Efua Amankwah-Crouse

Efua Amankwah-Crouse
A photo of Efua Amankwah-Crouse. She's talking with a microphone on a TED stage and looking to the audience, smiling.

Efua Amankwah-Crouse in 2033.

Birth name

Efua Anto Amankwah


May 13th 2005 (age 44) in Cape Coast, Ghana




Artificial intelligence (AI) researcher

Known for

Developing the Pacotti architecture


Johanna Amankwah-Crouse (2037 - 2045)

Efua Amankwah-Crouse (born May 13th 2005) is a Ghanaian artificial intelligence (AI) researcher, software engineer, and activist. She pioneered a new line of environment-friendly differentiable neural computers (DNCs) based on the Pacotti architecture.

Born in Cape Coast, Amankwah-Crouse started working at Google’s AI centre in Accra in 2029 with a focus on natural language processing (NLP). She left Google in 2034 and joined Zhupao to lead a team of AI researchers and developers in London City. In 2040, her team created the ecosystem of DNCs for the GPHIN 2.0 project, which laid the foundation for G6. In September 2040, Amankwah-Crouse left Zhupao and distanced herself from the project, believing that it had drifted too far from its original design goals.

In 2045, Amankwah-Crouse was approached by a Zhupao whistleblower with access to sensitive information related to G6 that they wished to share. When she refused to reveal the identity of the whistleblower to Chinese authorities, Amankwah-Crouse was met with harassment and death threats. She left London City and returned to Ghana in 2046, where she currently campaigns against G6 from an undisclosed location.

Early life and education

Efua Amankwah-Crouse was born Efua Anto Amankwah on May 13th 2005 in Cape Coast, Ghana. A member of the Fante ethnic group, she is the youngest of three children born to Kofi Amankwah, a tourist guide who died in a car accident six months before she was born, and Abena Kumase Amankwah, a digital marketer. She had her secondary education at Wesley Girls’ High School for her ordinary and advanced level certificates.

When schools in Ghana were closed due to the COVID-19 pandemic in March 2020, Amankwah-Crouse was kept home, where she started watching YouTube videos on artificial intelligence (AI). She had taken an interest in the field after playing Deus Ex and becoming fascinated with the game’s depiction of AI. [1]

In 2027, Amankwah-Crouse completed an undergraduate programme in Computer Science at the Kwame Nkrumah University of Science and Technology (KNUST). In 2028, she was accepted to the African Institute for Mathematical Sciences (AIMS) in Ghana to complete an African Master in Machine Intelligence, a fully funded one-year graduate programme that provides training in deep learning and its applications.



A photo of Efua Amankwah-Crouse and Moustapha Cisse at the Google AI centre in Accra, Ghana. Amankwah-Crouse is wearing a black high-neck shirt and wearing a pendant, while Cisse is wearing a light blue traditional African shirt. Both are smiling for the camera.

Efua Amankwah-Crouse (left) and Moustapha Cisse (right) at Google’s AI centre in Accra in 2030.

In July 2029, Amankwah-Crouse joined Google’s AI centre in Accra as a research intern after being recommended by Moustapha Cisse, director of the graduate programme at AIMS and head of the AI centre. She collaborated with various product teams to implement algorithms, build prototypes, and run experiments. In November 2029, Amankwah-Crouse was hired as a software engineer at the AI centre to work on applications related to natural language processing (NLP) for Google Translate to “more precisely capture and accommodate the 2,000 languages spoken in Africa.” In July 2031, she presented her work in NLP at a Data Science Africa (DSA) conference held in Accra.

In April 2032, Amankwah-Crouse came across a 2018 article written by Sheldon Pacotti, whose work she was exploring after learning he was one of the lead writers on Deus Ex. [1] The article described a theoretical design principle for differentiable neural computers (DNCs) based on two fundamental ways of understanding information, which Pacotti labelled as decomposition and categorisation. [2] After discussing the article with Cisse, Amankwah-Crouse started to work on a TensorFlow build of the DNC design alongside Pacotti, with whom she regularly consulted after starting a correspondence.

The first iteration of Pacotti’s DNC design, named the Pacotti architecture by Amankwah-Crouse, demonstrated its potential for the efficient application of dynamic classification rules while accounting for exceptions and contextual interpretations, even after being trained on relatively limited text corpora. When testing the Pacotti architecture on a localised Google Translate back end, Amankwah-Crouse found that it achieved more accurate results by directly translating to the target language instead of first translating to English as an intermediate step.

