AI Efficiency vs. Legal Rights: A Critical Look
- AI copyright infringement. Lawsuits highlight the clash between innovation and copyright, urging a legal reevaluation.
- AI ethical issues. Scaling personal outreach into spam reveals the fine line between efficiency and ethical misuse.
- Permission principle. Consent is key in distinguishing between innovation and intrusion in digital communication.
The New York Times and many other content creators and authors are suing generative AI pioneer OpenAI and Microsoft, claiming they trained their ChatGPT product on copyrighted material without permission nor compensation. These AI copyright infringement lawsuits have many people asking some variation on the question:
Should we penalize a chatbot for doing what all human beings do — that is, assimilating information to inform responses — just because it does so more efficiently?
I’m not a lawyer or legal scholar, but it seems like the answer is almost assuredly yes. That’s because superhuman efficiency and scale routinely lead to abuse when it comes to AI copyright infringement and other issues. In fact, it’s easy to find compelling examples without even leaving the world of marketing.
Let’s discuss a few AI ethical issues.
Cold Email vs. Spam
Sending one email to someone who hasn’t asked to hear from you is sending a cold email. That’s legal everywhere in the world. That’s just one person taking their precious time to reach out to another person.
However, if you do that same thing at scale, then you’re a spammer. Traditionally, that means you’ve used software to use a small amount of your time to send the same (or largely the same) message to lots of people, collectively wasting a lot of their time. Spam is a blight on the modern world, which is why the European Union, Canada, and many other nations outlaw spam.
Operating under the highly permissive and very antiquated CAN-SPAM Act of 2003, the US sadly hasn’t made spamming illegal yet. However, inbox providers aggressively block spam and consumers aggressively report unwanted email as spam.
In this new world of generative AI, I would argue that using generative AI to substantially automate the writing of cold emails is also spam, because you’ve similarly drastically increased the scale of your cold emailing.
Related Article: Is the Anti-Spam Law CAN-SPAM Now Meaningless?
Cold Calls vs. Spam Robocalls
When a person calls someone they don’t know personally and who hasn’t given them permission to call, that’s a cold call, which like cold email is perfectly legal. However, when someone uses a little of their time to record a message and then uses an automated call system to push that message out to tons of people on behalf of an organization that doesn’t have permission to call those recipients, that’s a spam robocall.
Robocalls are among the top complaints to the Federal Communications Commission. And robocalls that try to sell you something are illegal, unless the organization has written permission from the phone number owner to call them with such offers, according to the Federal Trade Commission.
Generative AI will surely take robocalls to a new level of irritation and deceptiveness. A robocall impersonating Joe Biden that tried to discourage New Hampshire democrats from participating in the state’s primary last month is a hint of what’s likely to come.
Related Article: Machine Learning and Generative AI in Marketing: Critical Differences
Individual Surveillance vs. Mass Spying
To depart from the marketing world for just a moment, most people don’t have a problem with individual surveillance, where a law enforcement officer shadows a person who’s suspected of wrongdoing to collect potential evidence. Limited personnel ensure that this tool is used with considerable discipline and prioritizes people who are the most likely to commit a serious crime.
However, when machines engage in similar activities, it becomes mass surveillance, where everyone is presumed to be guilty of something until proved otherwise. Less than a quarter of Americans approve of their government spying on them, according to Amnesty International. And generative AI will turn mass surveillance into mass spying, where AI helps comb through all of that data for potential wrongdoing.
To bring things back to marketing, privacy has been a huge issue in the digital marketing space for years, and there have been huge steps taken. Third-party cookies have been sunset by most browsers, with Google’s Chrome set to finally follow suit this year. Apple has launched Mail Privacy Protection, App Tracking Transparency, and Link Tracking Protection. And all of that is in the wake of the EU passing GDPR.
Related Article: Email Marketing’s Increasing Role as Third-Party Cookies Disappear
The Critical Value of Permission When It Comes to AI Ethical Issues
In all three of those cases, the essential element in turning something pernicious into something acceptable is permission from the person whose time you’re taking. We have clients that regularly email millions of people, and it’s fine because they secured permission from all of those email address owners. Robocalls are similarly fine when people opt in to get them. And tracking web, app and email activity are OK with permission, helping brands send their customers the relevant messages they expect.
And, of course, the lack of permission is core to the lawsuits against OpenAI by The New York Times and others. With OpenAI admitting that it’s “impossible to train today’s leading AI models without using copyrighted materials,” the discussion seems to have shifted to attacking overly generous copyright laws.
For example, Yann LeCun, Meta’s chief AI scientist, argued that “most books” — those that don’t generate “significant money” — should be “freely available to download” because of the good it would do for society. (And by “society,” he seems to mean generative AI and other big tech companies.)
