In our recently-concluded event MLDS 2019, one of the speakers quipped that AI is the tool that is making predictions cheaper. When machine learning models move from modelling phase to the production phase, many issues surface. The goal of democratisation of AI is enabling every individual with access to AI applications.
How Leading Tech Giants Are Democratising AI
The key functionalities of AI include vision, speech, prediction or forecasting. Many organisations now deploy vision and speech systems like a routine. But building recommendation engines requires both knowledge and tools. The knowledge of building a pipeline and the tools to handle data ingestion, storage, security, access and deployment.
What Amazon did was, they democratised recommendation engines. They allow interested entities to exploit Amazon’s recommendation pipeline. All the user has to do is make their data available to be ingested into the pipeline and run it through a custom API, built by Amazon.
The output, which is usually the customer behaviour prediction can be looped back into the first stage of data collection and improve the pipeline automatically without the user having any knowledge of the workings of the recommendation pipeline.
Amazon has solved the problem of forecasting on a very large scale and democratised it. The big billion days or the ‘prime’ days witness a surge in orders, to be delivered within two days. For such fast deliveries, the warehouses need to be stocked prior to the D-day.
Algorithmic Forecasting: This could be done only by algorithmic forecasting to get an estimate of which commodities will be high on demand and also what would a certain customer from a certain location would buy. Amazon has marvelled at this game for over two decades and for the past couple of years it is outsourcing their platforms for other customised purposes.
APIs: Instead of reinventing the wheel, new, higher level APIs can be developed by organisations without getting themselves into all the trouble with recommendation pipelines. This allows the companies to concentrate their resources on more pressing issues. In a way, a quicker more efficient way of launching innovations into the market ensuring an accelerated overall growth of the society.
We sometimes say that pretty much any relationship that happens in the world, happens via AWS platforms; whether it is Shaadi.com or Tinder, it runs on AWS,” quipped Madhusudhan Shekar, talking about the implications of democratising AI/ML and unlocking new today at MLDS 2019.
Another key rule to be considered before building any model is about how safe the platform is and whether the datasets being used comply with GDPR, whether the metadata is anonymised properly. Since anonymity will, in turn, generate bias in the machine learning models.
To guarantee information security and to keep in check all regulatory measures, Amazon built Sagemaker, which too, is outsourced for customised applications. After a certain point in building models, it is the inference that gets expensive and not training.
Building a Resnet image model would take 30 minutes on Amazon services and generating 360,000 image inferences would cost only 22 cents per hour. That is huge cost-cutting taking into consideration the overall picture.
For The People By The People
Since it is always the tech giants that fuel innovation in this digital revolution, they also happen to be the key people responsible for democratisation.
And, they are doing a great deal already by open sourcing datasets, tools, frameworks and other research that takes in a lot of high-quality man hours to build.
Google, too, with its Google.AI has been teaming scientists and developers to tackle problems across a range of disciplines, like improving the algorithm that detects the spread of breast cancer to adjacent lymph nodes or creating AutoML, to enable neural nets to design neural nets.
Google hopes that by democratising AI and making it for all, will take an ability that a few PhDs have today and will make it possible in three to five years for hundreds of thousands of developers to design new neural nets for their particular needs.
AI is expected to serve a major blow to the labour market in the coming years by taking away one-third of the current jobs. These newly unemployed unskilled labour will passively hinder whatever development these modern innovations promise. Since the surge in tech-innovation can’t be contained, it is only reasonable to make it more accessible.