All posts by morse2022

Wrapping Up: Seeing Earth, Talking to the Geon

To wrap up this bespoke residency, now that our 6 months (part-time) has come to an end, I thought it would be helpful to cover a bunch of ideas and workflows we’ve established, some problems, and future directions to explore. There’s a lot to unpack, but I’ll try and be succinct – with a few digressions.

Here’s a video that covers the main points, which I expand upon in the text below.

During the course of this residency I’ve conducted lots of experiments and had a great dialogue with Chris about how to approach things, leading down some inspiring pathways for future work.

Philip Island Multispectral QGIS

2D visualisation

Initially we started off with looking at how to visualise hyperspectral and multispectral satellite data – and it turns to be pretty straight-forward using Python and some programming help using AI – a great way to develop ~explained~ code for libraries one may be unfamiliar with.

You need to know how to iterate and debug a program, work through some documentation, and the data can be accessed in complex file formats and transformed into usable images. This can be done in Juypiter notebooks, running locally in e.g. Anaconda, or remotely on Google Colab or NCRIS Cloudstor notebooks. Pretty straight-forward. These programming assistants will significantly improve over time and complex tasks will become even simpler to prompt.

HDF View File Data
ChatGPT 3.5 HDF Python advice

What the data captures is an entirely different matter – and hyperspectral data is a lot more complicated than multispectral data. I explored a raycasting volume system for that in UE, but that is beyond the remit of this wrap-up. There are lots of different ways to approach it – the question is which one is most useful. It’s a bit “chicken and egg.”

In concert, the question arises as to what one can infer from the data – which is in itself a huge field of scientific and engineering research. It’s not just a matter of applying some kind of GIS colour palette to spectral data – there’s whole fields of analytics that can be applied. These range from naïve naked-eye approaches, through quantitative and statistical analyses to fascinating work in deep learning .

What we have attempted to do is to move this data from the exclusive purview of the specialist to the generalist, from the desktop GIS system to the spatially-located mobile device where an observer may ~actually~ be in a moment in time, and to establish how a human-AI interaction might be established that can create dialogue, queries and observations about the data and their immediate environment.

3D+ Visualisation

Similar approaches can be undertaken in Python for creating 3D+ representations of data, using e.g.  MatPlotLib or VisPy (amongst many others).

For our purposes, Python scripts can run in Unreal Engine/ Omniverse platforms (as well as Unity) and become involved in the creation of geometry, textures, actors and a whole range of actions and narrative entities, Very interesting to see how this is opening up as a result of Virtual Production pipelines, upon which art/science projects can piggyback. This includes creating complex time series animations, physics simulations and volumetric rendering, including interactions with other toolsets used in scientific visualisation and creative industries workflows. Python is the lingua franca.

Cesium

Cesium for Unreal has progressed a great deal. It is currently more stable and flexible than implementations for Omniverse or Unity. I surmise that this is because of the source-available nature of Unreal, as opposed to the closed-source model of the other platforms. Smart move by Epic. I hope it stays that way and becomes more open over time. Cesium is opensource, which is crucial to its flexibility and widespread adoption.

AI and ML

Of course, the elephant in the room is Artificial Intelligence and Machine Learning. It has been fascinating to watch this evolve over the past 6 months – a huge hype-cycle reminiscent of the Blockchain frenzy of the last few years – but at least blockchain was never represented as an ‘existential risk’ . Nevertheless, quantum blockchain technologies will, some day, become human-actor authentication and provenential authorities for lots of different types of data – it may become the only way to distinguish between ‘real’ and generative datasets.

AI  is the definitive enabling technology of our time. It present risks (not yet existential ones) and great opportunities. Like any powerful technology it must be treated with great circumspection and aligned with scientific and ethical interests for the benefit of ‘humanity’. It’s a mirror of humanity, and humanity is not all good. Artistic engagement can help explore and critique this new domain.

