Drake vs. Generative Artificial Intelligence

You may have heard that over the weekend a TikTok user uploaded a Fake Drake song called “Heart on My Sleeve.”?The new song was streamed over 20 million times .?People who listened to the new song marveled at how close it was to the real Drake.

To fans, it's sounded like another hit.?But the song was just another recent example of the advances in generative artificial intelligence.?The song that the TikTok user uploaded was made by training the system on Drake's music.?Then they simply generated a new version based on his existing songs.

In a sense, it was using the same technology that you see in ChatGPT and DALL-E 2.?Except instead of training on millions of songs, it limited itself to only Drake's music.

The Universal Music Group or UMG quickly responded to the new song by saying that

“Training of generative AI using our artists’ music (which represents both a breach of our agreements and a violation of copyright law) as well as the availability of infringing content created with generative AI….”

Streaming platforms then dutifully pulled the song from their catalogs.?So far, the track has been pulled from Apple Music, Deezer, and TIDAL, Spotify, SoundCloud, and YouTube.

UMG’s response to this song is understandable. They want to restrict TikTok users from sharing viral songs produced using voice recognition systems trained with their artists' voices.?But did this TikTok user violate copyright law??It’s not as clear as UMG says.?Why is this a violation of copyright law and not ChatGPT or DALL-E 2?

ChatGPT uses something called Generative Pre-trained Transformer (GPT) and Large Language Model (LLM).?These models work by processing billions of text compilations online.?A lot of these news articles and stories are also protected by copyright. OpenAI says it's Fair Use .

The same with DALL-E 2.?This also uses GPT, but instead of using text it trains itself on images it finds online.?Again, a lot of these images are protected by copyright.

It seems the only difference between these systems and the TikTok uploader is just a matter of scale.?OpenAI, the company responsible for ChatGPT and DALL-E 2, trains their systems off the work of millions of different writers and illustrators.?While Fake Drake was only trained on one artist.

So will the difference between whether you violated someone's copyright just come down to the size of the system's training data?

It makes you wonder what would have happened if the TikTok user hadn't been so brazen.?Imagine if the system trained the model of off thousands of different singers.?Then they used the system to generate a song that sounded a little bit like Drake but was also a mixture of different artists.?Then they simply titled the song “Heart on My Sleeve,” by SynthFlow (I generated this AI singer’s name using ChatGPT).

This is similar to OpenAI's approach, but instead of UMG's lawyers dealing with a single user on TikTok, they would be facing the rapidly growing generative AI industry.

But UMG has a point. They see as a struggle as two competing forces on one “the side of artists, fans and human creative expression, or on the side of deep fakes, fraud and denying artists their due compensation.” The statement was directed at one lone person on TikTok, but it's really a criticism of generative AI.

The knowledge and talent death spiral

This challenge isn't as new as it might seem. Back in 2016 Stanford professor Andrew Ng said that a

“highly-trained and specialized radiologist may now be in greater danger of being replaced by a machine than his own executive assistant.”?

The famous University of Toronto AI professor Geoffrey Hinton said

“People should stop training radiologists now. It’s just completely obvious that within 5 years, deep learning is going to do better than radiologists.”

Then Andrew Ng founded a company that tried to do just that.?He trained an artificial neural network with all the data from Stanford University’s hospital.?The system got very good at identifying anomalies in X-rays.?In fact, it was very close to its human counterparts.

But this new system also presented a challenge.?Let's say that medical students took Geoffrey Hinton's advice and decided not to become radiologists.?The AI system they created trained of off the images correctly classified by a radiologist.?This data was created by highly skilled professional.?In machine learning, this is called labeled data.

So how does the system continue to get better as it replaces human radiologists??There'll be no new radiologists to classify future X-rays.?That would be fine if skills and knowledge were static. But X-ray machines get better all the time.?In fact, Andrew Ng said the system became much worse once they went to a hospital with different equipment.

This is the same problem that UMG alluded to with the Drake song. If radiologists, writers, and musicians can't get "due compensation," then who will create future work to train future AI systems??These new generative AI systems can create a knowledge and talent death spiral that cuts off the very source that these systems need to grow.

This seems like just the beginning in a struggle between several large industries.

You have movie, music and media producers that base their business on creating content with professional talent.?Plus, you have hospitals, universities and pharmaceutical companies that rely on well-trained professionals.?Now you have technology companies that can generate new work off this existing content.?

This last weekend it was just one lone TikTok user, but in the future, you’ll see giant companies struggling to protect their industries.

In the future, you might see Microsoft as the largest movie company or Amazon as the largest music producer.?It would have seemed bizarre a decade ago, but imagine informing yourself back then that Amazon had become the biggest video streaming platform and Apple was earning Academy Awards. Then go back twenty years and explain how Google generates more ad revenue than all the newspapers combined.

It shows that companies that create great content often lose out to companies with superior information systems.

If you interested in this topic I have a course that is part of a three part series on Data Ethics. Data Ethics: Managing Your Private Customer Data.

This is my weekly newsletter that I call The Deep End because I want to go deeper than results you’ll see from searches or ChatGPT. Each week I’ll go deep to explain a topic that’s relevant to people who work with technology. I’ll be posting about artificial intelligence, data science, data ethics and agile software development.?

This newsletter is 100% human written ?? (aside from a quick run through grammar and spell check).