AI + IoT = AIoT, Multimodal Intelligence and Other Trends to Watch
2019 was an interesting year for the technology industry – many technologies were hyped heading into the year; however, their paths of adoption took different directions:
- Adoption of AI/ML continued to increase, with newer breakthroughs and areas of application – especially in biology, medicine and research.
- 5G, while discussed broadly by technology vendors and industry professionals, encountered a few challenges with infrastructure, security and adoption.
- Enabling intelligence on the edge gave rise to a new term AIoT – renewing vigour and focus towards enabling the next generation of industrial automation.
With these trends in mind, 2020 is expected to be a year of convergence and course correction for younger technologies, with an increasing trend towards sustainability and greener solutions.
Blockchain’s Niche in the Secure, Distributed Data Store
While 2019 saw quite a few interesting applications of blockchain, there have been two main challenges to its adoption: 1. Lack of standardization (platforms, specification, interfaces, etc.), and 2. The fact that benefits of blockchain are realized once a majority of the collaborating providers in a chain all start using the same – or interoperable – platform(s).
The current major players in the platform space all have their own standards for their products – design, components, contracts and implementation – thereby tying an early adopter down to a single product. This lack of standardization has been an area of significant focus/attention recently.
The International Organization for Standardization (ISO) and the Institute of Electrical and Electronics Engineers (IEEE) have both started standards initiatives that would be ready by 2021 – and we would expect the platform providers to start supporting these standards once they hit early access (hopefully by 2020).
In parallel, enterprises today have started adopting blockchain in a phased manner – the approach now, from the enterprise PoV – is to design the to-be state of the information architecture in a future-proof manner (keeping the application code as product-neutral as possible), and realize the true benefits of interoperability and data-sharing once the partners start their implementations.
With the purchase of a few blockchain products by the existing stalwarts in the market, cloud support and integration with other existing technologies are also on the rise. With all the above, we think in 2020 blockchain will enter mainstream adoption as the distributed store of the future.
AIoT and More Sustainable, Greener Solutions
2019 saw an increase in an infusion of IoT into existing scenarios – with most of the challenges around adding IoT/sensor capabilities and enabling intelligence on the edge to be resolved (this fusion of IoT and AI is known as AIoT). While the original purpose for enabling these capabilities may have had to do with early prediction of faults or optimizing usage patterns for efficiency, the large volume of data now available from these devices and sensors has opened up new avenues of exploration and optimization.
The evolution of IoT into AIoT progressed in three distinct stages:
- Enabling core capabilities on the edge – these included basic sensor development, integration with available devices, etc.
- Collecting the data generated from these sensors and storing them in a structured form on a central data store – typically on the cloud
- Realizing the synergy between AI/ML and IoT and combining them into AIoT
Focus in this area has also been evolving along with the core technology itself – shifting towards applications of AIoT and away from initial device capabilities and integrations. In other words while IoT provided access to a large base of information (“here’s the data”), AI/ML has brought in the intelligence and decision making (“here’s what you can do with it,” and “here’s where you are inefficient”).
It’s expected that this year will see continued focus on AIoT adoption, combined with the ability to move decision making to the edge, to drive a responsible, sustainable and greener approach to energy consumption.
From “Narrow” to Multi-modal Intelligence
2019 saw an increase in adoption of AI/ML solutions in newer and previously unexplored areas. This will continue into 2020 as algorithms become more intelligent. However, the scope of existing machine “intelligence” is still too narrowly focused on single objectives. To put it simply, the engine classifying a picture of a “cat” does not really understand what a cat is. This is semantic information that is understood by a different “narrow” engine that only understands the semantic concept of a cat.
There are already efforts underway to create multi-modal intelligence, exploring combinations of natural language with visual cognition. The goal is to expand narrow-ness of an AI solution, and to enable transferability so that we can prove understanding – in the example above, that would mean that the same algorithm will eventually be able to recognize a picture of a cat and understand the semantic details of the feline – much like how humans think.
We believe that in 2020, AI will shift towards multi-modal intelligence – such an achievement would open the doors to many more uses for AI/ML in future.
While 2019 has seen an increased adoption of AI/ML, there have also been unintended consequences – incidents leading to a “trust crisis” regarding decisions put forward by algorithms. Algorithms trained on data captured over the past few years naturally reflect biases inherent in the data – when we evaluate the predictions generated by these algorithms through an evolved values set, we suddenly end up discovering the inherent data biases – but somehow we blame the algorithm as biased! As a result, making AI/ML solutions interpretable has been an area of intense interest. If we can understand or interpret the steps an algorithm took to arrive at a decision, we can decide the limitations of the algorithm itself, or the missing gaps in the data that the algorithm was trained on.
In 2020, we will see two things help to address these limitations. First, we will see increased regulatory support to ensure AI/ML follow certain anti-bias principles. Second, solutions will be built to give an outside-in view of black box algorithms, helping humans better understand algorithms and the decisions they recommend.
Rajamani Saravanan is the chief architect at Mindtree.