XR

David Chalmer’s Reality+ (2022) stimulates  thoughts about the interfusion of world, data and intention, as does Jeff Malpas’ Place and Experience: A Philosophical Topography (2nd Ed. 2018) and Peter Otto’s Multiplying Worlds: Romanticism, Modernity and the Emergence of Virtual Reality (2011).

Apple’s Vision Pro is the latest technological offering in this long history of the intermediation of the self, panorama and place. And it is compelling – not because of the artificial gaze projected to the outside world (deixis to the other), but because of the disposal of controllers – that it can operate by coded hand-gesture alone. Its parts are not new, but the bringing together of systems is. It looks very interesting.

Yet the price of all ‘complete’ XR is total surveillance, even with a ‘secure enclave’: it’s still a head in a box, inherently panoptic and performative.

One wonders how long it will be before we all need to start wearing tinfoil hats to resist implanted thoughts or inception. Quite a long time, I expect, but not forever. Besides, the notion of implanted thoughts is epistemically ambiguous – often these are simply referred to as ‘culture’, ‘beliefs’ and ‘language’. It’s all quite problematic in the post-truth, post-human world of the Novacene. More to contemplate.

Some Practical Examples

In these naive and early days of AI XR, the world that is opening up is fascinating, as I hope the brief sketches above demonstrate. I think of them simply as sketches in exploring how XR will become continuous across mobile devices, HMDs/spectacles and desktop and large immersive screens. Each device format has its own affordances and content, interactions and UI/UX needs to be cognizant of that – lots of interesting design considerations. Natural interactions seem the most compelling, as the premise of ubiquitous/ambient computing is that it will disappear into the background and essentially become invisible – except for intermediation with the world via AI agents such as our idea of the Geon.

I hope you’ve found the material I’ve covered here as interesting, useful and thought provoking as I have! My sincere thanks to Prof. Chris Fluke, the SmartSat CRC and ANAT for facilitating this absorbing residency.  Lots to think about and lots of ideas for future work.

NVIDIA Omniverse or Unreal Engine 5?

Throughout this residency I’ve been working with two development platforms to try and evaluate which is the ‘best'(easiest?optimal?) workflow for the ideas we’ve been exploring. Even at this penultimate stage, things are unclear, so I thought I’d unpack some of the similarities, differences and interactions between the approaches.

Here’s two takes (after a few unsatisfactory tries)  from ChatGPT 4 and Google Bard, using the prompt:

“You are an expert computer artist and programmer. Explain what are the differences and similarities between NVIDIA Omniverse platform and Unreal Engine 5? Create a list of 5 (or so) dot points that clearly indicate when each approach would be the optimal one.”

They’re quite useful – generic (as expected) – and indeterminate because there is so much replication, cross-over and interaction of features.

<aside>reading these responses, I suppose this inevitably speeds the end of some national, regional, alternative and idiomatic versions of English – “ise/ize”, “colour/color”, “mould/mold” etc. </aside>

It seems to come down to a few points that I’ll detail at the end.


GPT4:

As a computer artist and programmer, you are looking at a blend of aesthetics, functionality, and the ability to customize and extend the features of a platform. As of my knowledge cutoff in September 2021, both Unreal Engine 5 and NVIDIA Omniverse have features that will appeal to artists and programmers, albeit with different strengths.

**Unreal Engine 5:**

1. **Game Development**: Unreal Engine 5 is built primarily for the development of games, with a wealth of tools for creating, scripting, and animating game worlds.

2. **Blueprint and C++**: UE5 provides both visual scripting through Blueprint and full source code access with C++, giving you a range of options for scripting and customizing your projects.

3. **Nanite and Lumen**: Nanite allows you to import high-fidelity, cinema-quality art assets directly into the engine, while Lumen provides dynamic, real-time global illumination.

4. **Multi-platform Deployment**: UE5 supports a wide range of platforms, making it ideal for game and application development across devices.

5. **Community and Resources**: The Unreal Engine has a large and active community, which means extensive resources, tutorials, and support.

**Optimal for**: Artists and programmers focused on game development or creating other interactive, real-time experiences. If you need access to source code or want a large community for support, Unreal Engine is a strong choice.

