Artificial intelligence (AI) has made the leap from a distant science fiction concept to a focal point of investor interest in recent months, in tandem with rising venture capital investments and media coverage of AI breakthroughs. We caution against much of the hype and overexcitement, as we believe major AI applications are still in their infancy when it comes to mass adoption. This is also supported by Gartner’s ‘hype cycle’ for emerging technologies, which is placing AI and related technologies at the ‘peak of inflated expectations’ phase.

Nevertheless, actual public AI use cases have been low-key and have ranged from improving online searches and product recommendations, to fraud detection and facial recognition. This is because the line between ordinary and AI-powered software has blurred and will continue to do so. Advances in AI technology are increasingly becoming the solution to mastering the Big Data challenge, providing companies an edge over their competitors. In this light, we believe that AI is undervalued in the long term, as the accelerating progress being made cannot be ignored, with cognitive AI increasingly being embedded in products and services going forward and being provided through the cloud.

AI currently covers a ‘narrow’ range of intelligence
Artificial intelligence is a subfield of computer science which aims to mimic human intelligence by processing data and autonomously drawing conclusions and/or acting on information. The main goal of AI is to make machines smarter and more useful. The concepts of AI and machine learning are several decades old, but only recently have the enormous datasets aligned with computing and storage capabilities to harness their full potential. This has enabled AI to cover a ‘narrow’ range of intelligence, which allows it to excel in specific domains at a human level or above. Examples include playing games such as chess, Go or poker, medical imaging or cybersecurity applications. Research efforts are now focusing on creating ‘general’ AI, which could be able to perform a broad range of tasks and apply knowledge to solve unfamiliar problems without being specifically trained in those tasks. However, we are currently far away from reaching ‘general’ AI.

Deep learning has helped the progress of AI
Major leaps in machine learning and deep learning capabilities have been one of the catalysts behind the current advances in ‘narrow’ AI. This has also led to new breakthroughs in image and speech recognition, which reached human parity levels in 2015 and 2016 respectively (see chart). Deep learning has been around for decades, but three factors have shifted over the last five to ten years which will continue to shape the developments of AI: 1) data abundance; 2) increasing computing power; and 3) open-source AI leading to improved algorithms.

1) AI is a tool to master the Big Data challenge
Immense and growing amounts of data are created by connected and sensor-packed devices and machines (the ‘Internet of Things’), as well as smartphones, videos and social networks. Machine learning becomes more effective the more ‘structured’ data they have, meaning that as the amount of data increases, so does the number of problems the approach can solve. We believe high-quality ‘training data’ (text, image, audio, user activity, etc.) is the new competitive advantage in AI. Unstructured data (i.e. unlabelled, low degree of organisation) is ineffective for training deep learning algorithms, but also expensive to store in large quantities, as traditional storage systems are struggling to keep up with their explosive growth. In earlier years, Alphabet (known as Google) and Microsoft had to use human labour to label sample image datasets. Nowadays, raw data is immediately labelled by natural language processing or speech and image recognition algorithms.

2) AI advances thanks to cloud computing and GPUs
The general availability of low-cost computing power, particularly through cloud computing services and new machine learning approaches, has dramatically increased the speed and accuracy of AI applications. Graphics processing units (GPUs) are also facilitators of deep learning and AI advances: they originated from the video game industry and have been repurposed to run computationally- intensive functions. The parallel computing abilities of GPUs has shortened the training time of a machine learning model by about eight to nine times according to NVIDIA, an American technology company which designs graphics processing units (GPUs) for the gaming and professional markets. This contributed to the improved results seen in speech and image recognition.

3) Open-source platforms improve AI algorithms
We have seen a clear acceleration in fundamental AI research, thanks to close collaboration between academia and industry. Large technology firms have recruited top researchers to lead their internal AI research and development departments, while AI tools and computing power have been made accessible to the public via open-source platforms. This has significantly lowered barriers to entry, especially for universities, to study more complex problems like natural language processing and neural technology.

AI will be offered as part of cloud services
For all three drivers listed above, cloud computing will be a key catalyst for the continued adoption of AI. AI-enabled applications will run in the cloud, be able to process enormous amounts of data at manageable costs and result in insight gains that allow companies to better understand their customers, potentially boosting the return on investment. The flexibility provided by clouds might allow companies to change their products and services much more quickly than legacy companies, speeding up the pace of disruption. We see AI use cases across industries, ranging from transportation to healthcare, advertising to finance, with the potential to impact profit pools totalling hundreds of billions of US dollars over the next decade. Currently, we see the greatest near-term impact from AI in improvements to human and machine productivity, while creating and potentially sustaining durable competitive advantages for firms that leverage the new technologies.

We believe the largest dataset owners and collectors currently have an edge in AI. Basic AI tools will be leveraged by all companies to boost productivity, but these basics are likely to become commoditised by cloud computing software vendors. Investors should focus on companies that are able to reap competitive advantages in AI applications via proprietary data, as Big Data is becoming a key differentiator. In the early race for dominance in AI, we see numerous AI enablers and platforms arising. Currently, we favour the integrated cloud computing providers and cloud software companies. These firms have been steadily investing in AI and related assets. Many cloud software companies will leverage AI platforms to deliver new insights and ultimately improve productivity and overall software efficiency.