Quantum Error Correction (QEC) is a daunting task of mitigating the sensitivity of qubits to noise in the environment. Qubits can experience errors in two dimensions: bit flips, where similarly to classical computing error the qubit’s computational state flips from 1 to 0 or vice versa, and phase flips, where the different possible amplitude states may change. AWS introduced at re:Invent 2023 a new quantum chip where it was stated "We've been able to suppress errors by 100x by using a passive error correction approach".

The advancements in error rectification and the efficiency of hardware are paving the way for a future where quantum computers can tackle intricate challenges that were once thought to be beyond our reach.

For those interested in exploring Quantum Computring on AWS, there are free Braket learning paths on the AWS Skills Builder platform. You can gain hands-on experience with the AWS Quantum ecosystem while earning a digital badge demonstrating the knowledge learned.

One of the big advancements is in display brightness, compared to an NXTWEAR S, it has +25% increase, to an impressive 600 nits. This enables to use it outdoors, and with the added option to make the shades permanent part of the glasses it creates a stylish look reminiscent of dark sunglasses.

Furthermore, it has an increased FoV, to 46 degrees, this matches XReal Air 2 and comes close to Rokid's 49 degree FoV. Although pure VR glasses has more extensive FoV, smart glasses like this one have a portable advantage, making them ideal for on-the-go media consumption and enhanced productivity devices.

When comes to software, it's still running the RayNeo XR software providing the AR experience - similar to Nebula for the Xreal glasses. It provides seamless integration for Samsung phones, tablets, with multi window browsing, media streaming, gaming, etc. Although their Android support is better, it still supports being a MiraScreen target to support mirroring iPhone experience.

Source: Unsplash

Sometimes it can feel like we need to switch masks to find the costume that fits our current AI audience. Or that we often have mixed feelings about their use based on our experiences using them, or the direct impact they have on our own lives. The reality is that we need to call for pragmatism and adhere to best practices in ethics to ensure AI technology. Only then can we ensure is used responsibly, securely and to the benefit of this world.

Given it’s October, and Halloween is vastly approaching, let’s consider some of the ghouls and goblins haunting the field of AI ethics, along with the state of best practices in the field.

Bias and Transparency​

Transparency of training datasets is an important consideration for assessing the potential for bias in the results generated by a given machine learning model or algorithm. When using algorithms in our applications without the ability to scrutinise the training datasets, we open ourselves to building applications that contain the biases mirrored in our society.

There are many stories about this that have existed for a considerable period of time. The book and podcast Invisible Women by Caroline Criado-Perez talks about the emergence of various technologies and innovations throughout history where their design negatively impacts women, from the use of crash test dummies based around men from the 1950s until 2011 to AI within the healthcare industry being less likely to diagnose medical conditions in women and other diverse populations. Or even this admission from Tom Cools at Devoxx UK that his attempts to automate the attendance register went wrong when racial bias in the machine learning algorithm he used wouldn’t register one of his students. But we also need to consider other points of bias such as political affiliation as covered in this paper from Uwe Peters or even the skewing of results by misinformation.

To build applications and technologies and automate away the complexities of life in this AI world in a way that reflects our amazing and diverse society, we need visibility into how they are trained, as well as potential outputs. Model cards are a crucial first step to solving this problem. Quite simply, model cards are documentation that describes useful aspects of a model including:

1. Intended use cases
2. Limitations
3. Ethical considerations
4. Datasets used in training
5. Results of experiments to test the model

Not all models have a publicly available card. Prominent repositories are adopting this practice, including Hugging Face which has a publicly available template. Not all models out there have full specifications or details of their datasets to balance with competition considerations. The situation is also the same with dataset cards, for which there is also a template available, where often details of biases in datasets are not transparent. For both, we need an agreed standard and charter committing to provide these assets.

Model Interference and Validation​

Just like applications built without cutting-edge technologies, AI technologies are susceptive to security attacks and manipulation. With the emergence of LLMs, OWASP has created an official Top 10 for Large Language Model Applications.

Many of the threats listed relate to tampering with data, prompts or even the outputs which can lead to the spread of misinformation or malicious activity in the wrong hands. Software libraries exist allowing for the manipulation of these models that, in the event of model theft or duplication, could be used to provide malicious results that cause reputation damage to institutions.

Data Privacy​

It’s not even the case that knowledge of programming is required to potentially manipulate these algorithms. Staying with the LLM example, researchers at IBM reported in August results from an experiment where they were able to hypnotize several LLMs to trick them into releasing confidential information and creating malicious code. In heavily regulated industries such as financial services, this can lead to not only significant reputational damage but also sanctions and fines by regulators across the world.

