Our outlook on the AI Landscape

Our Outlook on the AI Landscape

An overview of our investment focus in the coming year

We witnessed a remarkable acceleration in the field of Artificial Intelligence in 2023. Milestones that once seemed years away were achieved with astounding rapidity, surprising even those who have spent decades in the industry. The potential for AI to unlock value across every industry is a tangible reality, and we have just scratched the surface. The influx of exceptional talent, attention, and investment in both AI research and commercial applications underscores the broad applicability and importance of AI technology.

In the last year, we have been encouraged by conversations with amazing founders and the problems they are setting forth to solve with AI. We continually discuss as a team where we think there is open opportunity and what problems we are eager to see solved. In this dynamic landscape, here’s what we’re most excited by and where we’re focused on partnering with founders.

The Augmentation to Automation Continuum

Last year we witnessed varying levels of AI co-pilots and agents across nearly every industry and role type in the boom of genAI. Most co-pilot solutions work in an assistive manner and are not yet able to handle full task completion. User trust simply is not there yet in most scenarios and in many cases it shouldn’t be. We believe co-pilots are phase I of the revolution and not the end point. As adoption continues, we believe solutions will iterate beyond co-pilots into proactive agents. Even with augmentation, adoption across industries will vary with some readily open to augmentation technologies and others moving at a slower pace until trust in the AI increases.  

As we move along this continuum, we believe there are sectors poised to leap more swiftly into automation. We’ve identified three key characteristics that indicate an industry's openness to automation:

1. Data-rich environments, where abundant homogeneous data enables the training of systems that meet necessary accuracy thresholds.

2. Significant improvement over current practices, meaning automated AI systems that are a step change more efficient than the existing state of the art, thereby justifying the risk/reward of full automation.

3. Minimal human-to-human interaction requirements, as the risk of reputational damage from errors can impede full automation in knowledge work.

On both ends of the spectrum, we expect the continued rise of both domain-specific solutions and multimodal models that solve targeted issues across a variety of industries. We are excited by the prospect of multimodal platforms that can ground language in the real world, providing complete solutions to customer pain points that aren’t able to be addressed with LLMs alone.  

Examples of industries we believe are Augmentation ready (GenAI): Finance, Legal, Education, Marketing, Gaming, Customer Support, Security, Software Development

Examples of industries we believe are now or will soon be Automation ready (GenAI, Deep Learning, Reinforcement Learning): Supply Chain, Logistics, Agriculture, Manufacturing, Retail, Insurance, Q/A

‍

The Essential Role of Data Quality in the AI Value Chain

Underpinning both augmented and automated solutions is data and the tooling required to prepare it. This underscores the importance of quality data as the most vital asset for startups building along this continuum. We see this as a gradient of data preparedness across industries; there are those who are ripe for the transition, those where they face challenges with data that is not yet machine learning ready, and lastly industries where sparse data limits the applicability of AI. Startups that have accumulated or have access to unique datasets, particularly in legacy industries, will be in a strong position to win.

We believe that solutions and platforms addressing data quality issues (e.g., incomplete, inconsistent, mislabeled, duplicate data), data cleansing, and data preparation for AI models will continue to rapidly evolve.

As more personalized applications (for both enterprises and consumers) come to fruition, we believe the use of 1st party data for both training and inference will continue to increase. For these personalized models and solutions to occur, we believe there is opportunity for increased data security solutions to be developed to prevent attacks and data leakage in ways that federated learning and encryption are not solving today.  

Computational Discovery

There are several industries that are experiencing technological breakthroughs in their respective fields apart from AI. However, in combination with the AI revolution, these advances will be significantly amplified.  Groundbreaking tools like AlphaFold and RoseTTAFold in protein design are revolutionizing drug discovery and will continue to power even greater advances. Discovery, driven by machine learning, at the intersection of research and commercial opportunity is where we believe this combinatorial disruption will be most felt.  

We firmly believe these advanced science and engineering technologies will bring technological progress in their respective industries as well as immense benefit to the planet and humanity. We think of computational discovery as machine learning led solutions exploring novel designs for microchips to materials to drugs to significant advancements in the energy transition. We believe AI can be applied to solve some of the toughest challenges in these industries and will bring vast improvements to current development processes and timelines. We see opportunity both in these AI-centric discovery platforms as well as supporting tooling for researchers, biologists, chemists, and those facilitating further breakthroughs. For long-term success, both platforms and tools in these areas will require clear use cases and well-defined commercialization strategies.

Underlying Shifts

With all the progress we have seen in the last year, the exciting reality is that there is still so much more to unlock and discover in AI.  We've identified several 'underlying shifts' that we believe will catalyze significant opportunities, introducing new use cases and potential across various industries. Here are some big shifts we're keeping an eye on:

Semantic Understanding:

We're looking for breakthroughs akin to transformer-level advancements in knowledge graphs or structures that resemble human learning and understanding more closely. While tools like GPT are somewhat limited in their current state, lacking factual knowledge and prone to inaccuracies, combining an LLM with a Knowledge Graph (as seen in Bing Chat, Bard, or various bespoke enterprise systems) can provide accurate and specific knowledge. However, these are challenging to build and serve. A breakthrough in building, storing, and serving these could revolutionize many scenarios and mainstream enterprise LLM usage, among other applications.

Zero-Shot Reinforcement Learning:

A zero-shot reinforcement learning agent that can master any task immediately upon learning the rules, without requiring additional training. Enhancing these systems to learn and adapt with minimal input could open new possibilities in robotics and physical world automation, as well as use cases in energy distribution, network optimization and beyond.  

Reasoning:

Improving AI's capability to comprehend and interact with complex concepts and environments is crucial. We view enhanced reasoning as the next phase, succeeding the initial co-pilot era. Breakthroughs in this area could enable solutions that currently fall short of full task automation to integrate more deeply into workflows, solving complex customer problems end-to-end with reduced need for human supervision. We see this believe this could present in two main ways:  

Advancements in causal cognition that allow systems to better anticipate actions and response patterns as opposed to just associated inference. This could have applications from predicting customer churn to increasing model interpretability.  

Progress in the understanding of entity relationships, similar to how humans learn, as opposed to learning from massive training sets. This may indicate a new architecture but could lead to faster, more automated construction of knowledge graphs and the discovery of hidden connections and data relationships.

Conclusion

Since our firm’s inception we have had the privilege of backing pioneering AI startups. We are excited about collaborating with early-stage founders who are driving innovation in the sectors mentioned and beyond. Our investment in these areas is driven by our conviction in the transformative potential of AI to revolutionize these industries. This transformation, we anticipate, will not only generate substantial value but also offer widespread benefits to humanity. We're grateful to play a role in this journey alongside visionary founders and are excited for the years ahead.

Building something? Connect with our team.

Heather Gorham, Principal: hg@flyingfish.vc

‍

‍

‍