**NVIDIA Omniverse:**

1. **Collaboration and Interoperability**: Omniverse is designed as a platform for collaboration, allowing multiple users to work together in real-time across different applications, thanks to its use of USD (Universal Scene Description).

2. **Physically Accurate Rendering**: Omniverse provides highly accurate, real-time ray tracing rendering with the NVIDIA RTX platform, which is ideal for projects requiring high-quality visualization.

3. **AI Integration**: Omniverse leverages NVIDIA’s AI technologies, offering unique possibilities for AI-enhanced simulation, animation, and more.

4. **Infrastructure Management**: Omniverse can be used to visualize and manage data center infrastructure, an application unique to NVIDIA’s platform.

5. **Integration with other NVIDIA tools**: Omniverse is designed to work seamlessly with other NVIDIA tools, such as Deep Learning Super Sampling (DLSS), enhancing the platform’s capabilities.

**Optimal for**: Collaborative projects requiring real-time contributions across different applications, projects requiring high-fidelity, physically accurate rendering, and tasks involving AI-enhanced graphics or simulation.

Please keep in mind that both platforms continue to evolve and have likely added new features or improvements since my last training data in September 2021.


BARD:

Sure. Here are the similarities and differences between NVIDIA Omniverse platform and Unreal Engine 5:

Similarities

  • Both are powerful 3D creation tools.
  • Both can be used to create realistic and immersive 3D experiences.
  • Both support a wide range of features, including 3D modeling, animation, rendering, and simulation.
  • Both are used by a wide range of professionals, including game developers, architects, and filmmakers.

Differences

  • NVIDIA Omniverse platform is a more collaborative platform, designed to help teams work together on 3D projects.
  • Unreal Engine 5 is a more game-oriented platform, designed to create high-performance, real-time 3D experiences.
  • NVIDIA Omniverse platform is currently only available for Windows, while Unreal Engine 5 is available for Windows, macOS, and Linux.

Here are 5 dot points that clearly indicate when each approach would be the optimal one:

  • NVIDIA Omniverse platform is the optimal choice for:
    • Collaborative 3D projects: The platform’s real-time collaboration features make it easy for teams to work together on 3D projects, regardless of their location.
    • Large-scale projects: The platform’s scalability makes it well-suited for large-scale projects that require multiple users and high-performance computing.
    • Complex projects: The platform’s powerful tools and features make it well-suited for complex projects that require a high level of detail and realism.
  • Unreal Engine 5 is the optimal choice for:
    • Game development: The platform’s real-time rendering and performance make it well-suited for developing high-performance, visually stunning games.
    • Virtual reality and augmented reality: The platform’s support for VR and AR makes it well-suited for developing immersive experiences.
    • Filmmaking: The platform’s powerful tools and features make it well-suited for creating realistic and immersive visual effects for films.

I hope this helps!


It does indeed – thanks Machine Intelligence! 🙂

The main differences are subtle and strategic, and perhaps more effectively understood in the context of industrial competition and the territorialisation of artistic and developer activity – driven by business. It’s basically technological evolution driven by capitalism,  not social good or woo-woo philosophical outcomes. Was it ever not thus? Well, yes – opensource provides an important alternative model, but it has significant infrastructural constraints.

Omniverse provides – at a basic level – an incredible toolset that enables the development of bespoke applications for realtime visualisation and simulation. The workflow is quite different to UE, but it provides access to all its features in Python (above) and C++ (below)- and a graph visual language that is easy to use, including standardised UI features. So it is nice and fast, even if it’s a bit crashy at this stage. It’s pretty compelling and I am interested to learn more, especially as it develops in the direction of UI’s for XR streaming across multiple platforms and the integration of AI and physics simulations (e.g, via Paraview). It’s very modular – reminiscent of Opendoc – but not open! Forgotten dreams of a better world. Really, Omniverse could be access-controlled opensource, like Unreal is. And much more free (not only in $$, but in principle).

