New Foundations: Exploring AI for assisted production

Article 1/4

PCH Innovations
5 min readSep 20


There is a lot of hype surrounding AI right now.

This hype seeds fear and over-inflated expectations — which helps no one.

At PCH, we believe the best way to learn about what lies in the unknown is to explore, to play, to experiment, and to share what we learn.

Last summer, we shared our some of our initial learnings from applying LLMs to client manufacturing problems. (See the article: “LLMs: The opportunity for manufacturers’)

This summer, our team has expanded their exploration into recent AI developments and how they might integrate into tricky manufacturing workflows. Over the next few weeks, we will share some results of our explorations into new applications of AI in our New Foundations series.

Why the hype? Shocking capability expansion.

The last 18 months have seen a dramatic improvement of AI capability across domains. Accelerated generation of images, video, animations, text, fully-coded applications, genetic-sequences, protein folding, mathematical proofs, and more. This accelerated improvement in capability is making everyone go coo-coo bananas over what’s coming next.

Are these AIs conscious? Do they have feelings or desires of their own? Should we ascribe identities to these systems? We don’t know.

We do know that the field of AI is undergoing a paradigm shift. A shift away from small, highly tuned models, trained on highly niche, pruned datasets, and towards models that are trained (self-supervised) on broad spectrums of data (e.g. TBs of image data taken from the internet) that can be adapted to a wide range of downstream tasks (e.g. language, vision, robotics, reasoning, human interaction) via fine-tuning.

Foundation models are creating unique potential to build solutions applicable to a wide variety of domains.

Generally called Foundation Models, these systems are shifting the dynamics of what is possible. Coupling foundation models with smaller, more targeted models enables AI systems to combine prediction and classification with increasing ‘reasoning’ capability. This hybridization expands the potential capabilities of AI systems dramatically!

Specifically, there have been:

  • Unexpected improvements in AI reasoning ability, curious emergent behaviors, and expanding abilities to use tools
  • Rapidly developing tool stacks that are increasing the pace of development (model API ecosystems, orchestration frameworks, vector databases).
  • Unexpected LLM cost and size reduction that is enabling the possibility distributed deployment across domains and devices.

These new capabilities are bringing many companies toward a potential inflection point.

The leadership challenge of the decade.

The social, cultural, and economic impact of this technology is indisputable. From design to film production — AI does present a threat to job security in many creative industries, so the anxiety around it is not entirely unwarranted. OpenAI’s own findings indicate that approximately 80% of the U.S. workforce could have at least 10% of their work tasks affected by the introduction of GPTs, while around 19% of workers may see at least 50% of their tasks impacted. 80% of white-collar work will be shifted. [OpenAI — GPTs are GPTs].

But AI also creates the opportunity to recalibrate our work towards human strengths. To free millions from jobs that are dirty, dangerous or require work that misdirects humans from achieving their true creative capability. For example, there are over 10 million unsafe or “undesirable” jobs in the U.S. alone, and an aging population will only make it increasingly difficult for companies to scale their workforces. As a result, the labor supply growth is set to flatline this century. If we want continued growth, we need more productivity — and this means more automation [ref: Figure Master Plan]

Foundation models are lowering barriers to AI application across all industries. Leaders need to recalibrate how they manage resources.

The root of the challenge is to seek an intelligent balance between how human and machine attention is directed towards human progress.

Okay. Lots of grand talk. But how should we get our hands dirty?

Three categories of inquiry.

Our starting point.

At PCH, we assist innovation teams worldwide in exploring what kind of future they should build and prototyping how they can build it.

Most of the teams we work with are pushing the limits of how they can design and manufacture new products. In the past few years, we have built everything from enterprise software that uses LLMs to accelerate manufacturing configuration workflows, to robots that rejuvenate your Nikes.

The common thread among these client teams is context-rich communication. From managers and marketers, to designers, manufacturers, all the way to sales — creators apply knowledge to transform raw materials into beautiful products that serve us.

With that in mind, we look at AI through the lens of immediate application to the worlds of design and manufacturing. Through conversation with innovation teams around the world, we’ve seen increasing interest in the development of systems that can augment aspects of creative work. To expand our ability to support our customers with up-to-date context, over the past few weeks our team has explored new developments along three categories of inquiry:

  1. Organization: How can AI systems give us more clarity on what to build? We previously built a knowledge extraction and management system using traditional NLP techniques and initial LLMs. How much more easily could this be built with emerging LLM tool stacks e.g.LLM APIs, LangChain, and Vector stores?
  2. Flexibility: Can we control robots via language? Developing intuitive robotic experiences is continuous effort at PCH. Rapid recent progress of LLM-enabled reasoning has shown unique functionality in the applied robotics space.
  3. Accuracy: How can AI help us improve positioning accuracy of robotic tools? Last year we launched a robot that could place repair patches on shoes. Since that time, the capability of vision systems have expanded dramatically. How much more accurately might we pick and place components using new vision and representation methods?

This is the first of four articles in a series we call New Foundations where we will dive deeper into the three themes listed above. We’ll be sharing research notes into our current explorations into the state of AI and how it can be applied to the realm of manufacturing and complex system assembly. Read the next article here.

Recommended reading:
On the Opportunities and Risks of Foundation Models

Words by Timo Gmeiner. PCH Innovations is a Berlin-based, creative engineering studio for exploratory technology and innovation strategy.

We are always open to find more curious, talented, and motivated people to help us translate cutting-edge AI research into meaningful applications and products. Does this sound like you? Get in touch



PCH Innovations

PCH Innovations is a creative engineering studio for exploratory technology based in Berlin.

Recommended from Medium


See more recommendations