On The Shoulders of Giants

July 16th 2025

“If I have seen further, it is by standing on the shoulders of giants,” attributed to Isaac Newton, has been invoked by tech companies and AI researchers to justify using copyrighted works for training AI systems.

This argument – broadly that human endeavor builds upon all human endeavor preceding it – has now moved from Scientific Revolution era Europe to current day Silicon Valley philosophy and seems to have been echo generated in the modern-day courtroom in Bartz v. Anthropic (1). In the publishers’ case against the LLM company, Judge William Alsup of the United States District Court for the Northern District of California said, “Like any reader aspiring to be a writer, Anthropic’s LLMs trained upon works not to race ahead and replicate or supplant them — but to turn a hard corner and create something different.” His point aimed at transformative use but the philosophy is perhaps identical: it’s acceptable for a machine, like a human, to learn from preceding works.

Well, sure, but even ignoring that GenAI systems have the ability to memorise and replicate vast amounts of data – see for example blocks of word-for-word text LLM’s have repeatedly been caught spitting out in NY Times v Microsoft (2) or the Darth Vader and Minions characters shown in the complaint in Disney & Universal v Midjourney (3) – whereas humans don’t have the perfect recall of millions of works, the argument is flawed in other key ways.

 

Money, money, money

Humans typically have to pay a price to consume copyright protected work. That can be the cost of a book or entry to a museum or the price of a cinema ticket or subscription to HBO or a Spotify subscription. More indirectly we can even consider the collective cost through taxation for a library or state museum. However you look at it, the individual creator has more often than not incurred a cost to access preceding work – imperfectly and generally speaking. And this compensation goes to the creators. Granted not always, and not always in full, and not always in proportion with the merit of the creation. However, these attempts at, and flawed system of, compensation are a far cry from GenAI models which have often resorted to wholescale lifting without payment of copyright protected work so that they can be trained.

 

Give me that green light

Secondly, the works we might consume simply moving through life as humans are (not always but mostly) available to us following the creator’s consent. I’m not in any way arguing that we live in a perfect society where all creators operate without being subject to power imbalances and offer fully informed consent to each specific and prescribed use of their art, before walking off into the perfect Pissarro sunset, but again this is easy to contrast with GenAI developers (not always but sometimes) ripping through pirated caches of online books or circumventing paywalls and security measures to obtain what they need.

 

Roll the credits

Thirdly, I mentioned Pissarro above. The artist’s name is burned into my brain because it’s on a plaque next to his paintings in Tate Britain where my mother forced me to visit as a child. I, mostly human, walk through the world and learn and am influenced by copyright protected works and the creator is usually credited. A journalist’s name is under the byline, the director of photography is listed in the film’s credits, “Jonas Brothers” is on the concert tickets I don’t buy, etcetera. I live a few miles from Hollywood where the classic novel What Makes Sammy Run by Budd Schulberg tells me that writers have been shafting each other and stealing screenplays for nearly 100 years. So as with compensation and consent again the credit system is imperfect. But What Makes Sammy Run is so poignant because the writer artfully details how poorly Sammy Glick treats people; one particularly chicanerous move Sammy makes is stealing the character Julian Blumberg’s screenplay, making superficial or no changes, and calling it his own. The reader knows this is abhorrent, and this makes Sammy terribly unscrupulous, which is the point of the book. So, if lack of credit is enough of a problem to serve as a plot point in a novel, why is it not enough for legislators or the courts to consider GenAI systems’ MO pretty…terribly unscrupulous?

So, ok, as a human I can absorb the preceding works of the world, learn from them, stand on the shoulders of giants in creating something new, but those giants I stand on have been compensated for their art, have controlled their art and have been credited for their art, again imperfectly but the principles have stood. Until now, it seems.

 

What is my life?

The way AI learns is compared with the way humans learn, as in Bartz v. Anthropic. This is natural given we are our most obvious point of reference, and we anthropomorphize as a general rule in order to make sense of uncertain environments. But AI developers seem to lean into this with a flawed and forced linearity. They don’t differentiate or ask whether our learning process – the human experience – is something of inherent value itself. From birth to death humans typically absorb information through touch, taste, smell, hearing, and sight. We may associate something we see with a taste or smell, with synesthesia some people consider sounds to have colours, and so on. We are informed by these experiences with a side of genetics. I don’t want to preach we are all very special and unique but even identical twins have differing tastes and personalities. There are infinite possibilities in what we might create, as with GenAI, but is there an unknown, an X factor, or spark of humanity, whatever that is, in getting there? Meanwhile, so far at least, GenAI systems are fed, bluntly speaking, through one plane of reference (language, image, sound etc) and are trained to predict or reason accordingly.

Also, our human experience even during moments of transformative reinterpretation or inspiration, has the factor of time baked into it along with the full stop of mortality. We have (hopefully) a lifetime of learning while GenAI systems can contract that learning into a comparatively instantaneous timeframe. We have to daub our paint onto the canvas before we die. Is the human lens and all work created through it coloured by it being finite while, so far at least, GenAI systems have no concept of what it can mean to die? Is there a layer of beauty in the point that we all must come to an end? And again, without getting too far into questions of the soul, should this matter if the output is sometimes or often the same? I really don’t know the answer to this.

What I do know, though, is that the writer Budd Schulberg considered stealing someone’s work to be unfair when he published his book What Makes Sammy Run in 1941. Today there are arguments thrown at us daily about why this no longer applies, from the need to win the geopolitical AI race, to the pointlessness of resisting the onward march of technology, the benefits to society from embracing AI or simply, as above, that AI is only doing what we humans do ourselves. I can’t ask the now deceased author what he thinks of the strengths of these arguments in reframing the taking of someone else’s work as GenAI systems are doing. However, I strongly suspect Sammy’s foil, the decent and good Al Manheim, would think it’s not on; that he would believe of course creatives should be compensated, control, and be credited for their work.

I think Manheim would shake his head, perplexed, and ask why we are now supposed to collectively pretend that the foundational pillar of creative ownership has disappeared.

 

References:
(1) Bartz et al. v. Anthropic PBC, No. 3:24-cv-05417 (N.D. Cal.)
(2) The New York Times Company v. Microsoft Corporation, 1:23-cv-11195, (S.D.N.Y.)
(3) Disney Enterprises Inc. v. Midjourney Inc., 2:25-cv-05275, (C.D. Cal.)