The age of the ‘intelligent assistants’ is well and truly upon us. Machine learning (ML) has already emerged as one of the key elements of global digital transformation – with cumulative investments (on artificial intelligence and ML) projected to reach $58 billion by the end of 2021. In the US alone, the market for deep learning software will jump from $100 million in 2018 to a whopping $935 million in 2025. The worldwide machine learning industry is growing at a CAGR of ~42% , and will be worth just a shade under $9 billion by the third quarter of 2022.
In the enterprise space too, the growth of machine learning use cases has been remarkable over the past few years. Total enterprise-level adoption of ML tools and solutions is expected to touch 65% before the end of the decade – and spendings will go up to $46 billion (according to a IDC report). On average, 55% corporate CIOs have identified ML as one of the core priorities for business acceleration. Over here, we will highlight how machine learning will continue to evolve in 2019:
Newer use cases of ML are coming up
Earlier this year, it was announced that the US Army will be using customised machine learning software tools (created by the Chicago-based Uptake Technologies) for predictive maintenance of combat vehicles. In other words, ML would be able to indicate when, and what type of, repair services a vehicle might require at any time. This ‘intelligent’ functionality will be powered by advanced sensors embedded in the vehicle engines. Yet another interesting use case of ML is the prediction of stock market fluctuations – based on the records of previous stock earnings. A recent research showed that such stock market predictions with ML have a 60%+ accuracy meter – which is impressive enough. Moving over to medical science and healthcare, ML models are being used to estimate the probability of death of a person (the accuracy in this case is well over 90%). Progresses are being made to expand the scope of ML further, in retail, marketing & sales, and industrial/manufacturing sectors. ‘Reading’ and ‘interpreting’ past data for forecasting the future – that’s the essence of machine learning – and the technologies are definitely getting more refined.
Note: The concepts of AI applications and ML tools are no longer limited to external robots. Instead, they have become natural extensions of business workflows and everyday applications.
Adoption of ‘hardware optimised for ML’ set to rise
2019 might very well be the year when specially prepared silicon chips – with custom AI and ML capabilities – become mainstream, at least for enterprises. The market for AI-optimised hardware will continue to grow rapidly in the foreseeable future. A series of new, powerful processing devices will be launched – and we would also get to see high-end CPUs and GPUs being used. Taken together, these tools and platforms will enhance the usability of ML hardware in a big way. In 2018 Q1, SambaNova Systems – an AI chip startup – raised a massive $56 million in a Series A financing round. By the end of 2025, global sales of AI-powered hardware will cross the $120 billion mark. The biggest of players – from Nvidia and Google, to IBM – are already in the game, and the machine learning hardware market will be one to look out for next year and beyond.
Cloud adoption to rise with ML
A yearly growth rate of ~25% will see the worldwide cloud computing market soar to $410 billion+ by 2020. The growing adoption of ML in enterprises is a key driver behind this surge. For the successful implementation of a ‘machine learning culture’, businesses have to focus on innovation more than ever – with particular emphasis on improved cloud hosting and infrastructure parameters. Over time, more and more ‘AI-specialised tools & systems’ (apart from business critical information and big data) have to be stored on the cloud – and the latter needs to have adequate security and usability standards for the purpose. A robust, scalable cloud support will help enterprises seamlessly move on from machine learning to deep learning, deliver greater value to end-users, and improve their ROI figures.
Note: Starting from 2019, the general user will start to get a clearer idea on how AI and ML processes work – thanks to the detailed ‘AI audit trails’. Given the critical nature of the domains (say: medical science) in which AI is making its presence felt, it is only natural that people would want to know how the technology arrives at its conclusions/predictions.
Moving ahead with capsule networks
For all the merits of neural networks, they often do not factor in the relative orientation or the position of select objects. As a result, ‘information gaps’ might remain in the machine learning models based on them. To tackle this, capsule networks have already arrived – and they are likely to replace many conventional neural networks in 2019 and beyond. In terms of performance, these capsule networks are a cut above the traditional neural network systems – with more accurate pattern-detection capabilities, and that too, with lesser data and a much-diminished probability of errors. What’s more – capsule networks do not require repeated training iterations either, to ‘understand’ variations. The size of the overall neural networks market will be more than $23 billion in 2024, and capsule networks will be right at the center of this growth.
