Radical Reads: The Algorithm That Could Take Us Inside Shakespeare’s Mind

Nick Frosst, Co-founder of Cohere
    .   

Cohere Shakespeare

 

Image Source: The New York Times; Cohere; Radical Ventures

Editor’s note: Today’s issue of Radical Reads is guest-authored by Nick Frosst, Co-founder of Radical Ventures portfolio company, Cohere.

 

This week The New York Times featured a collaboration between Cohere and author Stephen Marche in pursuit of some Shakespearean detective work. Stephen, a Shakespeare expert, used Cohere’s platform to determine the authenticity of various pieces supposedly written by Shakespeare. You can read the piece here.

 

It turns out that the bibliographic practices of the publishing industry were not so great back in Shakespeare’s day. As a result, many versions of the bard’s works exist today. But determining the authenticity and accuracy of those versions remains a subject of debate amongst scholars.

 

For example, there are three versions of the famous ‘to be or not to be speech’ including a version that starts like this:

 

       To be, or not to be, ay there’s the point,

       To Die, to sleep, is that all? Aye all:

       No, to sleep, to dream, aye marry there it goes.

 

It’s a far cry from the iconic version. But how can we determine if he actually wrote it?

 

Stephen used the Cohere API to create a custom model by fine-tuning our largest language model on all the text he is certain that Shakespeare wrote. He was able to calculate the probability of each of various versions of the “To be or not to be” speech under the probability distribution parameterized by the model. 

 

Using Cohere, he was able to confirm that the most popular phrasing (“to be or not to be, that is the question”) is the version most likely to be written by Shakespeare. Or at least it is the version that a large language model fine-tuned on only Shakespeare thinks is the most likely.

 

Determining authorship through probability analysis with a fine-tuned language model is not a common approach in academia, but it confirmed the suspicions of a Shakespearean scholar. Importantly, this kind of research is now accessible to any academic. Thanks to the Cohere API, Stephen was able to conduct this research including the fine-tuning and the analysis without any coding expertise. 

 

5 Noteworthy AI and Deep Tech Articles: week of November 29, 2021

 

1) How Peter Jackson used AI to strip out the guitars and uncover The Beatles hidden studio conversations on Get Back (Guitar.com)

The Get Back documentary series leverages AI to uncover private conversations between the members of The Beatles during the recording of Abbey Road and their final album release, Let It Be. The exchanges are drawn from 100+ hours of audio and 60 hours of video footage recorded by Michael Lindsay-Hogg for the 1970 documentary Let It Be. The AI has been particularly useful in reducing George and John’s guitars which they intentionally played to obscure private conversations. The custom-built AI was taught the sounds of various instruments and the human voice to isolate the desired sound in the recording – a concept known as ‘demixing.’  

 

2) How AI will change what happens when you see a doctor (The Globe and Mail)

“AI and machine learning represent the most promising technology that can transform the current medical practice and therapy designs.” The Royal College of Physicians and Surgeons of Canada predicts sweeping changes will come sooner than later, and no area of medicine will be untouched by AI. There is potential for Al to relieve the workload on doctors by streamlining mundane tasks like image analysis, medication reconciliation, and note-taking. Increased data access can support rapid and accurate clinical predictions that benefit doctors and patients, making diagnosis more straightforward. AI is already enabling rapid drug development and discovery platforms such as that created by Radical Ventures portfolio company Genesis Therapeutics, where a breakthrough deep learning platform and proven biotech leaders work together to reinvent drug research and development.  

 

3) First draft of the Recommendation on the Ethics of Artificial Intelligence (UNESCO)

All 193 Member states of the UN Educational, Scientific and Cultural Organization (UNESCO) adopted a first-of-its-kind recommendation on AI ethics. At the heart of the text is a warning to governments to avoid dangerous applications of the technology that may threaten civil rights. Recommendations also include providing individuals more transparency, agency, and control over their data. The draft explicitly bans the use of AI systems for social scoring and mass surveillance, emphasizing that AI actors should favour data, energy and resource-efficient methods that will help ensure that AI becomes a more prominent tool in the fight against climate change. UNESCO is the first international organization to get Beijing to sign onto principles that call for the end of pervasive mass surveillance powered by AI. 

 

4) Robot tackles the puzzles of Pompeii (The Times – subscription required)

Can AI solve the world’s most intricate jigsaw puzzle? The frescoes of Pompeii were destroyed first by the eruption of Vesuvius in 79AD and further damaged in bombings during World War II. The thousands of delicate shards remaining are extraordinarily fragile and likely impossible to reassemble manually. Machines equipped with robotic arms can scan these archaeological fragments and create a 3D digital model. The machine can identify nuances in the structure and paint and organize the related pieces to reveal a complete picture of the Roman art without touching the artifacts.


5) New deep learning method adds 301 planets to Kepler’s total count (NASA)

 

ExoMiner is a new deep neural network that leverages NASA’s Supercomputer, Pleiades. It can distinguish exoplanets from imposters or false positives. The neural network supplements human experts and learns from past confirmed exoplanets and false-positive cases and applies the ‘knowledge’ to data gathered by NASA’s Kepler spacecraft and K2, its follow-on mission. The neural network uses the same techniques as its human counterparts and is not a black box – critical features in the data used for identification can be shown. ExoMiner has validated 301 planets with the Kepler Science Office that, until now, have been unverifiable by human or machine.

 

 

Radical Reads: 10 AI predictions for 2022

Radical Ventures Partner Rob Toews published his predictions for AI in 2022 including a deepening of the east-west technology divide, the next steps in language AI, and climate change becoming a spotlight for innovation.

Read More »

© 2021 Radical Ventures Investments Inc.