Digital assistant Siri responds by recognising speech. The more it is spoken to, the more competent it becomes. Speech recognition has lesser error rate today. Digital assistants recognise even accents. All this has happened over a period of time. To accelerate this process, one has to feed great amount of data and compute power to the system. There are instances where this has to be done instantly. AI assisted systems do this. In a car braking system of an autonomous car, there should be an instant response.

These days chips are made with AI acceleration. Semiconductor industry is developing such chips. Big Tech too develops such chips. India is in the forefront of designing such chips.

An aircraft in flight generates a lot of data, which is to be analysed in real time to detect if something is failing. In plants, a lot of equipment generates data that is to be analysed. It is humanly impossible to do this in real time. Here some data is processed by edge computing and some at the server. Thus it is a hybrid system. Intel’s new processors with AI acceleration address such kind of applications. In software such analytics is not fast enough. Even GPU are used for graphics processing. They too fall short of the emerging requirements.

Such situations require an ML algorithm running on hardware. A software programme launched on lap top competes for time from the processor (CPU), as there are other programmes running. If the information is hardware-based, the hardware does it immediately. Since that is what it is designed to do.

Different functions require different AI enabled chips. IBM’s Telium processor has on chip acceleration for AI is targeted to financial services. Whether to use a chip designed for a specific AI application or a general purpose chip is taken in the light of the software capabilities of an organisation. A specific chip requires more software capabilities.