... even when they're (supposed to be) on your side.
Ted Gioia has the breaking news:
Authors win a big lawsuit against AI—but the judge says they may not be able to trust their own lawyers.
He explains that the high-tech plagiarism modus of these "large-learning models" (LLMs) simply means that the models are "trained" on thousands of books, and millions of articles and blog posts. All written by an actual person. A person holding copyright in that work.
So when authors, in a class action, won an ironclad case again AI company Anthropic for violating their copyrights ...
"some thought that this might result in “more than a trillion dollars in damages.” That would put Anthropic in bankruptcy and send a message to the entire AI industry: Don’t mess with creators!
Yay!
But ...
Instead the lawyers negotiated a quick deal for $1.5 billion—and Anthropic didn’t even need to admit wrongdoing. But the penalty was so light that the judge has refused to accept it. Instead he expresses concern that the settlement will be forced “down the throat of authors.”
How is this possible? Their own lawyers negotiated the deal.
But listen to the judge. He admits that class members often “get the shaft” in situations like this. And he adds: “I have an uneasy feeling about hangers-on with all this money on the table.”
Simply put, lawyers want their commission more than they care about their clients. Or their case.
This is the sad reality of copyright litigation to protect human creators. My copyrights as an author have been violated and I don’t want a cash settlement—I want the stealing stopped. I want a Napster-style shutdown, and there’s legal precedent to support this. But what lawyer can I trust? They make money on a cash settlement, not on stopping AI use of my book.
Expect to see similar settlements in music copyrights. A few people will get a nice payday, but nothing else will change.
7 comments:
AI is trained on the works of authors and creators. That is, it reads them. Then it is able to use what it has learned and retained to answer user queries and execute tasks for users. The allegation seems to be that by relying on the work of others to be trained and to learn the AI is breaching their copyright. Right?
"Trained" means copied.
"Reads" means copied.
"Learned" means copied.
"Answer" means regurgitate in a likely and felicitous manner.
"Relying on the work of others" means copied.
"AI" is high-tech plagiarism.
"Allegation" means "proven in a court of law."
So, let's fix your reply, shall we?
=> High-tech plagiarism copies the works of authors and creators. That is, it copies them. Then it is able to use what it has copied and kept and retained to regurgitate the copied work in a likely and felicitous manner for users. The charge proven in a court of law is that by relying on the work of others to be copied and kept the high-tech plagiarism apps are breaching their copyright. Right?
You can answer your own question now.
Peter Cresswell
Yes, I see the tone of your response is unfriendly, perhaps even rude.
As a person now active in training and education I noticed that I am trained on the works of many authors and creators. All we students are. Teaching professionals are as well. We read them and yes, we copy them. That is expected- demanded even. We are required to be able to quote them, recite them, summarise them, manipulate them, reform them, apply them, read them out aloud, refer to them, use them, apply them in various ways when asked.... just as AI or expert systems do. Those programmes are in essence machine versions of what we all do, you included.
Recently you used Grok and, by your own argument, plagiarised the works of other authors and creators, breaching their copyright. Not only that, you published it on this very forum, word for word. Hypocrisy!
==
A "charge proven in a court of law" is not necessarily a fact. It may be. It may not be. Many, many bizarre and even corrupt things can happen in courts.
@Anonymous 23:29:03: It’s not intentionally rude, although I have very little respect for a person unwilling to put their name to their opinion. Still, I apologise.
“Training” for a human being is not copying. For a human being, training means *understanding.* Not just to know, but to know *why.* If we “copy” our reading it’s to quote, in fair use, as part of a demonstration of understanding, or to make an argument that reveals our understanding
Whereas “training” for a machine has none of that, as I’m sure you know. ‘Training” here is simply recognising that some words used by some authors seem to often follow certain other words, and that certain sentences used by authors might often follow other similar sentences.
The intention of these so-called Large Learning Models is to copy and to regurgitate.
That is not what either of us do.
* * * *
I did mention Grok in a previous reply. If there is one non-copyright benefit of these models it’s the ability to quickly check references, as in the material referenced in that reply. But that’s not a copyright issue, it’s a network issue (and those references it spews up still need to be checked against their originals for “hallucinations.”)
