Tractica spent two engrossing days at the AI Summit in London last week. Clint Wheelock, Managing Director at Tractica, and I were fortunate to chair the two conference session tracks during the two days, giving us a chance to interact with many of the speakers and companies at the event. Tractica is an exclusive research partner with the AI Summit global conference series, which has become one of the leading enterprise artificial intelligence (AI) conferences to attend.
The presence of more than 1,400 delegates underscored the point that AI is indeed one of the hottest topics in technology and business today. The majority of the attendees had a decent understanding of AI and were looking at specific information and best practices around applying AI in their own domains. Most conference attendees have been witness to companies like Google, Amazon, Microsoft, and Facebook embracing AI over the past few years. These companies, also referred to as “hyperscalars,” are at the leading edge of AI, driving research breakthroughs and investing heavily in building the biggest data centers with best-in-class AI hardware, and hiring the brightest minds in AI to stay ahead of the competition. Most of the money spent by the hyperscalars in AI is linked to improving existing products and services targeted at the business-to-consumer (B2C) market. Improved search results, product recommendations, movie recommendations, speech recognition, photo tagging and search, and chatbot assistants are just some examples of where AI is impacting consumer-facing, internet-era services.
Artificial Intelligence Spending Forecasts and Market Sectors
Tractica’s latest AI forecasts show that the consumer sector represents about 29% of the $1.4 billion AI software spending in 2016, representing the biggest piece of the pie. The rest of the market, leaving out defense at around 9%, is what we call enterprise AI. This includes industries like healthcare, business services, finance, investment, logistics, advertising, automotive, telecommunications, retail, oil and gas, media and entertainment, legal, life sciences, government, gaming, fashion, energy, education, construction, building automation, agriculture, transportation, sports, real estate, and information technology. These industry sectors represent enterprise AI spending, which turns out to be a highly-fragmented market, with sectors like advertising, automotive, retail, finance, investment, telecom, and business services seeing the bulk of the activity today, although there is no clear winner. These are traditional enterprise sectors, many of which are still on the path toward digitization, and now are faced with the task of implementing AI.
The good news is that Tractica’s forecasts are validated by what I heard and saw at the AI Summit last week. Looking at the mix of attendees and speakers at the AI Summit, and the level of AI adoption within enterprises, it is good to know that our research is aligned with realities on the ground.
AI Summit Themes
To add more flavor to the forecasts, here are some of the key highlights and themes that I gathered from the event:
- AI Transformation Is Separate from Digital Transformation: Digital transformation, for the most part, has been about using digital tools like social media, mobile platforms, and the web to help companies better align to the changing nature of the customer, and therefore, it is better understood in the B2C context. While there is a business-to-business (B2B) aspect of digital transformation in terms of improving workflows, processes, supply chains, and ecosystems, those aspects take lower priority. This is well highlighted in a recent McKinsey study on how companies are implementing digital across their organizations, with customer marketing being the number one digital transformation area. Many organizations are still coming to terms with understanding what digital means, and how it could impact their business, but are now faced with the emergence of AI, which frankly seems to be a bigger priority area. While it is clear that, without digital, it would be hard to implement AI, do not confuse one with the other. If a company is on a path to digital transformation, it should ensure that AI is part of the ongoing process, rather than something that is done subsequently. The key difference between AI and digital is that AI is purely driven by data and represents the intelligence that comes from data, while digital only represents the digitization of information, which is the first step toward implementing AI. Also, AI within an organization is likely to extend across both B2B and B2C processes, possibly making B2B a higher priority.
- Start Lean and Fail Fast: There were many speakers from banking and financial institutions, including Citibank, J.P. Morgan, UBS, ING, BNP Paribas, and others. Most of them have small teams in place to explore the application of AI. The applications in finance and investment are quite varied from customer service and engagement, investment and trading, cybersecurity, regulatory compliance, operations and employee services. Most of them prefer to start lean and small, taking fewer risks and having a “fail fast and break things” approach. While this is straight out of the Silicon Valley playbook, it also makes sense in large traditional corporate institutions. Banks are highly regulated complex organizations, so I can see the hesitation in being cautious in applying AI. Furthermore, for any organization that is large and complex, such as healthcare, insurance, or even government entities, this is a good mantra to follow, as it reduces risk and allows one to experiment with multiple approaches to implementing AI.
