Investing in artificial intelligence: What’s all the fuss about?

Artificial intelligence is one of the most exciting and transformative opportunities of our time, argues Nathan Benaich, an investor at London-based investment firm Playfair Capital.

While the field has gone through two cycles of bull (‘50s-’70s and ‘80s) and bear (‘70s and ‘97-’93) markets, where expectations were hyped up only to eventually fall short of their promises, I think this time is different.

From my vantage point as a venture investor at Playfair Capital, where I focus on investing and building a community around AI, I think it’s a fantastic time to build and invest into companies creating or leveraging AI for their products and services.

There are three key reasons. First, our addictive usage of the internet, software products and smartphones – all two billion of them – has created data assets that describe our behaviours, interests, knowledge, connections (and more) at a scale and granularity that has never existed before.

This is the key raw material with which we can build learning systems. Second, the costs to compute and store data are both falling precipitously, making computationally-intensive applications affordable and tractable.

Third, research communities are publishing significant improvements to the design and development of learning systems that accelerate the pace at which we see real world applications emerge.

Where is AI applied?

Many of the underlying tools employed to build AI systems, such as natural language processing (NLP), computer vision, machine learning and deep learning, are already out in the wild powering software services we know and love.

The vast data silos that exist within the public domain, enterprises, research institutions and government data portals present analytical opportunities both for their owners and for companies able to address new questions with a holistic view of this data.

Indeed, companies such as DueDil, in business intelligence; Stratified Medical, in life sciences; or PeopleGraph, in search; have employed NLP, graph theory, machine learning and information retrieval to extract new high-value signals in their domains that would otherwise be missed.

There are also many types of tasks that are inefficiently completed by knowledge workers because they are repetitive, cumbersome and data-heavy. Companies such as Tractable, in auto repair insurance; Gluru, in contextual file management; and Kreditech, for consumer credit scoring; demonstrate AI’s ability to drive significant efficiency gains. AI can also improve problem solving in domains that otherwise rely on deep expertise developed over many years.

For example, both Ravelin, in online fraud detection/prevention; and Unbabel, in machine translation; are cognisant of the value in having human experts-in-the-loop. This framework is powerful because it complements AI-driven results with human expertise and enables active learning from feedback to improve performance over time.

AI is also a key enabling driver for autonomous agents on land, air and sea, as well as virtual and augmented reality. Indeed, computer vision, deep learning and reinforcement learning are pivotal for companies like Mobileye, which supplies the autopilot technology for Tesla, and SkyCatch, a drone data software platform.

Given the R&D-heavy nature of this space, there is a rather unique opportunity that focuses on productising core technologies that are part and parcel of the AI ecosystem. For example, Cloudera and Databricks market enterprise data analysis solutions using Apache Hadoop and Spark, respectively, while Seldon and Ayasdi provide AI infrastructure and tools as a service.

Further along the R&D spectrum, we find the likes of Google DeepMind and Vicarious working on general learning architectures that will bear fruit in the longer term.

Building and investing in AI

While many of the same company-building principles apply, a specific set of operational, commercial and financial challenges manifest in the AI space. Working on AI problems requires significant domain expertise and R&D efforts.

While barriers to entry are being chipped away by popular open source frameworks like Torch, Theano, Caffe or TensorFlow, they don’t erase the problem altogether. The first issue this poses is one of talent sourcing and retention.

Here, founders should talk about the hard research problems their company solves to attract interest from the best engineers who care about advancing the status quo in their field.

Second, it’s important to balance the longer-term R&D route with monetisation in the short term by scoping out paid client work early on. This approach can also be a means to bootstrap a learning system with much needed data.

What’s more, given that users will often be benchmarking an AI product’s performance against a result achieved by a human, the best teams develop their technology in close collaboration with end users.

Indeed, dedicating resources to product research, design and experience as an afterthought is not an option in a world where many products exist that solve the same pain.

While it’s clear many industries are awash with interest in AI, the buyer’s knowledge and adoption of this technology is often nascent (exceptions being FinTech and AdTech). Hence, it’s key to understand how value is measured and benchmarked in a use case, how data can be accessed for training and learnings shared across multiple clients and which business models work as a function of sales cycles.

On the financial markets front, last year was a record year for investments (312 rounds in 267 companies, +20% vs 2014) and exits (42 liquidity events, +41% vs 2014) in private AI companies.

However, this sector accounted for only 1% of the $129bn of invested capital and 3,411 exits in the broader technology market in the same year. Indeed, we’re still in the early days of value creation with AI; two thirds of the financings were at the Seed and Series A stage.

Of note for entrepreneurs, North American companies accounted for 85% of all capital invested, 71% of exits and raised 5x larger late stage rounds compared to their peers in the rest of the world.

Reassuringly, the UK is pulling its weight as the second most active country after the US. Indeed, SwiftKey, VocalIQ and DeepMind were all founded and run in the UK and later acquired by US-based technology giants Microsoft, Apple and Google, respectively.

Bullish outlook

I’m bullish on the value to be created with AI across our personal and professional lives. More support is needed for companies driving long-term innovation – VC was born to fund moonshots after all.

Given that access to technology is increasingly commoditised, the winning strategy for building a sustainable advantage may in fact be non-AI and non-technical in nature, eg the user experience layer.

As such, there’s a renewed focused on core principles: build a solution to an unsolved, high-value, persistent problem where AI is the best fit tool as a means to this end.

This article first appeared on the 11th edition of Tech City News’ popular tech magazine, which focused on the topic of artificial intelligence. You can read the magazine online here. Subscribe to receive future editions delivered to your door for free.