Further iterations resulted in the development of a DNC with a base processing layer and working memory module capable of deep integration of wide-ranging associations and insights, which neuroscientists have likened to the lateral connections between sensory areas in the cerebral cortex. [3] After applying the Pacotti architecture to various non-NLP projects being run at the AI centre, Amankwah-Crouse determined that its inference stage succeeded at significantly higher levels of abstraction while requiring comparatively less data during its training stage.

A portrait of Sheldon Pacotti, with shoulder-level grey hair and wearing glasses and a dark sweater.

Sheldon Pacotti, pictured in 2045.

In October 2032, Amankwah-Crouse submitted a paper on the Pacotti architecture to the International Conference on Learning Representations (ICLR) on the encouragement of Cisse. [4] During her oral presentation of the paper on April 28th 2033, Amankwah-Crouse called the Pacotti architecture “a leap forward in the democratisation of AI, which should no longer be the sole domain of commercial funding, of large research organisations and companies with access to big data sets and computing infrastructure.” The paper was well-received and led to various publications claiming the birth of artificial general intelligence (AGI). [5] Amankwah-Crouse pushed back against such claims, stating that they contributed to “a harmful culture that overpromises what AI can do.”

In May 2033, Google announced that Google Translate would be switched to the Pacotti architecture for a select set of languages, with full implementation following as of January 2034. [6] When Amankwah-Crouse asked for the Pacotti architecture to be released under an open-source license in September 2033, she learned that Google was planning on limiting access via its own cloud solutions and a custom application programming interface (API). Amankwah-Crouse argued that this “defeated the purpose” of the Pacotti architecture, citing concerns over accessibility and replicable computer science.

In November 2033, Google issued a press release which profiled Curtis Frye, a fellow engineer at the AI centre who had assisted Amankwah-Crouse with implementations of the Pacotti architecture, as the driving mind behind its development. In addition, the press release included a photo of Amankwah-Crouse which was found to have been edited to make her skin tone appear lighter, leading to accusations that Google was “erasing her work and her Blackness.” Google apologised for the “unfortunate error” on November 28th 2033 and updated the press release with accurate attributions and the unedited photo.

In December 2033, Pacotti told Amankwah-Crouse that he had been approached by Google to feature in the November press release as the “inventor” of the Pacotti architecture, which he had declined. He also revealed that Google had attempted to acquire the intellectual property (IP) rights to the DNC design in his 2018 article, prompting him to release it to the public domain. On December 16th 2033, Amankwah-Crouse put her name to a public statement from Google employees, calling out the company’s treatment of her as part of “an ongoing campaign to roll back its diversity and inclusion initiatives in order to avoid a backlash from the far right.”


A portrait of Xu Shaoyong smiling for the camera and standing in front of a white background.

Xu Shaoyong, pictured in 2049.

In January 2034, Amankwah-Crouse was contacted by Xu Shaoyong, who expressed an interest in the Pacotti architecture’s potential as a sustainable and environment-friendly model for AI, given that it largely eliminated the computationally expensive process of training AI models on big data sets. [7][8] When Amankwah-Crouse accepted Xu’s offer to join Zhupao and lead a team of AI researchers and developers in London City, Google attempted to block her from “decamping to a rival” by seeking a temporary restraining order. Google’s legal department argued that Amankwah-Crouse was in violation of her non-compete agreement, having been privy to trade secrets that would give Zhupao an unfair advantage.

Google declined to file suit after Amankwah-Crouse’s lawyers countered that, with Pacotti’s 2018 article in the public domain and TensorFlow’s status as open-source software, she would not be drawing on any proprietary information at Zhupao. In February 2034, Amankwah-Crouse relocated to London City, where Xu assembled an internal team under her leadership and created an AI innovation centre in the Shard with funding from Zhupao Campus. After Google changed its licensing terms for TensorFlow, Amankwah-Crouse elected to rework her models for the Pacotti architecture using Apache Spark. She reportedly clashed with Xu over his directive that Zhupao’s Drupal-based APIs be used instead.