Moving away from attacks, exposure to proprietary customer information can also be a concern. As covered in a recent FINOS Zenith Brain Trust meeting, the state of licensing for models, and also the code for these technologies, is not mature or well understood.

Even with well-understood licensing, we need to consider the usability challenges this can introduce to developers using these technologies in applications, as well as employees using the tools as part of automating their own work. Just in our personal lives when we click accept on the license terms whose length rivals War and Peace without reading them, there is a potential that ignorance of the terms may lead to the leaking of private information as training for models and algorithms.

Many organizations are already providing company guidelines on the use of AI tools and rules on the exposure of private data, including lists of allowed tools. They are also looking to use on-prem technologies and self-hosted LLMS to prevent data leaks. Even in these cases, the license terms must be carefully reviewed to ensure compliance.

Model Governance​

Transparency in the use of these models within organizations is pivotal to determining which AI technologies are used within the company. Much like the need to track software dependencies to remediate security vulnerabilities quickly, models and AI software must be included in software inventories. Consider the case where a particular model is found to be compromised, or open to a particular security issue.

We need to know exactly where it is used, as well as which environments (production versus development) to act quickly to remediate the problem. AI Factsheets and inclusion of models in existing software inventories are recommended as the approach to handle these concerns.

Another consideration would be the potential accreditation of these technologies, or the organizations building the AI software you use. Governance and accreditation by leading organizations such as AI For Good could help in the decision-making process alongside license review on which models from which organizations meet the required ethical standards that you wish to adhere to, as well as following best practices from these organizations on their usage.

Application Terms of Use​

Terms of use of not just AI technologies, but the applications we build that make use of them, is the final consideration within the ethical AI space. Part of the reason that we hear such terrifying stories of the impact of their usage is because of a lack of published and enforced codes of conduct stating how this technology should be used.

There is a balance in the ethics of usage. From the extreme misuse such as the example in Spain of AI image technology being used to generate naked images of young girls to the more grey area of uploading AI-generated videos of dead children to TikTok for education on child violence that caused legitimate upset to families, applications without explicit terms of use and blocking of violators can find themselves inadvertently being used for nefarious activities. Much like sharing software such as Napster and LiveWire were shut down for copyright infringement due to users sharing music, or the use of data on social media platforms by Cambridge Analytica to influence voters, software intended to change the world and benefit people can often be used in unintended and unethical ways.

There is guidance out there from some model makers, such as the Responsible Use Guidelines from Meta for developers which is referenced in the model cards of their models. But much like the communities and meetups that developers belong to, established codes of conduct in an easily digestible format that outline the acceptable activities in that group are also required to make clear the intended usage of these technologies. This must also extend the applications we build using these technologies, for example, AI image generation applications, stating what is and is not okay.

Conclusion​

While these ethical challenges exist within the current ecosystem, and make adoption by financial services institutions difficult, with clear ethics and terms of use there is no reason that these technologies can't be used to improve the state of our lives and products.

AI is not all about replacing humans. One great recent example that leaves me with hope is the use of AI software trained on keyhole brain surgery to help train neurosurgeons without putting patients at risk, as well as highlighting key areas and tumours. Far from aiming to eliminate surgeons, it sets out a future to make brain surgery safer.

We still need to think about the malicious applications of our software. But we should also take advantage of the benefits AI brings.

From a Financial Services standpoint, ensuring that your ability to recreate the results of a particular query is essential; especially in the eyes of our regulators! This problem is exceptionally complex, compounded by the complexity of models, presenting a significant barrier to full reproducibility and adoption.

Transparency and Reproducibility Challenges​

AI models, particularly deep learning models, are known for their intricate architectures and data-intensive training processes. During the discussion, it became evident that the sheer complexity of these models presents a significant barrier to full reproducibility as well as transparency into their inner workings.

One key challenge lies in the stochastic nature of AI model training. Random processes, such as weight initialization and data shuffling, introduce variability into the training process. Even slight variations in these random factors can lead to different outcomes when attempting to reproduce the model.

The conversation highlighted the critical role of software environments in reproducibility. Managing dependencies, libraries, and software versions becomes crucial when trying to replicate a model's results. Any changes in these components can result in variations in model performance. Providing consistent results in regulated domains such as ours is important given the impact misleading results can have on clients and materials produced using these models.

Furthermore, the use of hardware accelerators like GPUs adds another layer of complexity. Driver and firmware updates for these accelerators can impact model training, necessitating careful management of hardware configurations.

Data Considerations​

Data is the cornerstone of AI model training, and it presents its own set of reproducibility challenges. Datasets can evolve over time, with new data points being added or existing ones modified. Ensuring that the data used for model training remains static or adequately versioned is essential for reproducibility.