- Focus on Big Ticket Areas for AI-Led Cost Savings: Simon Thompson from British Telecom provided a great tip for anyone just starting to think about using AI in their organization. If cost savings is what you are after, focus on the big dollar items that cost “hundreds of millions” or billions of dollars. If you can use AI to make an existing process more efficient, and extract a few percentages of savings, a few percentages of a billion is better than a few millions. Anyone who understands decision-making at the C-suite level should resonate with this approach, which is likely to improve the chances of AI gaining traction within an organization.
- AI Will Help Transform Business Models: Although AI was largely seen as a cost savings tool, or an extension of robotic process automation (RPA), there were a few speakers who saw AI as part of a bigger business model transformation. For example, the banks are seeing a lot of disruption down the pipeline as banking products become commoditized. For them, AI becomes a key differentiator to transition them toward becoming a “cognitive bank,” one that better understands and anticipates customer needs and aligns its internal workings to adapt quickly to dynamic environments. Insurance companies also see a wider applicability of AI than simply being able to do faster claims processing or spotting fraud, helping them become “AI first” organizations. Over time, insurers will create individual insurance plans for each of their customers, moving them away from pooling customers in different bands, giving them more flexibility and adaptability. An AI first insurer will be able to better predict illnesses, disasters, crashes, etc., changing the nature of underwriting itself.
- Machine Reasoning Might Be More Applicable than Deep Learning for Certain Cases: Most enterprises are finding it difficult to compete with Google or Facebook in developing deep learning talent, which is a highly specialized and niche area of AI that also requires big investments in data storage and processing. However, there is another branch of AI that might be better suited to the “baby steps” approach. Whereas deep learning is fundamentally a machine learning approach built on probabilistic learning models performing pattern matching on the data, machine reasoning takes a logical or memory-based approach to understanding data. Deep learning improves as the amount of data increases, and if the data set is relatively small like an Excel file with a few thousand rows of data, then machine reasoning might be a better bet than deep learning. Tractica provides a separate forecast for machine reasoning, estimating it to reach $1.8 billion in software revenue by 2025, with use cases in defense, manufacturing, healthcare, energy, government, education, and finance.
- Ethics and Bias Issues Should Not Be Afterthoughts: As systems start to make decisions on behalf of humans, whether it is an AI insurer deciding on a claim, or a driverless car making decisions about avoiding crashes, it is important that these systems are designed with human ethics in mind, while avoiding bias toward a certain demographic or data set. The nature of AI is closely linked to the training data; therefore, avoiding bias in training data is a key issue that enterprise AI practitioners will need to tackle. One of the speakers, Dr. Bertie Müller, suggested the use of a federated learning system (similar to a recent paper from Google on federated learning) where end points like a mobile phone perform their own learning individually in a secure, efficient manner and then update the system with new information. This would avoid the risk of a centralized AI algorithm overriding decisions, and in some ways, will also reduce the bias as learning is performed in a decentralized fashion. Dr. Müller also talked about how an ethics certification of the code would be needed, where unless the certification is verified, the code will not execute. Unfortunately, most speakers at the conference hardly mentioned ethics and bias, which is similar to the topic of security in the context of the Internet of Things (IoT), for example. IoT security is, for the most part, treated as an afterthought, and if the recent WannaCry attack is not ample warning, we are closer than ever before to the possibility of a large-scale IoT cyberattack. Along similar lines, unless enterprise AI practitioners start to take AI ethics and bias seriously, we are more than likely to end up with a situation where millions of AI systems are vulnerable, ultimately impacting business and human lives.