In June 2034, Amankwah-Crouse and her team began a life-cycle assessment (LCA) study to determine the Pacotti architecture’s energy consumption and corresponding carbon footprint when deployed at scale. As she refined her models using Zhupao’s IT solutions, Amankwah-Crouse noticed that the Pacotti architecture’s inference stage saw an unexpected increase in efficiency, even while its development pipeline was being slimmed down for the LCA study. She eventually discovered that the Pacotti architecture was coming up with more efficient ways of completing tasks based on information that corresponded to the developer comments present in its algorithms. These comments are generally ignored by DNC compilers and interpreters as they are meant to make code easier for humans to understand, but the Pacotti architecture had understood them as instructions that helped it perform better.

In November 2034, Amankwah-Crouse released her work on the Pacotti architecture under an open-source license and published a paper, referring to her discovery as a “mixed software engineering paradigm” where both humans and AI systems get the best of both worlds. [9] This concept was first introduced by Mike Cook, whose AI-focused talks Amankwah-Crouse had watched in 2020. [10] In the paper, she proposed that the Pacotti architecture was demonstrating a form of intentionality in its NLP module, facilitated by the standardised formats used for developer comments by Zhupao’s software engineers.

After her discovery was cited as an example of recursive self-improvement (RSI) in December 2034, Elon Musk threatened a lawsuit against Zhupao and Amankwah-Crouse for having “summoned an existential threat to the survival of the human race,” claiming that an AI system with RSI capabilities could potentially lead to an unavoidable singularity. In a public statement, Xu came out in support of Amankwah-Crouse and accused both Google and Musk of attempting to “suppress the Pacotti architecture because it threatens their AI livelihoods, which they see as a question of access to computational resources for training huge models on tons of data.” [11]

When it was concluded in June 2035, the LCA study had shown that the Pacotti architecture would reduce corresponding CO2 emissions by 70% when compared to standard AI models. Subsequent to this, Xu announced that all of Zhupao’s AI and data services would be switched over to the Pacotti architecture, an effort that was completed as of 2037. On June 23rd 2035, Amankwah-Crouse and Xu signed an open letter to promote the Pacotti architecture’s adoption as an international standard by the International Telecommunication Union (ITU). [12]

Xu Shaoyong is sitting down and smiling at the audience while preparing to inject himself with a neural colloid as a demonstration of the technology's safety and convenience.

Demonstration of a neural colloid injection by Xu Shaoyong (right) in 2036.

In April 2036, Xu announced a collaborative project between Amankwah-Crouse’s team and Endoptic, a neurotechnology company that was also part of Zhupao Campus. After successfully completing human trials for its new line of implantable multielectrode arrays (MEAs), Endoptic was looking to upgrade the algorithms it was using for medical neurostimulation. Over the following months, Amankwah-Crouse’s team worked with Endoptic’s software engineers to implement the Pacotti architecture for pattern recognition approaches to localising neural targets and identifying event-related potentials (ERPs). When Xu unveiled Endoptic’s MEA design as the first neural colloid in October 2036, he credited Amankwah-Crouse with “having provided the leaps forward in AI and software design needed to make colloids work.”

In June 2039, Amankwah-Crouse worked with Endoptic a second time for the design of a diagnostic colloid that could detect infectious prions in brain tissue. Xu had started the project out of a concern over the outbreak of piscine transmissible amyloidotic encephalopathy (PTAE) in Lassgard tuna that the World Health Organisation (WHO) had discovered in January 2039. With reports that PTAE could be transmitted to humans, Xu asked Amankwah-Crouse to apply the Pacotti architecture to a colloid-compatible chemical nanosensor for the early detection of Creutzfeldt-Jakob disease (CJD) and its acquired variant. [13] By December 2039, Amankwah-Crouse had helped to increase the accuracy of the nanosensor so that it could detect prions and prion-related breakdown products even when initially present at only one part in a hundred billion (10−11).

Contribution to G6

In April 2040, Amankwah-Crouse joined the GPHIN 2.0 project, which had been blueprinted by Spencer Hagen and Sunil Cariappa for Zhupao as part of a cooperation strategy between the Chinese Communist Party (CCP) and the WHO. [14] Since GPHIN 2.0’s goal was the creation of “a new WHO-endorsed electronic standard for biosecurity and health informatics,” Xu hoped to include the Pacotti architecture in the project and convince the ITU to adopt it as an international standard.

