Take a guess: what’s the hottest tech role in 2019? According to a LinkedIn report the most promising job in the United States is data scientist; the ranking is based on salary, number of job openings and annual growth. The same report lists Cloud Computing and Artificial Intelligence as the two most in-demand ‘hard skills’.
It goes without saying that for a modern business, technology is essential both to the running of that business and for interaction with customers. Such businesses find themselves accumulating an increasing amount of data, which in turn drives the demand for data scientists whose job it is to extract value from that data.
Data enables the use of AI. We can write software that’s capable of learning and train that software using lots of data.
Let’s take language translation as an example. Early translation software was the butt of many jokes, but today tools like Google Translate , while not perfect , are worlds apart from their early predecessors. To train their software, Google have collected hundreds of thousands of example translations made by human experts. These examples are fed into an AI program that learns the underlying patterns in its input and uses those patterns to make new translations.
If you wanted to build your own translation software from scratch then, in addition to writing the software, you would also need to gather training data, prepare it, and use it to train your software. The same applies to other AI applications and this naturally creates a high barrier of entry for somebody wishing to deploy AI in their business.
The three major cloud providers , Amazon Web Services, Google Cloud and Microsoft Azure, have recognised both the demand for AI and the difficulties in implementing it. Because these companies have already made the investment in building things like machine translation, image classification, or recommender systems, they can easily make these services available as part of their cloud offering. Not everybody needs a bespoke AI solution; many problems are readily solved using the existing offerings from these cloud providers.
Amazon, Google and Microsoft are locked in a struggle for cloud market share. Right now Amazon are in the lead but Microsoft and Google are catching up. In 2018 Amazon remained static on 32% of the market whereas Microsoft and Google both increased their share, the former from 14% to 16% and the latter from 6% to 8%. This growth could be attributed to the focus on Artificial Intelligence and Machine Learning from both Microsoft and Google. Which is not to say that Amazon don’t care about AI or that they don’t have a good AI offering, but it’s Google and Microsoft who are really pushing it. Google consider AI to be one of the main differentiators between themselves and other providers.
By offering an AI platform that is easy to use and covers most use cases, these cloud providers are making AI available to everybody. Even so it can still be daunting to get started, so to make things simpler we’ve summarised the offerings from the three big cloud providers into two broad categories.
AI as a Service (AIaaS)
All of these solutions use pre-trained models and offer easy integration methods for you to use them with your existing application. The use of pre-trained models means the training for a specific task has already been performed, like our Google Translate example, which has already been trained to translate across more than 100 languages. Other examples include image classifiers, recommendation engines, document analysis and conversational agents.
AIaaS (everybody loves crazy acronyms!) is great if your requirements can be met by an existing pre-trained model, and it’s always a good place to start. The use of pre-trained models means there isn’t really much need for data scientists; the cloud provider has already done that work.
Machine Learning as a Service (MLaaS)
These services are aimed at users that want a more bespoke solution. Suppose you want to automatically classify different brands of shoes. The AIaaS pre-trained model might only be trained to identify a ‘shoe’ from other objects in an image, but not different shoe brands. For this task you’d need to train a special model that is aware of the various attributes that distinguish brands of shoes, such as logos.
If you feel like you might need a mathematics degree to figure this out, don’t worry. The big cloud providers offer two types of MLaaS: a hardcore version for the experienced data scientist (maths nerd) and an easier version that doesn’t require you to know your naive bayes from your logistic regression. You can start by simply uploading your data and selecting the attributes from that data you want to train with. What next?
For many applications, artificial-intelligence-as-a-service (AIaaS) offerings are a great place to start. There’s a wide range of pre-trained models that make it easy to get up-and-running with artificial intelligence and you don’t need a data scientist in order to benefit from these solutions.
When you need to go further than the pre-trained models allow, the MLaaS offerings are there for you to train using your own data. In many cases here also, you don’t need a data scientist as the MLaaS service can automatically determine which attributes to use for your custom model.
The cloud providers have put a huge investment into their AI platforms and the learning material that goes with them so go and check them out; it’s really easy to get started.
Here at Fuzzy Labs we help businesses get tangible value from AI. We’ll work with you to explore what’s possible, build a prototype and go from there to a fully production-ready solution that integrates with your existing software.
In future blog posts we’ll go into detail about some of the applications of AIaaS. In the meantime you can follow Fuzzy Labs on Twitter, or LinkedIn, or visit us at fuzzylabs.ai.