At Microsoft’s annual developer event Ignite, last week, every session touched up on the use of artificial intelligence (AI) in software products. David Carmona, general manager of AI at the IT major, seemed very much the man of the moment. In an interview, he dwells on how pervasive the concept has become and how he sees the future. Excerpts:

Would you have predicted 20 years ago we’d be where we are now with AI?

We have been suffering from winters of AI for many decades now. Many times before we have said, ‘this is the moment of AI’ and then suffered a winter. The difference between now and earlier is the compute power, data availability and the progress we have made in algorithms; we are getting real business results that we could not get 20 years ago. This helps start a cycle of investment from companies. That keeps driving AI to the next level.

In the next five years, do you see AI with capabilities not seen now?

The two new frontiers we will breach with AI is one, enabling more people and democratisation of AI for developers. That is has begun to happen. We will aim to enable every user and employee with AI.

Second, as we increase compute capability around AI and as we increase research in areas such as machine teaching, we will see AI go beyond not so much in the realms of data scientists to business users. For example, instead of training algorithms with large data, we will teach algorithms as we teach humans, at the concept level. Currently, how computer vision works is you provide 1000s of images of a chair, and then the system will eventually learn to recognise a chair. Now, if you teach it that what has four legs, with a certain shape is typically a chair, the system will understand the concept and dissect the problem in smaller pieces. It’s like teaching humans. It is more efficient, needs less effort as we don’t need that kind of data.

What is AI to your users now?

Two ways of looking at it. The technical definition looks at technology that enables AI, new advancements as machine learning and deep learning… As a concept, it is broader than that. It is about how machines complement humans in their capabilities. Split it into three fundamental areas: perception — ability for machines to see the environment around them or computer visions; understanding those things in the world — not just perception but turning that data into knowledge; and three, enable a way to communicate back with humans, with concepts like natural language processing.

We have seen major progress in the same techniques we have been using for a long time now, because of computing power, progress on algorithms such as deep learning, and the availability of data.

Cloud is not only bringing compute power but is making it available to anyone in the world. A cluster of GPUs (graphics processor units) is available to anyone for training an algorithm.

AI as we understand today is focussed on training data. Deep learning depends on huge amounts of data that it uses to train. Availability of data has exploded, not only in enterprise but in the world in general.

How do you help a first-time AI customer?

We first help infuse AI in existing business applications. Business applications can be improved with AI. We provide that ability to developers, who can use pre-built and customisable AI services, called cognitive services in our cloud platform Azure. This is a short-term activity and gives quick RoI. Expectations are that 90% of business apps would get AI capability by 2020.

This is in five areas: vision (image classification) – popular in enterprise for detecting defects in manufacturing, and the like; Speech to Text, Text to Speech; natural language understanding; knowledge, and search.

At the next level, AI is applied to your business processes, not just the apps.

Every business process can be improved and refined with AI. For example applied in sales, it can it help seller identify leads and to manage those leads; our customers use AI to improve customer support and services.

You can develop your own tools with Azure Machine Learning (ML). One of the things we announced was automated ML.

It helps ML happen through ML. The machine learning capability helps generate machine learning models.

ML is a very manual process. You first need data, and then you identify the right ML algorithm There are 100s of algorithms and it takes time and effort to test and identify the one best suited to your need. So you provide the data set, and the system will automatically find the ideal algorithm using ML and then fine-tunes that algorithm. The resulting rise in productivity is just great. You even get improvements in the results of those ML algorithms You can use these without needing a lot of expertise on those algorithms.

There are 1.2 million developers using our cognitive services, while 300,000 use conversational AI.

If someone has spent 15 years in .Net and knows Azure, can that developer graduate to AI programming?

That developer can do AI. We released ML.Net last year. It’s a framework for .Net developers to do machine learning on AI ; all with the same tools they are familiar with. They need additional training and we provide that.

(The writer was in Orlando at the invitation of Microsoft)




Please enter your comment!
Please enter your name here