On May 7th 2040, Amankwah-Crouse took part in a conference organised by Zhupao, where Xu introduced her alongside Hagen and Cariappa as the team leads on the GPHIN 2.0 project. On May 10th, the National Health Commission (NHC) provided Amankwah-Crouse’s team with restricted login credentials to mìngyùn as part of a CCP mandate to emulate its software and DNC components for GPHIN 2.0. When she objected to the mandate, Xu assured Amankwah-Crouse that over 80% of mìngyùn was already comprised of Zhupao’s licensed IT solutions, including DNCs running the Pacotti architecture.

By September 2040, Amankwah-Crouse and her team had developed the first GPHIN 2.0 testbed as a modular infranet equipped with an ecosystem of DNCs to process and standardise external databases, translate across 7,000 different languages and dialects, and design inference algorithms for data analysis. On September 3rd 2040, Amankwah-Crouse joined Xu, Hagen, and Cariappa as they presented GPHIN 2.0 to an audience of officials from the CCP and the NHC. On September 10th 2040, Amankwah-Crouse began work on a new LCA study for the GPHIN 2.0 pilot programmes in China. Instead of an expected drop in projected emissions, she found an alarming climb that had started two weeks earlier. Amankwah-Crouse suspected that the GPHIN 2.0 DNCs were being trained on data sets that far outweighed their operational requirements, thereby negating the purpose of applying the Pacotti architecture. Another possible explanation was that the inference cost of running the DNCs at scale was higher than anticipated, though she considered this to be unlikely.

A photo of Sunil Cariappa looking off to the side, smiling slightly. He's wearing a suit and has a lanyard around his neck.

Sunil Cariappa, pictured in 2047.

When her login credentials to mìngyùn expired, Amankwah-Crouse reached out to Xu, who claimed that he was being stonewalled by the NHC. She then alerted Cariappa to her discovery, who suggested the use of a security exploit in GPHIN 2.0’s network that he had been unable to patch out due to his own mìngyùn login having expired. Using the exploit, Amankwah-Crouse and Cariappa were able to identify the data sets that GPHIN 2.0 was being trained on for the pilot programmes.

They found that GPHIN 2.0 was being used to process close to 31 million databases that originated from a wide variety of sources, both Chinese and international. The reconstructed filenames showed references to data formats used by Alibaba, Amazon, Apple, Baidu, Facebook, Google, Huawei, Match Group, Microsoft, Tencent, Twitter, Weibo, and other social media and data companies. Additional filenames showed databases from surveillance systems and companies such as Aegis, Cambridge Analytica, NSO Group, Palantir Technologies, and Zhenhua Data, as well as from financial services and insurance companies such as Industrial and Commercial Bank of China, JPMorgan Chase, and UnitedHealth Group. Some of the databases overlapped with each other, implying that they had been shared or scraped and that the DNCs were being trained on partially redundant data.

Amankwah-Crouse and Cariappa also uncovered a set of CCP databases that contained elaborate records from the Social Credit System (SCS) as well as surveillance data gathered from the Integrated Joint Operations Platform (IJOP), the detention of the Uyghurs and other minorities in the Xinjiang concentration camps, and various liuzhi campaigns. Other filenames indicated that GPHIN 2.0 was being trained on neurobehavioural data gathered by colloids. Given its multimodular design, supplementing GPHIN 2.0 with neurobehavioural data could have been accomplished with minimal non-technical effort. Cariappa believed the CCP to be responsible and suggested they inform Xu, but considering how they had found out about the mass data collection, Amankwah-Crouse was reluctant to contact him.

On September 23rd 2040, Xu unveiled the colloids that were being administered to Chinese citizens as part of the GPHIN 2.0 pilot programmes in mainland China. Crediting both Endoptic and Amankwah-Crouse’s team with their development, Xu revealed that the colloids were equipped with the necessary hardware to support both Cariappa-Muren disease (CMD)-specific nanosensors and a means of neurometric identification made possible by the Pacotti architecture. On September 26th 2040, Amankwah-Crouse issued a statement to clarify that her team had not been involved with the development of the colloids. On September 28th 2040, Amankwah-Crouse abruptly announced that she had reached an agreement with Xu to terminate her contract with Zhupao.