The discussion emphasized the importance of transparency in AI model development. Understanding the data sources, preprocessing steps, and hyperparameter choices is critical for others attempting to reproduce the results. The call for more comprehensive model documentation, akin to research papers, is growing louder within the AI community through mechanisms such as data model cards and model factsheets

Model Transparency Best Practices​

While full reproducibility in AI models may be elusive due to inherent complexity and randomness, there are steps organizations and researchers can take to enhance it, as illustrated in the below diagram:

Let’s dive into each of these approaches in turn:

1. Embrace Version Control​

Implement robust version control for both code and data. Track changes meticulously to ensure that others can replicate your work accurately.

2. Prioritize Comprehensive Documentation​

Maintain detailed documentation that encompasses the model architecture, training process, data sources, and hyperparameters used. This documentation serves as a crucial guide for reproducibility.

3. Leverage Containerization​

Utilize containerization technologies like Docker to encapsulate your software environment. Containers facilitate sharing and ensure consistent execution across different setups.

4. Advocate for Standardized Practices​

Support and adhere to standardized practices and guidelines within the AI community. These standards promote consistency and improve the reproducibility of AI models.

5. Promote Transparency​

Transparency is key to reproducibility. Share not only the final model but also the knowledge and best practices associated with it. Document the entire pipeline, from data collection to model deployment.

While achieving full reproducibility in AI models may remain a challenge due to the inherent complexity and randomness, these recommendations provide a solid foundation for enhancing the replicability of AI research and applications.

Licensing & Legal Considerations​

The realm of AI is rapidly evolving, and with it come a host of licensing and legal considerations. A recent discussion with our Brain Trust delved into the complex (and undefined) landscape of AI model usage, copyright, and the evolving role of corporate lawyers.

At the core of all of our discussion and concerns was the need for explainability in AI systems, even when humans are involved in decision-making. The challenge is that many AI technologies currently lack the ability to provide clear explanations for their outputs. This gap raises concerns about transparency and accountability.

Without the ability to explain how AI models arrive at certain decisions, organizations may face hurdles in using these technologies. This, in turn, has significant legal implications, especially in highly regulated sectors like finance.

On top of this, we noted that existing legal tools and processes are not quite caught up with the sorts of questions large enterprise have around copyright, provenance, and licensing concerning AI models. These tools simply haven't been tested thoroughly in this emerging field. No amount of selfie-taking monkeys are going to help satisfy these queries until we really get deep into the subject matter.

Our colleagues in Corporate Law find themselves at a crossroads, unsure whether to adopt a liberal or conservative approach to licensing. An interesting case cited during the discussion was Microsoft's acceptance of copyright liability for all uses of its Copilot AI, a move that leaves corporate lawyers pondering the implications.

As AI technologies become more integrated into industries, there's a growing need for legal frameworks that can adapt to the nuances of AI:

• Current legal structures and copyright laws might not adequately address the complexities of AI model usage.
• Establishing legal guidelines and frameworks for AI model licensing, copyright, and accountability will be essential.

It's a dynamic field where the law is racing to catch up with technology. This is going to be the subject of a Brain Trust subcommittee to be planned for the near future (So stay tuned!).

Security & Data Integrity in AI Models​

The rapid advancement of artificial intelligence (AI) technologies brings with it a pressing concern: security and data integrity. The Brain Trust had a number of thoughts on the intricacies of ensuring the safety and reliability of AI models in a rapidly evolving landscape.

One of the key points raised during the discussion was the importance of securing AI models. AI models can be vulnerable to various types of attacks, including:

• Data poisoning
• Adversarial attacks
• Model inversion attacks.

The security challenge extends beyond protecting AI models from external threats. It also involves implementing robust identity and access management (IAM) systems to ensure that only authorized individuals or systems can interact with these models, especially when considering privileged access on the cloud.

Participants highlighted the need for establishing provenance and a chain of custody for AI models. These mechanisms are crucial for tracking the origin and lineage of AI models, ensuring their integrity throughout their lifecycle. In sectors like finance, where accountability and auditability are paramount, having a clear record of an AI model's history becomes essential, allowing organizations to demonstrate compliance with regulatory requirements.

The discussion underscored the dynamic nature of AI security. As AI technology evolves, so do the threats and vulnerabilities. Organizations must adopt a proactive approach to future-proof their AI security measures. Collaboration among industry experts and researchers can play a crucial role in staying ahead of emerging threats. Regularly updating security protocols and mechanisms is essential to mitigate risks effectively. This will definitely need more discussion and scrutiny in the future!

