The AI market continues to evolve at an unprecedented pace. In 2019, more than $36 billion was invested into AI startups, and hundreds of AI companies were acquired. In this blog, we highlight six key trends in the AI landscape and propose what they mean for 2020.
1. North America remains dominant in broad AI innovation, but geographical adoption of AI in specific technologies and sectors varies significantly
What happened in 2019: Countries like the U.S. and Canada remained powerhouses in AI research. Although many were predicting China to take over the AI industry, our analysis suggested that this will not be the case in the foreseeable future. Indeed, most of the buzz around AI in China narrowly focuses on a few platform technologies like computer vision and voice recognition and a few industry sectors like finance and government where there is significant activity. Likewise, countries like the U.K. have rapidly adopted AI in specific sectors like healthcare.
What it means for 2020: Many countries will continue to set up AI initiatives and set aside funding for AI; however, most of these efforts will focus on the adoption of existing technology rather than generating fundamental research. Given this dynamic, and that leading in the AI field requires both technical innovation and adoption, there is unlikely to be a significant change in the broadly dominant countries in AI.
2. Enterprise adoption of AI gains steam
What happened in 2019: With companies outside of the technology sector, such as McDonald's and Nike, making major AI acquisitions, it's clear that enterprises from nearly every industry are quickly embedding AI throughout their businesses. Such rapid adoption has led to significant traction of leading enterprise AI companies like DataRobot and H2O.ai, which raised $206 million and $72.5 million, respectively, in 2019 alone.
What it means for 2020: AI adoption will continue to grow, but as many companies transition from the research phase to production, scaling AI and all the challenges that come along with that will become major bottlenecks. Companies should expect AI platforms that offer the ability to operationalize and monitor AI systems in production to emerge and gain traction; on the other hand, generic AI platforms with capabilities focused on prototyping AI will see commoditization.
3. Cutting-edge AI research leads to significant advances in natural language processing and robotics but doesn't come cheap
What happened in 2019: In 2019, we saw major developments in the field of natural language processing (NLP) using deep neural network Transformers for language modeling, which are quickly moving into real-world systems like Google Search. Robotics applications also saw impressive results, including OpenAI's creation of dexterous robotic hands capable of solving a Rubik's cube. These developments have come at a substantial cost, however. Google's AI research arm DeepMind was reported to have lost $570 million in 2018, while OpenAI raised $1 billion from Microsoft, which it plans to spend on research over the next few years.
What it means for 2020: While AI research can be expensive, such developments rapidly become democratized for wide use given the open nature of the field. Whether it's for knowledge management platforms or customer support chatbots, nearly every company will begin adopting new NLP capabilities in its organization. Intelligent robotics applications will begin to show more impressive demos but will still be limited to fairly narrow domains. Multimodal learning, or the ability of machines to learn from multiple types of data – including text, images, and structured data – will become a key area of research.
4. The AI chip battle heats up
What happened in 2019: Companies like Tesla showed the value of developing custom ASICs for AI applications like autonomous driving. Startups in the hotly contested space of AI processors began to ship their first silicon, and new architectures such as optical computing saw an increase in funding. Activity in 2019 ended with a bang as Intel acquired Habana Labs for $2 billion in December.
What it means for 2020: The computing demands of deep learning applications will continue to grow faster than the chips on the market can support. Every cloud provider will begin offering AI ASICs to customers, and there will be other major acquisitions of AI chip companies in 2020. Companies in nearly every industry will start investing in new hardware for edge computing as they encounter the limitations of software-based edge computing.
5. The limitations of AI systems become apparent
What happened in 2019: Most companies in the autonomous vehicle race pushed back their predictions for when fully self-driving vehicles will become available. Companies in other industries, such as pharmaceuticals, that overpromised the current capabilities of AI saw a drop in their valuations. Additionally, new research in the development of challenging datasets showed deep learning's inability to generalize far beyond similar data to what it's seen before.
What it means for 2020: The hype behind the terms "AI" and "machine learning" will slightly wear off. Despite continuing developments in the space, companies will need to take a sober look at what AI can and can't do today. AI still struggles in new environments that are unlike the training data and can't move far beyond pattern recognition tasks to those that require deeper levels of reasoning. Therefore, companies like Tesla that are promising fully capable self-driving systems in 2020 are highly unlikely to hit their goals.
6. Bias in AI models and privacy concerns make headlines, driving the need for trustworthy AI
What happened in 2019: Major news events like the potential for gender bias in Apple's credit card spending limits and unflattering attention around workers listening to recordings of speech recognition systems have put central AI issues into the mainstream. The lack of transparency from companies developing AI products is creating a negative customer sentiment that may significantly impact the use of machine learning systems going forward.
What it means for 2020: This will become the year when trustworthy AI makes significant headway. As AI moves into increasingly impactful applications, companies will demand features like explainability and interpretability in their AI systems. Companies need to assume that every dataset they work with contains bias and will need to vigilantly fight against it. Moreover, companies will need to be able to quickly prove that their systems are unbiased, driving demand for tools like Fiddler.ai. On the privacy side, the growing backlash from consumers will create an opportunity for companies like Sonos, which are taking a privacy-first approach to AI development. While many companies are taking an increased interest in privacy-preserving AI techniques, the landscape is still in its early stages and will require companies to tie together multiple solutions.
From these six trends, it's clear that the AI landscape is filled with both challenges and opportunities. We'll continue to analyze what happened in 2019 in an upcoming research to dive deeper into these challenges and identify more specific opportunities for 2020 and beyond.
- Download the Executive Summary: How to Avoid AI Failures
- Download the Executive Summary: Key Players and Business Models in the Connected Car