Note: Advanced healthcare modules based on ML algorithms, for the comparison of medical images of a patient with that of others, are already being used. AstraZeneca, a biopharma company, has plans to use robotics and machine learning extensively – for developing smart diagnostics systems in China.
Rise and rise of AI assistants
Siri and Google Assistant and Alexa have become pretty much a part of our everyday lives, right? In another five years or so, the value of the worldwide AI assistant market will touch $18 billion. More importantly, each of the top ‘intelligent assistants’ are becoming smarter, on a year-on-year basis (on the basis of 5000 general questions, Siri managed to answer around 31%, among which nearly 80% were correct responses; in the same survey, Google Assistant answered over 67% questions, with an accuracy of a shade under 88%). With the scope of machine learning expanding, AI assistants are ready to move beyond the smart homes and users’ pockets. From the next year, Hyundai and Kia will start to provide built-in, AI-powered virtual assistant systems in their new car models. These assistants will be able to perform a myriad of tasks – ranging right from remote home and car control functions (through voice), to destination suggestions (based on previous preferences) and navigation guides. In all scopes of life, ‘intelligent assistants’ with ML capabilities will be making lives simpler than ever before.
Note: Smart chatbots (with artificial intelligence) are also witnessing rapidly rising adoptions. There is, however, cause to be wary – since biases in training datasets can cause serious damages in user-experiences. The ‘Tay’ chatbot by Microsoft is a classic example of such a failure.
Developers will focus on solving more ‘real problems’ with ML
When it comes to a fancy technology like artificial intelligence (multipurpose drones and automated surveillance cameras and self-driving cars, and the like), it is very easy to go overboard. However, it is important to realise that – while all of these things CAN become a reality – the steps towards a full-fledged data-driven ecosystem have to be gradual and systematic. In 2019, app developers and AI specialists will be eyeing to use machine learning to successfully address real, important needs (personal and business) – instead of simply churning out new prototypes of deep learning tools. Put in another way, developers have to understand that AI and ML are much more than just a couple of tech buzzwords – and when implemented properly, their potentials can be endless. There are many other technologies that are vying for attention at present (4d printing immediately comes to mind), and unless the developments in AI solve actual problems – investors might start looking elsewhere. It will be crucial to separate the ‘AI overhype’ from the ‘AI facts’, and act on the basis of the latter.
Note: In a recent study, it was found that 89% of all CIOs have plans to implement ML tools and applications in their businesses.
World of the robots?
Okay, that sounds a bit too dramatic, doesn’t it? In truth though, the roles of intelligent robots in workplaces are gradually increasing – and the improvements in ML are the primary cause for that. In Japan, three-fourths of all elderly care services will be delivered by AI-robots by 2025 – replacing human caregivers. Tianyuan Garments – a China-based t-shirt company – has plans to use ‘sewing robots’ at its Arkansas factory. In general, many labour-intensive tasks (particularly the repetitive activities that do not require much specialised skills) will be performed by ‘intelligent robots’ in the not-too-distant future. Apart from making workflows smarter, improving availability and reliability, and shortening the time-to-market, ML-powered robots would also significantly bring down operating expenses (as well as outsourcing costs, if any). Greater productivity should be a direct result of full-blown AI adoption at workplaces.
Note: Machine learning can also play an important role in precision farming. Smart poles for agriculture, with deep-root sensors and dedicated ML module(s), can help farmers take more ‘informed’ decisions.
Voice technology to the fore
Whether ComScore’s prediction of 50% of all search activities to be powered by voice by the year 2020 comes true remains to be seen – but there is no getting away from the fact that speech recognition (and interactions based on that) has emerged as an important element of machine learning. Unlike the early days of voice technologies, present-day speech recognition has a sub-5% error rate – which is more than serviceable. Interactive voice response (IVR) systems are becoming smarter than ever – thanks to iterative learning, and voice-based ML systems have the capability to transcribe a wide range of languages/accents. The trend of developers coming up with voice technology-powered mobile applications is also expected to gain further momentum in 2019. Already, assistants like Amazon Alexa and Google Home ‘understand’ our voice commands – and they are paving the way for more such platforms to enter the market.