Of course, all these systems ~could~ be like a better improved version of Opendoc. But $$$ – it’s in their interests, currently, to be non-interoperable.

Unreal Engine is more mature – and much more stable, in my experience. But far more monolithic – perhaps it could be modularised. There is an emergent UI Library, but UE UI (UMG) seems counter-intuitive to me – it’s complex to get your head around (I understand why it is as it is, because of C++ legacy, but feel it needn’t be this way – UI could be an intuitive MVC plastic layer, not casting, binding and widgets).

Cesium runs both in Omniverse and UE – is it possible to create an Omniverse USD scene using Cesium and pipe that into UE  using an Omniverse connector or is it best just to use the UE Cesium plugin?

An interesting Omniverse/UE co-simulation here.

It would be nice if UE had ~easy~ realtime server capabilities like Omniverse, without all the asset issues of version control with Git, LFS, Bitbucket, Github, Plastic, Perforce, Subversion etc cross-platform. This is something I need to investigate to find an optimal solution for our use-case. Of course, everything is complicated in a non-commercial research context.

A useful talk comparing the two approaches can be seen here (registration may be required)

https://www.nvidia.com/en-us/on-demand/session/gtcspring22-s42162/

Omer Shapira, a Senior Omniverse Engineer from NVIDIA discusses: “Learn about designing and programming spatial computing features in Omniverse. We’ll discuss Omniverse’s real-time ray-traced extended reality (XR) renderer, Omniverse’s Autograph system for visual programming, and show how customers are using Omniverse’s XR tools to do everything together — from design reviews to hangouts.”

A useful diagram from the PDF slides:

It provides an insightful abstract overview of the relationships between data sources, pipelines and developer/artist/user activity – and, of course, computers. Somewhere there must be cloudy-bits and AI that will soon complicate the picture.

I must try and find the abstract view from an Unreal Engine engineer somewhere – I’m sure the diagram would be different. Nevertheless, the territorial play is the same – a stack from hardware engineering, through software engineering, to ‘experiences’ is basically the monopoly. It comes down to engineering choices – the physical constraints, scientific research, economics and politics  of engineering. That defines everything. But it is driven by human desire, a seeming-paradox, top-down but bottom-up.

USD (Universal Scene Description) seems to be on track to be the lingua franca of the metaverse/omniverse/whatevs. A universal file format is an incredibly important idea – not only because there are literally thousands of incompatible, difficult to translate formats, but because of obsolescence. Like human language, digital ‘file language’ evolves and changes; it is currently much more fragile. A significant solar storm might wipe out AI and all digital knowledge. Biology might not be fried in the deep ocean. We’ve had a few billion years of experience.

Nevertheless, the Hyperscale/Game Engine counterpoint is instructive. From my research into Cloud XR, either via Google Immersive Stream or Azure Cloud XR or Amazon etc. – the problem at scale is simply that you need lots and lots of individual virtual machines to run instances of a ‘world’ and lots of low-latency network traffic to synchronise apparent time (with a sprinkling of predictive AI for the ~20ms perceptual lag barrier).

Strangely, this sounds like an ecosystem.

This is clearly a problem that Omniverse tries to address, but won’t work until it is much more open. UE may beat it via the Omniverse of the expanded Fortnight platform, if they can colonise the hardware and mindware at speed. Or they may collaborate or parasitise each other – hard to tell. Are the metaphors appropriate? There are, of course, many other potential players in this space, even, I expect, disruptors or disruptive technologies that have yet to emerge from someone’s loungeroom.

No doubt agentive AGI systems would approach this in entirely different ways, given their own interests.

 

Safe and responsible AI in Australia: discussion paper

Illustration of Supporting responsible AI: discussion paper
Supporting responsible AI: discussion paper. SD/AI Image by Peter Morse

In case you have found it impossible to find via either the ABC news article or Industry and Science Minister Ed Husic’s  media release page, or via the website of the Government of Australia here’s the actual Safe and responsible AI in Australia discussion paper:

https://consult.industry.gov.au/supporting-responsible-ai

Submissions can be made until July 23rd 2023.

