The latest microchips mimic cerebral function. Smaller, faster and more efficient than their predecessors, they have the potential to save lives and help insurers, argues Amarnath Suggu
Microprocessors are the heart and soul of modern-day computing. Their processing power has grown exponentially over time, while their size has diminished. While this processing power has been adequate for most general purposes, the advent of the fourth industrial revolution has set the bar higher. Models and machines powered by artificial intelligence (AI) require more powerful and energy-efficient processors with smaller form factors, but traditional processors have reached their limits in terms of miniaturisation and energy consumption. This has resulted in the need for a new chip design that can cater to the present-day requirements of smart devices and autonomous vehicles.
Neuromorphic computing tries to mimic the functioning of neurons in the human brain, using biologically realistic computational models. It achieves a brain-like function by using electronic nodes that act as neurons and electrical signals that act as synapses connecting the neurons. An input from a sensory organ triggers the brain through a series of electrochemical reactions. Similarly, neuromorphic processors are triggered by discrete events and powered by a series of electrical impulses or spikes. These spike trains, as they are known, are propagated by the third generation of neural networks, called spiking neural networks.
What’s the advantage?
Neuromorphic computing co-locates both the memory and processing together, made possible by the use of special electrical components known as memristors. This architecture eliminates the Von Neumann bottleneck (the idea that computer-system throughput is limited by the ability of the processor), greatly improving the processing speed and reducing power consumption. Event-driven computation makes these processors all the more energy efficient.
The neurons’ architecture enables a high level of parallel processing. All neurons and synapses can work simultaneously on the same or different activities. Multiple neuromorphic processors can be scaled to form one single large network of neurons and synapses, drastically increasing computation power.
Studies carried out by Intel showed neuromorphic processors outperforming traditional processors in terms of processing times and energy efficiency: they were nearly 100 times faster and 1,000 times more efficient. They also learned subtle differences in images, including different human gestures, very quickly when compared with existing processors.
Use in insurance
Risk management is inherent to the insurance business. With the advent of the Internet of Things (IoT) (everyday objects with WiFi capabilities built in) and the abundance of data coming from connected devices, insurers have shifted to risk prevention. By analysing the data from IoT devices, insurers can alert customers to risks in a timely manner, or trigger actions to prevent loss of life or property damage.
Unfortunately, most IoT data is analysed by power-hungry AI models that are hosted on servers. Owing to their size and energy needs, they are usually hosted in the cloud or on the edge (at or near the user or data source’s physical location). As a result, the data has to be transmitted from the source to the cloud or edge. Data analysis outcomes are then transmitted back to the source to trigger an alert or a necessary preventive action. The use of traditional computing processors thus involves delays to the triggering of alerts or preventive actions.
Due to their smaller form factors and power efficiencies, neuromorphic processors enable AI models to be hosted at the source of the data itself. This eliminates the transit delays that exist in earlier models. More importantly, it opens the doors to newer business models and encourages insurers to explore innovative ways of preventing risks and saving lives.
Pure progress
Neuromorphic computing has come a long way since its inception in the early 1980s. Many organisations have successfully conducted experiments demonstrating that these new processors are computationally more powerful and energy efficient than traditional ones, as well as requiring a much smaller footprint.
Neuromorphic processors have improved AI’s processing capabilities and allowed it to be deployed at source. As a result, it can identify unknown and known risks much more quickly than earlier models. Insurers should recognise the advantages of neuromorphic processors and adopt them to prevent accidents – and, more importantly, to save lives.
Big Brother or Good Samaritan?
Some of the many possible everyday uses of neuromorphic chips
- All vehicles – Connected cars and trucks are equipped with cameras, Internet of Things (IoT) sensors and telematic devices. The data from these is transmitted to a server, where an AI model senses distracted or reckless driving and sends a response back to alert the driver and the authorities. Given the speeds at which cars travel and the possibility of a no-signal zone, the alert can sometimes be too late.
- Autonomous cars – Self-driving cars generate nearly 40 terabytes of data per hour and navigate using camera footage. A fully autonomous car (with Level 5 Advanced Driver Assistance Systems) performs nearly 4,000trn operations per second. Neuromorphic processors can make onboard AI engines a reality, helping to process the data faster and prevent accidents in real time while conserving power.
- Drones – Insurers use drones to inspect roof and crop damage, and determine the extent of flood damage. A drone’s flight is usually limited by data storage capacity and visibility, necessitating multiple trips and making it difficult to assess damage quickly. Neuromorphic processor-based AI engines on board can help drones to auto-navigate and process the feed in-situ, so damage can be seen in real time. This could expedite claim processing or underwriting.
- Accessories – IoT wearables collect vital statistics from the human body and upload this data to an edge server (usually a mobile phone) for analysis by an AI engine. Neuromorphic computing enables this data to be processed within the item itself, without draining the battery. This eliminates transit and processing delays, and can let a user know, for example, that they need to see a doctor immediately. This could improve the survival rates of patients with critical heart illnesses.
- Long-term care – Patients in care are monitored, either in person or via a camera, and help is dispatched when required. Neuromorphic processors can identify subtle human expressions, gestures and positions in footage to determine whether a person needs assistance. They can also be incorporated into a robot, which can remind the person to exercise or take medication. This could reduce medical claims and lower premia.
Amarnath Suggu is a senior consultant in the BFSI Technology unit at Tata Consultancy Services
