One of the key challenges in scaling AI revolves around the need to foster conditions for the trustworthy adoption of new types of big data-driven algorithmic processing. If we want the Internet of Things (IoT) health revolution to realise its full potential in the long term, patient data provenance and integrity, as well as that of their processing, need to be guaranteed to minimise undesired bias and cyber-security risks.
Medical devices will accelerate patient connection with healthcare providers and self-manage pre-diagnostics and part of the treatment due to intelligent solutions. Nevertheless, this future would require new data sharing models, preserving our data privacy and security, and eventually handing control of our health data back to us. Relying on Distributed Ledger Technologies (DLTs) such as the blockchain will offer such opportunities in fostering AI-driven solutions; healthcare could be the first to benefit from this.
The rise of the IoT
The semiconductor industry has been religiously following Moore’s law and kept on delivering its promise of doubling the integration level and performance of electronic circuits every two years or so. Combined with a massive increase in volume production and an aggressive competitive landscape, it has resulted in a significant component’s price erosion. Consequently, we have observed the arising of far more diverse and capable electronic equipment (features, processing speed, and memory storage) at a lower manufacturing cost across all industry segments.
The IoT consists of three main semiconductor functions: processing, sensor and connectivity components. Therefore, the IoT “revolution” is a logical consequence of the aforementioned, together with a major improvement in connectivity network infrastructure (Internet and Telecom). Once tens of billions of newly connected smart devices begin communicating all of their different kinds of data, it will by far surpass the volume we witness today. The era of pervasive IoT has started and also applies to healthcare, called the Internet of Medical Things (IoMT), which will bring positive contributions to lower healthcare costs in the years to come and improve treatment and services to patients.
AI, a general-purpose technology in need of a sustainable model
Similarly, the recent acceleration and excitement around artificial intelligence have been the result of converging trends: the growing flows and stocks of high-resolution data, increasing availability of high-performance computing power, advances in machine learning – especially deep learning across domains such as computer vision or natural language processing – and high-speed communication networks providing widely available large data bandwidth. From that perspective, AI can be analysed as a general-purpose technology serving as the pivot point for the enablement of the next phase of the digital revolution.
So far, the industrial added value chain of AI implementation has relied massively on that of cloud computing. It has enabled the rise of a good amount of innovations and solutions in autonomous mobility, medical diagnosis assistance, facial recognition, and virtual personal assistants. But this is coming at the expense of a model that is neither sustainable nor scalable since it is assuming unlimited computing power, data storage and networks resources. Looking at the current excessive volume of largely unstructured data generated, and projected volumes ahead of the deployment of billions of IoT devices in the coming years, it will be necessary to call for a newer more complementary approach and model and decentralise part of the value chain of AI computing.
Decentralisation as an alternative to the AI cloud-based approach
By 2020, up to 44 Zettabytes of data (44 billion Terabytes) could be generated daily. Such an unprecedented amount of data calling for continuous transfer, processing and storage will require far more infrastructure resources to mitigate the unavoidable effect of saturation and the limitations of cloud computing. Several other factors, like real-time requirements (minimum system latency response) for a critical application, or better user experience and the necessity of increased security, are colliding and highlighting the limits of current cloud-based AI as a solitary approach.
Therefore, a decentralisation of AI cloud computing at the edge level, materialised by one or multiple gateways (a communication network concentrator aggregating multiple sensors and higher computing power) and intelligent nodes (smart sensors able to run deep learning inference) would allow a more scalable and balanced AI framework implementation.
This complimentary solution approach would provide the following benefits:
- Improved real-time processing to ensure lowest-latency response with local autonomous action from systems (safety will be imperative in autonomous vehicles).
- Data privacy and security (less or no sensitive data shared over networks, insulated or self-contained system)
- Power consumption optimization for longer-lasting battery (battery-operated devices)
- Data sorting, filtering, pre-processing at Edge/Node level before Cloud (limit unstructured data and offload cloud processing)
- Improve efficiency usage of cloud connectivity based on availability and bandwidth
- AI’s processing load optimization between cloud and edge devices
A decentralised system could provide an answer to the current negative points of healthcare, such as the lack of adopted open standards for health data, the fragile interoperability, lack of handheld personal monitoring devices, and the unsuitable security architecture. For instance, European companies have started promoting and offering decentralised alternatives; the Slovenian start-up Iryo is one example amongst others.
Use distributed ledger technologies (DLTs) in healthcare
Relying on DLTs such as the blockchain could help us, not so much to secure the data itself, but to ensure its integrity, i.e. its immutability regarding potential tampering, alteration or removal. As processing relies on more complex, powerful and opaque algorithms like the now-famous deep neural networks, which mobilised hundreds of parallel layers of processing, DLTs can also play an instrumental role in creating the conditions for better traceability and explainability of the decisions made on the basis of advanced algorithms: healthcare could easily take advantage of such a system.
In healthcare, the use of DLTs associated with an emerging ecosystem of new legal frameworks can also power a wave of new data-sharing protocols, which are crucial to enabling the secure and free flow of data at scale. The emerging framework of “data trusts” is a good example: a bottom-up contractual mechanism whereby a very large number of data subjects choose to pool their data within a trust, and delegate data management rights (structuring, labelling, processing, storage, and portability) to professionally trained managers according to predetermined terms and conditions and a strict fiduciary duty.
Some practical example of the combination of the previously mentioned elements (IoT, AI, blockchain) can be found in the healthcare domain with a new generation of wearable devices. These devices are collecting the raw data from embedded motion sensors (accelerometer and gyroscope) and run deep learning inference to perform, for instance, the classification over different Parkinson disease’s patterns, especially the freezing of gait often responsible for patient falls. Thanks to its real-time processing capability (local deep learning inference), it could trigger a buzzer milli-seconds prior to the occurrence of the freezing of gait and help prevent the fall. The patient’s data could also be shared among different stakeholders (patient, doctor, hospital, clinic) through a secured and immutable ledger framework using blockchain technology. This will reinforce the trust among the various stakeholders of this ecosystem and allow new business models and incentives for end-users to be more proactively willing to share their data.
Franck Martins & Nicolas Miailhe – STMICROELECTRONICS & THE FUTURE SOCIETY