Alternative link to pdf:

https://apo.org.au/node/322938

and another:

Australia’s AI Action Plan

https://apo.org.au/node/318137

Analysis and Policy Observatory (APO or Australian Policy Online) – which looks like a credible resource as far as I can see (WHOIS)(Registrant)- has a bunch of useful policy documents on it – and it’s searchable!

Bit of a discovery failure from the Australian Government website. How would you know where to look? Clearly they need an AI assistant such as SwiftSage (something similar coming your way soon).

 

privateGPT

An interesting github repo to keep an eye on – a way to run a private LLM instance and train it on a local document repository – with no internet connection required (once it’s installed!). It uses GPT4all, LlamaCpp, LangChain and a local vector database (not Pinecone, I assume).

https://github.com/imartinez/privateGPT

I can imagine lots of use cases for this: training upon your own files could enable a useful personal assistant that is ~private~. Imagine it as a memory assistant for all those millions of forgotten files on your computer.

Or as a Zotero assistant for working through academic papers and making connections.

Or as a way of making searchable documentation of software – one could scrape all the Unreal Engine documentation using beautifulsoup and then train on it.

Or for an Arts organisation (like ANAT!) – train on years of office data to create a kind of corporate memory.

Of course, it would still have the hallucination problem, but this could be mitigated by combining it with some sort of web search or other confirmation mechanism.

Interesting!

Ambient AI

A community of abstract digital minds

Ambient AI

It’s always problematic making predictions – especially in times of huge technosocial change and disruption. We are experiencing one now (though it may not seem so apparent in day-to-day life) – important aspects need to be identified, and how they will evolve and move forward over the next decade. There are going to be huge, transformational impacts over the immediate future and, especially, our children’s lifetimes.

The principle emergent technology to understand is definitively AI – nothing else comes close, because it is so all-encompassing. Large Language Models are springing up everywhere, with surprising competencies, many of them opensource [1][2][3] and suitable for training upon domain-specific knowledges [4] as well as general tasks.

Interesting approaches in prompt-engineering (PE), chain-of-thought reasoning (CoT), reflexion[5] and, fascinatingly, ‘role-playing’[6] using LLMs, also seem to be improving benchmark performance in concert with reinforcement learning with human feedback (RLHF) [7]:

Thinking of these emergent capabilities, in the context of the current AI arms-race, the issue of human-AI alignment  [8] [9] is of crucial regulatory importance:

Ultimately, to figure out what we really need to worry about, we need better AI literacy among the general public and especially policy-makers.  We need better transparency on how these large AI systems work, how they are trained, and how they are evaluated.  We need independent evaluation, rather than relying on the unreproducible, “just trust us” results in technical reports from companies that profit from these technologies. We need new approaches to the scientific understanding of such models and government support for such research.

Indeed, as some have argued, we need a “Manhattan Project of intense research” on AI’s abilities, limitations, trustworthiness, and interpretability, where the investigation and results are open to anyone.  [9]

Placing this in the context of existential threat, it is well worth absorbing this interview with Geoffrey Hinton in it’s entirety:

Monoliths vs iOT

Although the above concerns sound like science fiction, they are not, and the consequence is that anyone working with AI development (which basically means anyone who interacts with AI systems) must situate themselves within an ethical discourse about consequences that may arise from the use of this technology.

Of course, we have all been doing this for many years through social media and recommender systems – like Amazon, Facebook, VK, Weibo , Pinterest, Etsy – and Google, MicrosoftApple, Netflix, Tesla, Uber, AirBnB etc. – and the millions of data-mining subsidiary industries that have built up around these. Subsidiary re-brands, data-farms, click-farms, bots, credit agencies, an endless web of information with trillions of connections.

In reference to Derrida, I might whimsically call this ‘n-Grammatology”  – given that the pursuit of n-grams has arrived us at this point for the ambiguous machines [10]. A point where the ostensive factivity of science meets the ambiguous epistemology and hermeneutics of embeddings in a vector space – the ‘black box’.

