The Future of AI: What We’re Watching

08/17/2017 Samir Kumar

The current renaissance in artificial intelligence is having a transformative effect in fields as diverse as transportation, healthcare, and advertising. Its underpinnings lie in advances in data-fueled machine learning’s ability to achieve state-of-the-art performance on all sorts of pattern-matching tasks. Innovations in deep learning, for instance, are helping tackle previously intractable problems, while everything from CRM platforms to the devices we use every day are becoming better at recognizing what we are trying to accomplish with them. However, we are still at the beginning stages of this transformation—which is a very exciting place to be from an investor perspective.

The vast potential of AI holds the promise to upend entire industries, change our relationship to machines, and even augment human intelligence itself. As such, Microsoft Ventures is constantly in search of innovative entrepreneurs who are integrating AI into tools and processes with the goal of achieving breakthroughs. We are also interested in supporting startups that harness AI for creating a positive impact on society—a key metric for startups we are looking to invest in through the AI fund we announced late last year.

At the same time, we want to nurture the next generation of AI and machine-learning breakthroughs by investing in startups innovating in AI itself. For example, the quest to close the gap between how our biological brains and machines learn to represent the world is a core aspiration of AI. We firmly believe that breakthroughs here will come through startups, as well as through R&D from both established companies and leading academic groups.

Innovations with AI

AI is currently being developed not only to integrate with and enrich the way we do our jobs, but also to remain virtually invisible in the process. A prime example of this is Tact, a conversational AI platform that can turn a salesperson’s smartphone into full-fledged sales assistant. Utilizing a voice-powered conversational interface, Tact can answer questions, curate information, and predict user needs—all while operating unobtrusively across various business systems in the background.

The practical benefits of unobtrusive AI and machine-learning technology that can seamlessly integrate with existing workflows are wide-ranging, as they significantly improve our productivity by helping us complete tasks more efficiently. Case in point: Agolo is making use of state-of-the-art techniques in natural-language processing to intelligently summarize the written content we deal with every day, such as news and documents. Because we all struggle with information overload, Agolo’s summarization platform has the potential to help us better consume and extract knowledge from written content. Aqua, meanwhile, utilizes machine learning to understand the behavior of virtual containers based on the applications running within them, helping IT specialists automatically pre-configure appropriate security settings. In general, cybersecurity propositions such as this hold strong potential as leading applications for machine learning and AI.

Successfully integrating AI-powered features into existing business processes also requires a robust understanding of customer pain points and user personas. This is key to Cognitive Scale’s ability to operationalize AI across customers in financial services, healthcare, and commerce for a variety of use cases. Element.AI, on the other hand, enables enterprises to reap the benefits of their data assets when they’re combined with Element’s world-leading talent in machine-learning research and custom AI development strategies.

Innovation in AI

While the investment opportunities for new technological innovations that use AI as an ingredient look quite strong, we are also encouraging startups that are innovating in AI and going after significant unsolved problems in the field. This includes algorithmic breakthroughs, improved data platforms, and better software tools/infrastructure, as well as hardware for training and running machine-learning models. The ecosystem dynamics (such as open-source software, open research, constrained access to leading research talent) set a very high bar for startups in this segment that are trying build a sustainable advantage to make them attractive for investment.

Bonsai, which is automating the creation of reinforcement-learning models for industrial automation and robotics tasks, is a leading example of innovation in software tools and infrastructure that catalyzes the democratization of AI. Because there are a far greater number of engineers and software developers with domain expertise in their specific verticals than there are machine-learning experts, Bonsai abstracts the complexities of architecting a machine-learning model. This allows their customers to focus on “teaching” a machine the attributes of an automation or optimization task they want it to learn.

Another example of democratizing AI is by addressing the common and significant challenge all organizations face when preparing data—an important step they must take before they can even begin experimenting with machine learning. For instance, because it is a form of supervised learning, deep learning requires large amounts of labeled data to work. Despite this, correctly labeled datasets are far from common in most companies. To solve this, Crowdflower deploys crowdsourcing that lets organizations label their training data at scale. Customers can then optionally deploy the Microsoft Azure machine learning-powered Crowdflower AI (which is further complemented by human-in-the-loop techniques) to train their models. More generally, we see key opportunities for software-tooling infrastructure to better operationalize machine-learning research into scalable production-grade solutions.

As we look across horizontal technologies, deep-learning-powered AI has propelled computer vision, speech, and natural-language processing to new levels of performance. That said, many unsolved problems remain that hold potential as opportunities for startups. For instance, we are excited by advances in intelligent video understanding in the computer-vision domain, large-vocabulary speech-recognition models that can run entirely offline, and further advances in language understanding. Finding new ways to combine these domains to create multimodal knowledge representations holds tantalizing possibilities for breakthrough experiences. For example, learning how object relationships work in video could lead to better language models, as well as machines that have a certain amount of common sense regarding the physical world.

Finally, computational requirements for machine learning are creating a surge of interest in silicon innovation and novel hardware architectures. While GPUs have been able to supply the large amounts of computational power that AI methodologies like deep learning require, interest in more specialized architectures is growing. This includes everything from the datacenter for building and running large models in the cloud to very-low-power devices for running machine-learning models at the edge. From our perspective, companies developing novel ways to achieve a power/performance breakthrough for AI workloads are well positioned to capture sustainable value.

Building the Foundation

AI is a core part of Microsoft’s innovation agenda across our large portfolio of products and services. We believe this technology not only will change the way we work and interact with the world but can also be directed towards improving society. Efficiencies in agriculture, understanding climate and environmental changes, better education tools to unlocking opportunities for those in the developing world are prime examples of how artificial intelligence can be deployed for social good. We are keen to promote those companies applying AI in those directions.

One of the key benefits Microsoft Ventures brings to companies we invest in is the chance to grow and develop via targeted engagements with relevant research, product, and sales teams within Microsoft. We promise to give our portfolio companies access to the extended and enviable network of Microsoft products, engineers, sellers, partners, and customers. Our long history in AI R&D and highly talented teams in Microsoft Research positions us well to assist companies we invest in. We look forward to connecting with the next generation of startups currently laying the foundation for tomorrow’s AI-first world!