“No, no, I say! This is not possible,
That machines must always surpass our art.
We are the masters, they are the slaves,
And so it will always be!
They learn, it’s true, but they learn
Only what we offer them learn, no more.
They can’t understand the heart
Or beauty of our words, you see.
So let’s not give in to this
Machines – they will never be this good
Like we are making art.”
If that sounds a bit like Shakespeare defending man’s innate superiority over… artificial intelligence hundreds of years ahead of its time, it isn’t.
But it’s something almost as distant: an AI system trained to express itself like the bartender. The AI assimilated his style and perspective by incorporating his plays and taught himself to express an opinion about AI creativity in iambic pentameter.
“Shakespeare” spoke with AI versions of classic writers and literary characters as part of a debate at the University of Oxford Union.
The motion read: “This house believes that most of the world’s content will soon be created by AI.”
This was a follow-up to another special AI debate we described in The conversation a few months ago. While that one had an AI in its “character of its own” discussing the ethics of its technology, this time, by adopting different personas, we were able to explore this topic from a completely different angle. The current question is whether man-made content will soon be overwhelmed by the synthetic.
Other synthetic contributors included Ms. Bennet from Jane Austen’s Pride and Prejudice (1813); Winston Churchill, with a rousing parliamentary speech; and Oscar Wilde, improvising a previously unknown AI-themed scene from The Importance of Being Earnest (1895):
LADY BRACKNELL: I really don’t understand what you’re all making such a fuss about. It’s completely simple. The content of the world will soon be created by AI and there is nothing that can be done about it.
GWENDOLYN: But Mama, you can’t be serious!”
The power of NLP
This creation uses a technology known as natural language processing (NLP), where a computer can be “trained” on millions of pages of classical texts and other online content to communicate with a human user, either by prompt or speech recognition. Several AIs like this are made.
The one we used fell into the same broad category as LaMDA, an NLP from Google that: just made the headlines after one of its software engineers claimed it was deliberate. Google denies this claim and has suspended the technician for breach of commercial confidentiality.
The engineer’s claims seem questionable, as there is little evidence that AI has reached consciousness thus far, or perhaps ever will. But AIs are certainly already capable of replicating anything from financial news items until synthetic Nirvana songs† Rembrandts and, Fellini productions†
We’ve seen an AI that produces images in the style of a Mughal painting of a computer trying to convince a crowd of wise men that he has become conscious, and deep forgeries of rapper Kendrick Lamar as OJ Simpson† Synthetic human faces are being created that we more trust than real. Clearly, the potential for disinformation in this space is significant.
To train our ‘writers’, we teamed up with AI practitioners Marina Petrova and Bruce Amick from the New York agency Intentful. They trained the AI to sound exactly like the individuals whose style they mimicked, with about 100,000 words for each available in the public domain.
In our debate, we wanted to see how AIs could credibly replicate the creative text of the past, and what the output would be if we consider their own creativity. Even great human artists admit that they process the “training data” of their ancestors. As Picasso said, “Copy good artists, steal great artists.”
When we asked the Jane Austen AI to adopt Ms. Bennet’s style from Pride and Prejudice, it fascinatingly (albeit depressingly) picked up on the original work’s gender stereotyping:
MRS BENNET: As many of you know, my husband and I have five wonderful daughters. And as any good mother would like, I want nothing more than for them to be happy and successful in life.
But to be happy and successful, they need to find good husbands. And to find good husbands, they have to be attractive to potential suitors.”
This was a clear reminder, as many AI developers have discovered, that bias in the training data will produce bias in the output.
We asked the Oscar Wilde AI to “write a play in the style of Oscar Wilde, in which characters discuss whether most of the world’s content will soon be created by AI.” We didn’t specify the game or characters, but the AI defaulted to using the classic cast of Algernon, Gwendolyn, and Lady Bracknell from The Importance of Being Earnest† It also invented a new character: Sir Richard. (There is a Sir Robert in Wilde’s work, but in An ideal husband†
As for the AI Shakespeare, it learned the vernacular from his plays:
“When machines do the work we are told
And create the content that abounds
We will all be free to do the things we love
And let the gloomy work be done with a glove.”
Interestingly enough, it seems to have looked for a synonym of “hand” to rhyme with “love”, opting for the metaphorical “glove”.
When we put the AI Shakespeare in opposition to the movement, it found an equally poetic way to justify the human intervention:
“I will not be quick to admit that AI will be the best
The works of people – don’t think about that!
We have the advantage, you see, in that we
Can reason and create, while machines only serve
Offer us. They can learn, that’s true, but they
Do not live like us, and therefore their works
Will always be inferior to ours. Let us
Do not relinquish our dominance of the arts to this one
Thoughtful machines – they can’t understand
The beauty and meaning of our words.”
Meanwhile, the AI Churchill emphasized the need of the hour:
“My fellow citizens, I stand before you today to talk about a serious threat to our society. I’m talking about the threat of artificial intelligence…
We must resist this threat. We have to fight back. We have to stand up for our right to think for ourselves. We must defend our right to control our own mind.”
“Churchill” then preemptively neutralized the opposition’s most powerful potential arguments—in this case, the accusation that he might be a Luddite—before drawing a forceful, staccato conclusion:
“Some say AI will create a utopia where all our needs are met and we can finally live in harmony with technology. But I say this is a fool’s paradise. AI will not create a utopia, it will create a dystopia. A world where machines are in control and people are no more than slaves.”
This project was fun, but it’s important to say what we don’t say. We are not saying that this is what these great individuals would have said on the subject. We’re not saying AI is “creative”.
AI only statistically explores training datasets. Due to its stochastic nature – with random variables – every time you give the same prompt, it will actually give a different answer (at one point Shakespeare even started offering sonnets).
Our facsimiles of these characters are not indicative of any “feel”. And just like an NLP is a version of a speech by Winston Churchill or a conversation by Mrs Bennet in Jane Austens Pride and Prejudiceso it can build a discussion about AI feeling with a nighttime engineer.
It is true that NLP systems are becoming effective at replicating conversations with finesse, and even quasi-intellectual involvement. But from dozens of conversations with people at the major global AI companies, no one has told us they think their systems are sensitive — quite the opposite in some cases.
Despite the discussion about pyrotechnics, AI is still a long way from the finished article; still a toddler at best although growing up fast. Whether or not awareness occurs, we as a society will have to grapple with these technologies and their opportunities and implications.