Click to learn more about author Andy Gray.
There has never been a better time to be a data scientist. As businesses seek to expand their use of Data Analytics, demand for Data Science experts and remuneration levels continue to grow. With this increased focus on Data Science and AI has come an influx of projects and heightened pressure on Data Science teams to deliver more solutions and deliver business value quicker. However, just because the number of projects is increasing, does this mean the amount of manual effort that goes into delivering more AI solutions needs to increase with it? Will larger teams necessarily deliver value quicker?
A recent Gartner survey reported that it takes on average four years to get an AI project live. For 58% of businesses it takes two years to get to the piloting stage. Furthermore, these big investments in data and AI projects are successful only 15% of the time. As a result, for many readers, delivering an effective AI app in one day sounds like an impossible pipedream. However, small AI projects can be produced in a short period of time utilizing the latest Auto Machine Learning (AutoML) and deployment automation technology. AutoML/AI as a service (AIAAS) platforms are now stepping up to deliver AI at speed, enabling data scientists to create competitive and researcher-level models in minutes.
Creating the Model
One feature of AutoML that allows this improved production speed is its ability to explore possible models promptly, enabling data scientists to drive value quickly. AutoML allows a data scientist to set the broad outlines and parameters for the task required, and the AI finds the best possible solution to the problem. It’s like having millions of apprentices working to solve a problem with an almost infinite number of solutions. AutoML does this at machine-level execution speeds, leaving any humans trying to solve the same problem in its wake.
The key to an effective AutoML platform is to get the right balance between defining the broad brushstrokes, in such a way that they are not cumbersome to define and change; and to give the data scientist enough control to create exactly the solution they want. By letting AI do this work at speed, data scientists can stay focused on the high level, evolving their solution rapidly and monitoring how model performance is affected. While AutoML is a great way to let data professionals, who may not be trained in ML or AI, produce machine learning models and solutions, more advanced data scientists also gain huge advantages in speed, insight, and productivity by using AutoML.
Typically, data scientists manage their experiments haphazardly in notebooks or raw code; the “hyper-organized” among us might even use source control to help track our progress. But, as in any exploratory process – where changes are tested and abandoned regularly, while iterating to the best solution – this process is not well catered for by traditional development tools that expect an inexorable march in one direction.
AutoML compounds this problem by creating thousands more models and data pipelines than a data scientist would be able to build manually. As such, automated tracking, ranking, and comparison of machine learning solutions becomes a critical requirement for any platform that uses AutoML.
Deploying the Model
As more AI projects are leaving the prototype and pilot stage, our industry is just waking up to the fact that creating AI models, which meet enterprise-level service level agreements (SLAs), is a different challenge than simply creating a model. Today, people with crossover code and Data Science skills are generally referred to as Data Science engineers. While we still read about the shortage of data scientists, data scientists currently outnumber Data Science engineers 100 to 1, according to my own search via LinkedIn’s Sales Navigator.
By using AI models in production to serve predictions, models can be published in a few clicks and are ready to meet any demand. The REST API makes it very easy for developers to integrate these AI/ML models into apps and services. As such, deploying the models takes only a few seconds work. When an API is used from the start, app development can kick off in parallel to the Data Science, shaving months off the prototype and pilot stage.
Delivering to Value
It is critical to remember that people can’t consume AI directly; they need a service or application to provide targeted processing and interfaces for their use case, and this means app code. The de-facto standard for writing robust systems that live in the cloud is microservice architectures. Typically, in the cloud we expect the hardware that we use to fail more than traditional computers, but by using microservices, AIAAS providers can provide near perfect up-time for a range of apps and services.
Microservice architectures are another new and relatively difficult skill that’s incredibly hard to hire for. Even for companies that have built several successful microservice-based systems, it often takes years each time. However, this code is not that different from project to project. The business and interface codes tend to be orders of magnitude easier to write and test. A sensible approach is to automate the parts that are common to most solutions and hard to do. By providing a robust microservice architecture and an API that’s easy to drop custom app code into, it’s easy for front-end and back-end developers to create new apps and services.
This type of technology is very new. It should also be noted that while all vendors use terms, such as Deep Neural Networks and AutoML, there are all sorts of variation in quality, in terms of model performance, stability, and scalability. If you asked me to recommend a way to tell vendors apart, I’d say look for quality of results with brands you trust.
The major tech firms have been late to the party on AIAAS. Start-ups, such as ourselves, have had full model explainability built in since 2017. Of the major players, only Microsoft has any support for this feature and theirs is a “build-it-yourself” SDK in preview, as of a month ago, not a built-in part of a highly automated AIAAS platform.
The pace of change in technology in AI and ML is phenomenal and nimbler organizations are doing a better job of staying on top of the latest advances and giving that advantage to their users. At this point the question of whether AI will remain hype or whether it is going to deliver on the bottom line has been well and truly answered. From automating the application of tax law 90% at Deloitte, to reducing wastage and costs for the NHS, AI is currently live across many industries delivering on the promises of massive ROI.
As Elon Musk said: “You have companies that are racing to build AI or they’re going to be made uncompetitive. If your competitor is racing towards AI and you don’t, they will crush you.” AutoML and AI as a service are technologies that can massively accelerate companies in this race. We’re seeing projects completed on average 24 times faster than industry averages. If you are trying to deliver AI, especially at enterprise scale, and not using these technologies, you are trying to dig a tunnel with your bare hands, while your competitors are using a tunneling machine.