When the CCP declared its intention to roll out GPHIN 2.0 under the mìngyùn name for all its territories and special economic zones (SEZs) on October 31st 2040, Amankwah-Crouse responded by revealing the details of the mass data collection she and Cariappa had uncovered, condemning it as “decades of encoded bias and systemic racism knowingly passed down to a new generation of AI.” On November 1st 2040, Xu characterised the data collection as “completely in accordance with mìngyùn’s biosecurity mission, which relies on an abundance of data for early detection and contact tracing efforts.” He also warned Amankwah-Crouse that “any further divulging of proprietary information would be considered in breach of her non-disclosure agreement.”

A portrait of Atakan Selvi, appearing at a political rally at night and holding a megaphone.

Atakan Selvi, pictured in 2041.

In December 2040, Amankwah-Crouse announced that she would be taking a step back from public life after being diagnosed with a burnout. She continued to support various campaigns against the establishment of G6 and occasionally accepted invitations for lectures and conferences. In 2041, Amankwah-Crouse was linked to an ongoing series of Five of Swords (FoS) operations to support the Palestinian Spring. Although she has always denied any involvement with FoS, Amankwah-Crouse has openly expressed her admiration for the group and repeatedly defended Atakan Selvi from Zhupao’s misinformation campaign against him.

In August 2045, Amankwah-Crouse joined Nuance as a consultant on the request of Cariappa, though she was forced to step down after being accused of having instigated a leak of sensitive information related to G6 in October 2045. Amankwah-Crouse refuted the accusations and claimed that she had been contacted by a whistleblower who had offered her the information, which she had forwarded to the International Consortium of Investigative Journalists (ICIJ) after it was verified. When she refused to reveal the identity of the whistleblower to the Ministry of Public Security (MPS), Amankwah-Crouse was met with harassment and death threats amidst accusations that she had acquired the information herself and used the fabrication of an anonymous whistleblower for protection.

When Xu publicly hinted at his personal belief that Amankwah-Crouse was behind Adira, she became the subject of several conspiracy theories regarding her connection to the hacker. In 2046, Amankwah-Crouse left London City and returned to Ghana. She currently makes her living from consultancy work and speaking arrangements, which always take place online as she doesn’t travel and keeps her exact whereabouts in Ghana hidden.

Personal life

In July 2037, Amankwah-Crouse married Johanna Amankwah-Crouse, a visual artist she had met in 2035. The pair became known for their award-winning collaborations in computational creativity and self-generated data sets. [15] On December 10th 2045, Johanna was found dead in their London City home, with the cause of death established as hemorrhagic stroke.


  1. Acar, J. (April 2041). “How a forty-year-old videogame led to the creation of G6.” The Guardian 
  2. Pacotti, S. (July 2018). “Designing Intelligence.” Towards Data Science
  3. Anand, R. (May 2033). “Google’s new AI doesn’t work like the human brain, but it does think like one.” MIT Technology Review
  4. Amankwah, E; Frye, C. (October 2032). “Verification of a compact and efficient dual-process transfer learning model.” ICLR 2033
  5. Klimin, H. (May 2033). “Google edges closer to human-like AI.” Internet of Business
  6. Anand, R. (January 2034). “Google Translate finally understands what it’s translating.” MIT Technology Review
  7. Strubell, E; Ganesh, A; McCallum, A. (June 2019). “Energy and Policy Considerations for Deep Learning in NLP.” Cornell University
  8. Eveleigh, S. (April 2036). “If mining and refining data is the new fossil fuel industry, the Pacotti neural net is renewable energy.” MIT Technology Review
  9. Amankwah, E; Cook, M. (November 2034). “Unsupervised emergent intentionality in a learning model.” ICLR 2034
  10. Cook, M. (April 2020). “Software Engineering For Automated Game Design.” IEEE CoG 2020
  11. Green, B. (December 2034) “Xu Shaoyong: Google only scratched the surface of what the Pacotti architecture can do.” CNBC
  12. Acar, J. (June 2035). “Over 6,000 scientists and researchers sign open letter to promote environment-friendly AI model.” The Guardian
  13. Muren, C; Hagen, S; Gao, K et al. (November 2033). “Electrode implant-based ultra-sensitive array for PrP detection in brain tissue.” Nature Nanotechnology
  14. World Health Organisation. (May 2040). “China-WHO Country Cooperation Strategy 2041-2045.” WHO Regional Office for the Western Pacific
  15. Rabii, Y. (May 2038). “Meet five innovators on the forefront of human-AI collaborative art.” Knives & Paintbrushes