What we know is that AI is a ‘black box’ and that our minds are a ‘black box’, but we have little idea of how similar those ignorances are. They will perhaps be defined by counter-factuals, by what they are not.

Myth

One of the mythologies that surrounds AI that is hard to avoid is that it occurs ‘somewhere else’ on giant machines run by megacorporations or imaginary aliens:

However, as the interview with Hinton above indicates, what has been achieved is an incredible level of compression : a 1-trillion parameter LLM is about 1 terabyte in size:

What this seems to imply is that the kernel will easily fit onto mobile, edge-compute and iOT devices in the near future (e.g. Jetson Nano), and that these devices will probably be able to run independent multimodal AIs.

“AI” is essentially a kind of substrate-independent non-human intelligence, intrinsically capable of global reproduction across billions of devices. It is hard to see how it will not proliferate (with human assistance, initially) into this vast range of technical devices and become universally distributed, rather than existing solely as a service delivered online via APIs controlled by corporations and governments.

AI ‘Society’

The future of AI is not some kind of Colossus, but rather a kind of of global community of ambient interacting agents – a society. Like any society it will be complex, political and ideological – and throw parties:

Exactly how humans fit into this picture will require some careful consideration. Whether the existential risks come to pass are out of the control of most people, by definition. We will essentially be witnesses to the process, with very little opportunity to affect the direction in which it goes in the context of competition between state and corporate actors.

The moment when a human-level AGI emerges will be a singular historic rupture. It seems only a matter of time, an alarmingly short one.

For the next post I will put aside these speculative concerns, and detail some of the steps we have made towards developing a system that incorporates AI, ambient XR and Earth observation. My hope is that this will make some small contribution to a useful and ethical application of the technology.

 


References

[1] “Open Assistant.” Accessed May 9, 2023. https://open-assistant.io/.

[2] “LLMs (LLMs),” May 6, 2023. https://huggingface.co/LLMs.

[3] “Introducing MPT-7B: A New Standard for Open-Source, Commercially Usable LLMs.” Accessed May 9, 2023. https://www.mosaicml.com/blog/mpt-7b.

[4] philschmid blog. “How to Scale LLM Workloads to 20B+ with Amazon SageMaker Using Hugging Face and PyTorch FSDP,” May 2, 2023. https://www.philschmid.de/sagemaker-fsdp-gpt. https://www.philschmid.de/sagemaker-fsdp-gpt

[5] Shinn, Noah, Beck Labash, and Ashwin Gopinath. “Reflexion: An Autonomous Agent with Dynamic Memory and Self-Reflection.” arXiv, March 20, 2023. https://doi.org/10.48550/arXiv.2303.11366.

[6] Drexler, Eric. “Role Architectures: Applying LLMs to Consequential Tasks.” Accessed May 9, 2023. https://www.lesswrong.com/posts/AKaf8zN2neXQEvLit/role-architectures-applying-llms-to-consequential-tasks.

[7] “Reinforcement Learning from Human Feedback.” In Wikipedia, March 30, 2023. https://en.wikipedia.org/w/index.php?title=Reinforcement_learning_from_human_feedback&oldid=1147290523.

[8] Bengio, Yoshua . “Slowing down Development of AI Systems Passing the Turing Test.” Yoshua Bengio (blog), April 5, 2023. https://yoshuabengio.org/2023/04/05/slowing-down-development-of-ai-systems-passing-the-turing-test/.

[9] Mitchell, Melanie. “Thoughts on a Crazy Week in AI News.” Substack newsletter. AI: A Guide for Thinking Humans (blog), April 4, 2023. https://aiguide.substack.com/p/thoughts-on-a-crazy-week-in-ai-news.

[10] Singh, V., 2015. Ambiguity Machines: An Examination [WWW Document]. Tor.com. URL https://www.tor.com/2015/04/29/ambiguity-machines-an-examination-vandana-singh/ (accessed 4.26.23).