Note: The traditional, suited customer service executives are also being gradually replaced by virtual characters. The latter offers more prompt responses – and since the conversation is intelligent (virtual agents learn from previous conversations), the personal touch is not lost.
AI markets in USA and China – the big fight?
North America has traditionally been the frontrunners, as far as artificial intelligence research and adoptions are concerned. This stranglehold, however, is growing weaker and weaker – with the Chinese market emerging as a serious force. In 2017, AI startups in China had a higher equity funding share than their American counterparts (48% vs 38%). The Chinese AI startup scene is holistic (unlike the slight fragmentations in the North American markets) – with the focus being on logistics, smart city projects, retail, healthcare, smart farming, and other domains. When it comes to deep learning too, China is clearly edging it – with 6X more patients issued than in the US. As per reports, China is looking to be at par with the American AI scene by 2020, and emerge as the undisputed leader of ML technologies within a decade of that. It will be fascinating to see how the US vs China race for global AI/ML supremacy pans out over the next couple of years.
Note: Instead of relying on third-party APIs, developers are increasingly turning to making their very own APIs for ML applications. There are plenty of developer-friendly assembly kits and mobile SDKs to provide the necessary help.
10. More machine learning platforms (and better ones too?)
Platforms like TensorFlow, H2O, ai-one and Torch are already making a difference to how ML functionalities can be deployed in different scenarios. In the year coming up, we can reasonably look forward to more powerful ML platforms – with cutting-edge analytics, classification and predictive capabilities. The capacity of these platforms work with other APIs and big data will also continue to improve. The constant developments in machine learning are opening up opportunities for computers and mobile devices to ‘learn’ faster and ‘interpret/analyse’ data in a better manner. In a February 2018 Gartner report, the total available market (or, TAM) of machine learning at the end of this decade was valued at nearly $26 billion.
Note: AI/ML applications are also facilitating automated decision management practices. Informatica and UiPath serve as great examples of this.
11. Revolutionising the way humans interact with technology
They might be present only in a handful of locations at present (<10) – but the ‘cashierless Amazon Go’ stores are completely changing the concept of shopping. In fact, by 2021, more than 2000 ‘Amazon Go’ stores might be present in the US alone. The manner in which we deal with, interact with, live with smart things (in particular) and technology (in general) is being shaped by the AI & ML revolution. Be it for a business, or for the society (read: surveillance cameras, smart city applications) or smart homes – deep learning is set to disrupt our lives everywhere, ensuring better performance across the board. Things that only seemed possible in sci-fi movies and our imaginations have been rendered possible with artificial intelligence. The key here has been the adaptability of the technology for different types of use cases. ML is solving problems and delivering value – and that’s precisely why it is growing in popularity.
Note: The development of ‘killer robots’ for warfare can be, potentially, alarming. A recent report predicted that the ever-increasing investments on AI for military applications might very well lead up to a nuclear war between 2040-2050.
12. NLP to become more nuanced
As a sub-domain of artificial intelligence, the importance of natural language processing (NLP) has gone up significantly over the last few years. By the end of 2020, the global NLP market will be valued at well over $13 billion – with the industry CAGR hovering around the 19% mark. Primarily used for converting data into text, natural language generation is a key feature of many deep learning systems – and for the preparation of detailed market summaries or reports – NLP is extremely handy. The fact that natural language processing has also become highly accurate is also worth noting, and automated systems are being enabled to communicate ideas in a seamless manner. Cambridge Semantics and Attivio are some of the notable companies that provide NLP services.
Note: NLP modules typically need to analyse three things: syntax, semantics and context.
As more progress happens in the world of machine learning and new application areas get unearthed, the demand for AI specialists (rather than tech generalists) will continue to rise. This will, understandably, be accompanied by increases in their average salary figures. There are certain grey areas – like the prospect of mass unemployment and maybe intrusive surveillance – but it is safe to say, 2019 is going to be a big year for machine learning. AI-as-a-Service has arrived!