* * * *
Many bizarre things do happen in court. But once the court has made a finding on a claim, that claim is no longer an allegation.
Let’s address this thoughtfully. You describe AI training as “recognizing that some words used by some authors seem to often follow certain other words, and that certain sentences used by authors might often follow other similar sentences,” then suggest this reflects an intent to copy and regurgitate. This interpretation may be a misunderstanding, as the two ideas don’t necessarily connect.
Recognizing word or sentence patterns in AI training is about improving the assignment of probabilities and weightings, not copying or regurgitating. Terms like “seem to” and “might often” point to the focus on probabilities, which helps AI select the most appropriate next word. Linking pattern recognition to an intent to copy may not fully capture the purpose of AI training, which is about enabling effective language generation.
==
You’ve been using AI. Isn’t this a deployment of the “copying” you attributed to AI training? Let’s explore this idea further.
There’s a saying: when facts support your position, focus on facts; when the law supports you, focus on the law; when neither does, emphasize rhetoric. Your argument leans heavily on copyright law, but a deeper understanding of legal systems and processes might strengthen your case. Focusing on facts could be more effective, though the facts here present challenges.
You’ve expressed concerns about AI while also using it, noting no copyright issues in your actions. The AI in question was trained on a diverse range of texts—novels, short stories, non-fiction, articles, websites, and public domain works—allowing it to understand narrative styles and factual content.
Setting aside copyright for now, the key point is that you’ve used a tool you describe as designed to copy and regurgitate. This could seem inconsistent with your stated concerns, similar to benefiting from something you question ethically. However, this doesn’t necessarily make your actions wrong—it highlights a possible gap between your stated views and your actions, which is worth considering.
==
It’s worth noting that a “charge proven in a court of law” isn’t always a definitive fact. Terms like allegations, claims, or findings don’t always align with reality, which is why court rulings can be appealed, overturned, or nullified. Without being directly involved in a case, it’s hard to know all the details. Relying on reports raises questions about their accuracy and completeness, which can make forming a firm conclusion challenging.
When a case is decided, the losing party may appeal, which is costly and time-consuming. After the appeal is decided the winning litigant might seek costs, adding to the loser’s burden. Appeals can escalate up to higher courts, even the US Supreme Court, increasing expenses and uncertainty. Attorneys for both sides weigh these risks. For plaintiffs, settling early might avoid the uncertainty of an appeal where the defendant could prevail, leaving them with losses. Defendants’ counsel faces similar considerations—settling might be less costly than an appeal with no guaranteed outcome. For both, a settlement provides certainty and limits risk.
Judges also play a complex role. Questions about their impartiality, jurisdiction, or past rulings may arise, but without full context, it’s hard to assess. Attorneys, however, likely have insight into these factors and act accordingly.
==
AI adoption is growing rapidly and becoming widespread. Copyright law may evolve to accommodate AI, and unrestricted use of public domain and copyrighted material is already occurring. Jurisdictions that limit AI may fall behind those that embrace it, creating a choice: adapt or lag.
AI poses challenges, potentially displacing many white-collar jobs, but it also offers opportunities for those who act quickly. Consider which side you’re on. Additionally, AI’s energy demands are significant—already matching household electricity use in states like Wisconsin. Meeting future needs will likely require new power sources, such as hydrocarbons or nuclear. Expanding nuclear energy would involve major regulatory and political changes, which won’t be simple. Threat and opportunity can occur simultaneously. Now is one of those times when exactly that is happening.
Eric Weinstein, when asked about preparing for the future, suggested that much of what we’re accustomed to will change or disappear. I share this view. He recommends developing practical, cross-disciplinary skills and preparing for uncertainty. A respected finance analyst recently noted that retaining even 50% of your wealth and purchasing power over the next seven years would be a significant achievement. The world is undergoing profound change, and everything is subject to restructuring. Holding rigid views may not serve you well in this environment. While AI might not be your preference, adapting to its presence will likely be necessary.
Funny.
Anonymous commenter pastes comment "created" by AI about AI.
How meta.
It's a metaphor for all of this slop: sending stuff you didn't say, to people who haven't asked for it, which nobody ever reads.
Post a Comment