Melbourne Visit 2

Last week I went on my second visit to Melbourne to work with Chris at SmartSat and the Swinburne Centre for Astrophysics and Supercomputing. It’s always a blast returning to Melbourne – I lived and worked there for many years, so it’s like returning to my old home town. The city remains familiar yet different – it grows larger and stranger between the years, but it retains both fond memories for me and a sense of nascent possibility.

Swinburne Architecture

You get quite a view of Hawthorne from these buildings.

Hawthorn skyline

And inside you get a view of the whiteboard on which we did most of our work – essentially laying out what we have covered in our collaboration so far and mapping out some possibilities for future progress. Yes, it’s a bit cryptic – so I’ll unpack it with some details in the next post.

Whiteboard diagrams

Sparks of Artificial General Intelligence

Sparks of Artificial General Intelligence: Early experiments with GPT-4 [1]:

https://arxiv.org/abs/2303.12712

Abstract

Artificial intelligence (AI) researchers have been developing and refining large language models (LLMs) that exhibit remarkable capabilities across a variety of domains and tasks, challenging our understanding of learning and cognition. The latest model developed by OpenAI, GPT-4 [Ope23], was trained using an unprecedented scale of compute and data. In this paper, we report on our investigation of an early version of GPT-4, when it was still in active development by OpenAI. We contend that (this early version of) GPT4 is part of a new cohort of LLMs (along with ChatGPT and Google’s PaLM for example) that exhibit more general intelligence than previous AI models. We discuss the rising capabilities and implications of these models. We demonstrate that, beyond its mastery of language, GPT-4 can solve novel and difficult tasks that span mathematics, coding, vision, medicine, law, psychology and more, without needing any special prompting. Moreover, in all of these tasks, GPT-4’s performance is strikingly close to human-level performance, and often vastly surpasses prior models such as ChatGPT. Given the breadth and depth of GPT-4’s capabilities, we believe that it could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system. In our exploration of GPT-4, we put special emphasis on discovering its limitations, and we discuss the challenges ahead for advancing towards deeper and more comprehensive versions of AGI, including the possible need for pursuing a new paradigm that moves beyond next-word prediction. We conclude with reflections on societal influences of the recent technological leap and future research directions.

Chat with the PDF:

https://www.chatpdf.com/share/xnNGlRJJmac5QC8kvFTiT


References:

[1] Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., Lee, P., Lee, Y.T., Li, Y., Lundberg, S., Nori, H., Palangi, H., Ribeiro, M.T., Zhang, Y., 2023. Sparks of Artificial General Intelligence: Early experiments with GPT-4. http://arxiv.org/abs/2303.12712

AI XR in the Cloud

Abstract XR. SD Image by Peter Morse.

Over the last couple of weeks I’ve spent some time virtually attending Nvidia’s GTC Developer Conference, which has been very illuminating. The main take-aways for me have been about how, now that we’re in the Age of AI, that it’s time to really start working with cloud services – and that they’re actually becoming affordable for individuals to use.

Of course, like most computer users, I use cloud services every day – most consumer devices already use them – like Netflix, iCloud, AppleTV, Google Drive, Cloudstor, social media etc. These are kind of passive, invisible services that one uses as part of entertainment, information or storage systems. More complicated systems for developers include things like Google ColabAmazon Web Services (AWS)Microsoft Azure and Nvidia Omniverse, amongst others.

In Australian science programmes there are National Computing Infrastructure (NCI) services such as AuScope Virtual Research Environments, Digital Earth Australia (DEA), Integrated Marine Observing System (IMOS), Australian Research Data Commons (ARDC), and even interesting history and culture applications built atop these such as the Time Layered Cultural Map of Australia (TLCMap). Of course, there are dozens more scattered around various organisations and websites – it’s all quite difficult to discover amidst the acronyms, let alone keep track of, so any list will always be partial and incomplete.

