Inspiration
A collection of ideas, papers and opinions in sectors and technologies that we follow. We hope these spark new ideas and perhaps lead to your next startup. Similarly, if you are already building in adjacent areas, please share your work with us.
Beyond being smart, the importance of new ideas
"If you asked people what was special about Einstein, most would say that he was really smart. [...] But that wasn't what was special about Einstein. What was special about him was that he had important new ideas. Being very smart was a necessary precondition for having those ideas, but the two are not identical." PG
Domain Data Science
Data science needs to become deeply integrated into every industry sector and almost every corporate function to drive productivity, revenue growth and market transformations. But a brilliant data scientist left alone in the realm of biology won't discover new therapeutics. Building a predictive model to identify errors in produced parts is useless without understanding quality control of a manufacturing process. And we can't expect data scientists to become domain experts in all these respective fields. We need data science to be fused with deep domain expertise. We need new collaborative software systems that enable industry domain experts, researchers and applied engineers to work hand in hand with data scientists.
What would "pair programming" look like between a biologist and data scientist?
Remoteness
We are all familiar with Zoom meetings and working from home, as well as the accelerated growth and now near-ubiquitous dependence on e-commerce. We can easily imagine the impacts these may have even following the pandemic on office space, business travel, and in-person retail. But the next-level implications of -- and opportunities created by -- “remoteness” are leading to exciting new enterprises in a wide variety of sectors.
In the “future of work” sector, most obviously, we are seeing entrepreneurs creating intelligent technology solutions beyond mere communications platforms to support distributed team collaboration and engagement and secure and effective customer, partner and vendor interaction and processes. In the “built economy” sector, we are seeing “remoteness” addressed through intelligent integration of computer vision and analysis for safety on construction sites, in manufacturing locations, and for expanded and intelligent monitoring of common spaces. We are also seeing a rapid rise in the creation of robotics applications for interior space management, waste management, and automated utilities monitoring and applications. And in digital health and health care we are seeing the introduction of novel component solutions enabling effective tele-medicine, digital and automated diagnostics, analysis, and processing. All of these are being unleashed by the requirements and possibilities of remoteness.
We expect “remoteness” to continue to drive innovation in these and other sectors in the coming years.
Toward Trustworthy AI - Bias Bounties
AI has enabled a diverse array of applications across commercial, scientific, and creative domains. With this has come a growing awareness of the large-scale impacts of AI systems, and recognition that existing regulations and norms in industry and academia are insufficient to ensure responsible AI development. Bias and safety bounties for AI systems could strengthen incentives and processes for broad-based scrutiny of AI systems resulting in improved societal accountability.
Dark Data
Gartner defines Dark Data "as the information assets organizations collect, process and store during regular business activities, but generally fail to use for other purposes. Similar to dark matter in physics, dark data often comprises most organizations’ universe of information assets. Thus, organizations often retain dark data for compliance purposes only. Storing and securing data typically incurs more expense (and sometimes greater risk) than value."
Some organizations believe that dark data could be useful to them in the future, once they have acquired better analytic and business intelligence technology to process the information. Because storage is inexpensive, storing data is easy.
Is now the time where the latent value of Dark Data can be realized? (See Inspiration block on GPT!)
Applied use of GPT (n+1)
OpenAI's GPT system appears magical at times. The path toward embedded AI assistance and conversational interfaces is becoming more and more plausible. Our fund's investment thesis, Intelligent Industry Solutions, is based on an observation that collectively over the last decade+ we've built out these amazing horizontal technology platforms. We're now poised to have domain experts apply these innovations to sector-specific use cases. We see a natural progression of the n+1 version of GPT and GenAI variants to be a cornerstone of new enterprise software systems.
(note: we wrote this block in the summer of 2020 shortly after GPT-3's launch. We've been interested in GenAI for years!)
$0 synthetic media
For years we've been following the work of Generative AI systems, first as interesting experiments and special projects, but now we're increasingly in an entirely new modality where nearly any piece of media can be produced at radically low costs. You no longer need to go to art school, be a trained graphic designer or a professional photographer. Over the coming years this will mean that the processes, workflows and partnerships around media will fundamentally shift. A single marketer can design, launch and analyze an entire multi-platform campaign. Local franchisees and other commercial, global partnerships can be empowered to hyper-personalize media for advertising and aligned-brand building. And that's only within the B2B realm, let alone new consumer experiences in gaming, spatial computing a la Apple Vision Pro, etc. As a second order effect, this inflection point creates opportunities for new platforms to deliver tremendous value adjacent to the more creation tools.
The CUP Theory of AI defensibility for Services-As-Software
We think Rob May's perspective on the applied use of AI technologies within Services-as-Software businesses has a lot of potential. Defensibility will arise from 3 things: Consistency, Ubiquity, and Personalization (aka CUP). Learn more on his blog.
Adversarial Attacks and Defenses
With the rapid developments of artificial intelligence (AI) and deep learning (DL) techniques, it is critical to ensure the security and robustness of the deployed algorithms. Recently, the security vulnerability of DL algorithms to adversarial samples has been widely recognized. The fabricated samples can lead to various misbehaviors of the DL models while being perceived as benign by humans. Successful implementations of adversarial attacks in real physical-world scenarios further demonstrate their practicality. Hence, adversarial attack and defense techniques have attracted increasing attention from both machine learning and security communities and have become a hot research topic in recent years. In this paper, we first introduce the theoretical foundations, algorithms, and applications of adversarial attack techniques. We then describe a few research efforts on the defense techniques, which cover the broad frontier in the field. Several open problems and challenges are subsequently discussed, which we hope will provoke further research efforts in this critical area.