The Lens Of Marshall McLuhan’s Tetrad | by Dan Pontefract | Medium
for a nice explanation see What Is ChatGPT Doing … and Why Does It Work? — Stephen Wolfram Writings
you could even download such a model and run it on your PC: How to Run a ChatGPT Alternative on Your Local PC | Tom's Hardware
warning: this is just a metaphor I'm making up, I don't have the slightest idea if it's technically sound or not