So this is where AI comes in in a strong way – providing the ability to ingest and summarise prodigious volumes of data and information – and hallucinate rubbish – and this is clearly going to be the way of the future. The AI race is on – here are some interesting (but probably already dated) insights from the AI Index by the Stanford Institute for Human-Centered Artificial Intelligence that are worth absorbing:

  • Industry has taken over AI development from academia since 2014.
  • Performance saturation on traditional benchmarks has become a problem.
  • AI can both harm and help the environment, but new models show promise for energy optimization.
  • AI is accelerating scientific progress in various fields.
  • Incidents related to ethical misuse of AI are on the rise.
  • Demand for AI-related skills is increasing across various sectors in the US (and presumably globally)
  • Private investment in AI has decreased for the first time in the last decade (but after an astronomical rise in that decade)
  • Proportion of companies adopting AI has plateaued, but those who have adopted continue to pull ahead.
  • Policymaker interest in AI is increasing globally.
  • Chinese citizens are the most positive about AI products and services, while Americans are among the least positive.

Nevertheless, it is clear to me that the so-called ‘Ai pause‘ is not going to happen – as Toby Walsh observes:

“Why? There’s no hope in hell that companies are going to stop working on AI models voluntarily. There’s too much money at stake. And there’s also no hope in hell that countries are going to impose a moratorium to prevent companies from working on AI models. There’s no historical precedent for such geopolitical coordination.

The letter’s call for action is thus hopelessly unrealistic. And the reasons it gives for this pause are hopelessly misguided. We are not on the cusp of building artificial general intelligence, or AGI, the machine intelligence that would match or exceed human intelligence and threaten human society. Contrary to the letter’s claims, our current AI models are not going to “outnumber, outsmart, obsolete and replace us” any time soon.

In fact, it is their lack of intelligence that should worry us”

The upshot of this, for the work Chris and I are doing, is clearly that we need to embrace AI-in-XR in the development of novel modalities for Earth Observation using Mixed Reality. It’s simply not going to be very compelling doing something as simple as, for example, visualising data in an XR application (such as an AR phone app) that overlays the scene. I’ve now seen many clunky examples of this, and none of them seem especially compelling, useful or widely adopted. The problem really comes down to the graphics capabilities of mobile devices, as this revealing graph demonstrates:

(screenshot from Developing XR Experiences for Every Device (Presented by Google Cloud) )

A potential solution arrives with XR ‘in the cloud’ – and it is evident from this year’s GTC that all the big companies are making a big play in this space and a lot of infrastructure development is going on – billions of dollars of investment. And it’s not just ‘XR’ but “XR with AI’ and high-fidelity, low-latency pixel-streaming. So, my objective is to ride on the coat-tails of this in a low-budget arty-sciencey way, and make the most of the resources that are now becoming available for free (or low-cost) as these huge industries attempt to on-board developers and explore the design-space of applications.

As you might imagine, it has been frustratingly difficult to find documentation and examples of how to go about doing this, as it is all so new. But this is what you expect with emergent and cutting-edge technologies (and almost everyone trying to make a buck off them) – but it’s thankfully something I am used to from my own practice and research: chaining together systems and workflows in the pursuit of novel outcomes.

It’s been a lot to absorb over the last few weeks, but I’m now at the stage where I can begin implementing an AI agent (using the OpenAI API) that one can query with a voice interface (and yes, it talks back), running within a cloud-hosted XR application suitable for e.g. VR HMDs, AR mobile devices, and mixed-reality devices like the Hololens 2 (I wish I had one!). It’s just a sketch at this stage, but I can see the way forward if/when I can get access to the GPT-4 API and plugin architecture, to creating a kind of Earth Observation ‘Oracle’ and a new modality for envisioning and exploring satellite data in XR.

Currently I’m using OpenAI GPT3.5-turbo and playing around with a local install of GPT4-x_Alpaca and AutoGPT, and local pixel-streaming XR. The next step is to move this over to Azure CloudXR and Azure Cognitive Services. Of course, its all much more complicated than it sounds, so I expect a lot of hiccups along the way, but nothing insurmountable.

I’ll post some technical details and (hopefully) some screencaps in a future post.