Creative Ventures

The logistics sector is no stranger to automation. Large-scale automated sorting, scanning, and the like have been adopted for decades.

Despite labor shortages and spiking wages recently receiving much more mainstream attention, the logistics industry has actually been confronting scarce labor for years, creating a constant interest in technologies meant to increase productivity in the industry. Still, there are significant, specific problems where labor is both critical and highly resistant to even the best efforts of numerous startups over the years.

More recently, ChatGPT has shown the enormous potential of AI’s ability to increase human productivity, but are the technologies that underlie it able to make a significant impact in the logistics industry?

Currently, no. Technologies like ChatGPT will be limited in the world of logistics, but to understand why, we need to look at why AI and robotics have struggled for so many years and why there has been such limited adoption. Ultimately, the greatest opportunity will be found in synthesizing many of these technological ‘waves’ the logistics industry has seen over time.

Manual Labor Is Different

Ironically, most areas of manual labor are much harder for AI to handle than typical ‘knowledge worker’ tasks. There are a few major challenges for AI in logistics and most other labor-intensive industries: frequent exceptions and lack of environmental constraints, manual manipulation being complicated and outside of AI’s ‘natural environment,’ and safety being a much more significant issue in physical environments.

The reason why we have not already been able to automate many areas like sortation fully is because of frequent exceptions.

In a space where items are rarely uniform, barcodes and QR codes get damaged, and environments constantly shift with fast changes in e-commerce trends or even major exogenous events like COVID, AI simply cannot accommodate or adapt. It’s an oversimplification, but even human beings struggle to adapt to these conditions, and they have years of practice (not to mention millennia of evolution behind them) to be able to manipulate and recognize objects adeptly. Not only do AI and robotics lack this advantage, but they often struggle to handle tasks effectively and without human intervention on any universal level.

" With continued rising labor costs and pressure on logistics industry margins, we see a high degree of openness and interest in piloting and adopting new technologies "

In addition, the stakes tend to be higher in these environments. Data is important, but physically putting humans in danger requires a higher safety standard than where AI is deployed in purely digital or knowledge-based environments. This is the same struggle that self-driving cars face, and while there are technologies over the horizon that may be able to help overcome some of these challenges, many of them are fundamental to the environment of logistics.

Robotic Arms, No-Code, and ChatGPT

Given this, let’s look at a few major technological trends, each of which had difficulty overcoming inherent challenges in logistics but have an opportunity in combination with each other.

Robotic Arms

One of the major developments over the past decade came from the proliferation of startups trying to create cheap robotic arms. We’ve seen many of these companies reach pilots but, ultimately, most failed to gain significant traction.

This is because the arms themselves don’t handle any of the major environmental challenges and are still extremely costly to program. In certain cases, programming costs much more than the arms themselves.

The Potential of No-Code

Given the specific problem of expensive programming, No-Code is a natural area of interest. There are many interesting modalities—we invested in one with a ‘wand-based’ robotic training tool called Southie Robotics—but the major challenge here is to create a system that is both flexible and simple enough to encompass the enormous variation that most segments of the logistics industry must face.

ChatGPT... and Other AI models

ChatGPT, obviously, does not operate within the physical world where most of the major challenges for logistics lie (and where labor shortages and rising wages bite deepest). However, we have fielded questions from the industry on whether or not the technology underlying it—transformers and LLMs—could potentially be used within problems in logistics.

There are many places where the fundamental technologies that drive ChatGPT (and the versions fielded by Google, Facebook, and others) could be used, but the commonality is a fundamental ‘grammar’ or structure that can be usefully understood, ‘remixed,’ and exploited. This includes areas in biology, pharmaceutical drug discovery, advanced material design, and similar.

Tasks in logistics like put wall sortation, co-packing, and the like do not have this structure. However, other techniques like simulated data and deep reinforcement learning could make robots (including, but not limited to, robotic arms) more capable. This could also be combined with No-Code to make the ability of No-Code to be more flexible in what is ‘explained’ or ‘taught’ to the robots, which further makes human beings more productive within logistics systems and processes.

Opportunities on the Horizon, with Humans in the Loop

With a synthesis of many of the waves of technology we’ve seen pursued by startups over the years, we believe there’s an opportunity to significantly increase automation and raise workers' productivity within logistics.

However, one fact to remember is that we do not see a viable path within the next 3-5 years (and likely longer) to actually take humans out of the loop. This parallels many industries—including knowledge-based industries with tools like ChatGPT—where automation can handle various low-level work but can only operate smoothly (and consistently) within human oversight.

With continued rising labor costs and pressure on logistics industry margins, we see a high degree of openness and interest in piloting and adopting new technologies. Although we have been skeptical as a firm of many of the prior technological waves because they usually did not do enough to warrant the kind of disruptive process change required to facilitate their adoption, we see many of the upcoming blends of technologies that have emerged over time as creating significant opportunities.