While the terms of copyrights are absolutely bloated (thanks, Disney), experts also acknowledge that copyright laws chiefly benefit corporations rather than individuals, who rarely have the legal muscle to defend their rights. In their own way, it’s hard not to see OpenAI, Meta and other tech companies as perpetuating that imbalance.
But given that there are powerful corporate interests in keeping copyright laws strong, generative AI companies are likely to argue that their models, while based on copyrighted works, are “transformative.” That is, their outputs create something distinctly new and entirely different from their models’ training materials.
Related Article: Generative AI for Email Personalization: A Hallucination Wrapped in Confusion
AI Copyright Infringement: Navigating Legal Complexities of Generative AI
Of course, hardly a day goes by without compelling evidence that generative AI models are routinely infringing on or even directly regurgitating copyrighted material. Even so, this is likely a winning strategy, given the courts’ poor understanding of technology and precedents like Authors Guild v. Google, where Google was ultimately allowed to scan books without the consent of authors nor the need to compensate them for their Google Books product.
That lawsuit took around a decade to be resolved. Imagine the state of generative AI in a decade — the number of users, the number of products it’s embedded in. And by then, perhaps the models will be much more transformative, rendering past infringements moot in the eyes of the court.
If that comes to pass, then the next step would likely be to get nations to grant copyright protections to content created by generative AI. Currently, the US and most other nations do not. Under current laws, transparency around generative AI usage is paramount, because usage forfeits copyright protections at a certain point.
Related Article: Is Your Brand Trusted Significantly More Than ChatGPT?
AI Copyright Infringement: The Value of Human Effort
The issue of whether AI-generated works should be protected just like human-generated works gets back to the core question at issue:
Are human-accomplished tasks and machine-accomplished tasks the same?
If generative AI works were seen as equal to human works in the eyes of the law, it would open the door to mass-scale copyright trolls, who would crank out and publish AI-generated images, storylines, songs, videos and other content and then sue artists, publishers, studios and others for copyright infringement after they spent weeks or months or even years working on a piece.
A Different Plane
Again, the truth of the matter seems clear: Tasks performed by humans are on a different plane than tasks performed by machines, whether they’re generative AI or robots or whatever. They’re just fundamentally different, yet so much of the rhetoric around generative AI seeks to put humans and machines on the same level.
Mostly, the argument is that these machines are acting and performing tasks like a human would, so the machines should be treated like people. That’s the core of the argument that some folks are making in regard to the New York Times suit against OpenAI and Microsoft. But mimicry isn’t humanity. Even if we give them people-names like Alexa, Claude and Jasper, they’re still just machines — devoid of emotions, dreams and families.
There’s also an argument out there — thankfully much more rarely seen — that people are just organic machines. Students of history should be alarmed by such statements that seek to dehumanize people.
But to pull things back from the edge of dystopia, our laws currently give great latitude to human-spent time, whether it’s spent calling or emailing strangers or spent writing a novel or painting a picture. Whether it’s deemed a success or failure by others, human-effort is always an achievement.
Just a Result
On the other hand, machine-effort is just a result. Like email-sending platforms and robocall software, generative AI engines are a force multiplier at such a scale that abuse is all but guaranteed. For instance, with 45% of all email worldwide being spam, it’s only through strong enforcement of permission by inbox providers that the tiniest fraction of that actually lands in people’s junk folders and inboxes.
Related Article: 7 Factors That Determine Email Deliverability
While the email industry has strong protections from abuse because the various players have clear loyalties, the same cannot be said of the generative AI industry. For example, inbox providers serve consumers and stand as a clear check on the desires of brands (and spammers), while email service providers operate in between those two parties.
In contrast, the social networks (e.g., Facebook, X), search engines (e.g., Google, Microsoft) and other players you’d expect to be a check on generative AI all have alliances with generative AI vendors or operate their own generative AI platforms. For example, Google’s recent “helpful content” update, which removed the requirement that content be written by people, caused some publishers to experience traffic drops of up to 70%.
This deep-seated conflict of interest makes self-regulation all but impossible.
What Protections Are Left?
If generative vendors are unwilling to secure permission from copyright holders for their training material … If current copyright laws ultimately prove unprotective … And if social media and broader content ecosystems are incapable of self-regulation … Where does this leave us?
New AI copyright infringement laws and regulations seem the only way — which is a frightening prospect since Congress struggles mightily to regulate any kind of technology and takes forever to come to a consensus on watered-down measures. However, I predict that more generative AI-powered deepfakes like the one deployed ahead of the New Hampshire primary will make it clear by the end of this election cycle that laws are urgently needed.
When they finally do write these new AI copyright infringement laws, let’s just hope our leaders can tell the difference between people and machines.
Learn how you can join our contributor community.