AI Taxonomy & Standards​

In the fast-evolving landscape of artificial intelligence (AI), establishing clear taxonomies and standards is emerging as a critical enabler for the industry's growth and interoperability.

One of the fundamental challenges in AI is the lack of a universal language to describe AI components, models, and systems. Taxonomy, in this context, serves as a structured framework that classifies and categorizes AI elements, making it easier for stakeholders to communicate effectively. Creating a shared understanding of AI terminology is vital for collaboration and interoperability, especially in an ecosystem where diverse tools, frameworks, and technologies coexist.

Participants emphasized the importance of developing standards that allow different AI systems to seamlessly interact with each other. This pursuit of interoperability extends to various domains, including finance, where data sharing and integration are paramount. Standardizing interfaces and data formats fosters an ecosystem where AI models and tools can work together harmoniously. It reduces integration friction, accelerates innovation, and enables organizations to leverage AI more effectively.

This will be the subject of further conversation with the Interoperability SIG and FINOS Labs at a later date. Taxonomies and standards also play a crucial role in ensuring the ethical and secure use of AI. Establishing guidelines for responsible AI practices is essential in addressing concerns related to bias, fairness, and transparency. In the financial sector, where regulatory compliance is a primary concern, adhering to ethical standards becomes a competitive advantage. It builds trust among customers and regulators while minimizing legal risks.

Developing comprehensive taxonomies and standards requires input from diverse stakeholders, including researchers, policymakers, and industry experts. This is going to become a key objective of the wider FINOS initiative as we explore many other emerging technologies in the future, so it’s going to be a good initial focus for future Brain Trust meetings. Looking ahead, the ongoing journey that demands active participation from the wider AI community and a commitment to ethical, secure, and interoperable AI systems will shape the future of AI. With the proper taxonomies and standards, we lay the foundation for innovation, trust, and responsible AI practices. It's a collective effort that promises to drive the industry forward while ensuring that AI benefits everyone.

The Zenith Team

While the equivalency of AES-192 may not provide a security improvement for many adopters today, the general inclusion of PQC algorithms is a welcomed improvement in what will be a multi-year adoption journey across the industry to protect user network traffic against future quantum cryptanalysis.

Across different VMs, GPUs are interconnected through ConnectX7 InfiniBand, enabling them to communicate with each other at 400Gb/s per GPU (i.e., 3.2 Tb/s per VM). These best-in-class GPU connectivity options substantially improve the training and inferencing performance of large language models (LLMs), which require heavy communication between GPUs. Azure allows easily scaling from 8 GPUs to a few tens/hundreds to many thousands of interconnected GPUs (referred as "super computers") depending on the compute needs of a particular inferencing or training workload.

The first Paper provides an development overviews in portfolio optimization, fraud detection, and Monte Carlo methods for derivative pricing and risk calculation while also providing a comprehensive overview of the applications of quantum computing in the field of blockchain technology. The second Paper focuses on stochastic modelling, optimization and machine learning.

This achievement marks a milestone in proving the utility of quantum computers as scientific tools capable of tackling complex problems that classical systems struggle with. In response to this breakthrough, IBM plans to upgrade its full fleet of IBM Quantum systems to large-scale quantum processors with a minimum of 127 qubits, enabling the exploration of an entirely new class of computational problems.

The impact of this development on financial institutions is potentially significant. With the availability of utility-scale quantum processors, financial organizations can explore the application of quantum computing to optimization problems. Collaborative working groups are being formed to identify and advance the use of quantum calculations in areas such as sustainability and finance, with the aim of leveraging quantum advantage to solve optimization problems. This indicates a growing recognition of the potential value quantum computing holds for the financial industry, as technology continues to advance and offer computational power beyond what classical systems can achieve.

• visionOS SDK
• visionOS Design Resources on Figma
• visionOS Human Interface Guidelines

These resources among with the 46 episodes of training announced at WWDC23 would be the base for the upcoming experiences people build.

The Meta Quest 3 VR headset, which was announced earlier this month and should bring enhanced capabilities, won't be available until September. Unity introduces new development tools for mixed reality applications on Quest headsets. The headset will have color passthrough and improved spatial awareness for more immersive and realistic mixed reality experiences that fit better into everyday life. Unity continues to be the most popular development environment for mobile VR applications and has just released a Meta OpenXR package that supports features such as passthrough, plane detection, raycasting, and anchors for Quest headsets. Unity has also updated some sample projects that use AR Foundation, a framework for cross-platform development of AR applications for mobile devices and XR headsets. Developers who want to create mixed reality apps for Quest headsets need to use Unity 2022 LTS, AR Foundation, Meta OpenXR package, and Meta's Presence Platform.