This article examines the possibilities and advantages of Artificial Intelligence as a Service (AIaaS). You are probably already familiar with the acronym aaS, which stands for “as a service”. Paste a random letter in front of it and you have a revolutionary cloud service model. While most readers probably are familiar with the concept of these cloud services, a new service has entered the cloud and has gained a lot of attention over the last years. AIaaS promises accessible AI tools for every company through the cloud.
Cognition through the cloud
AIaaS consists of artificial intelligence services delivered by a third party through the cloud, which allow users to incorporate automatic vision, speech and language understanding capabilities in their own services (Microsoft, 2019). AIaaS allows users to upload data, run state-of-the-art models in the cloud, and receive results by only using a cloud platform or API. Fortunately, Google offers free demos to try out their services yourself.
The pictures below show results of the classification of two pictures fed to Google’s free vision API. The nature of the input pictures has nothing in common, however, the API of Google is able to classify both input pictures accurately.
Besides classifying buildings and dogs for fun, the API has been used for counting endangered species (Wildlife Insights, n.d.) and automatically measuring the deterioration of shorelines (Anthony Reisinger, 2019). Furthermore, Uber is using a similar API from Windows to match drivers with Uber driver account files through face recognition to ensure that clients step into the right car (Microsoft, n.d.).
Figure 1 Original picture taken from Kaggle Dogs vs. Cats competition
Figure 2 Original picture taken from Tilburg.com
Google also offers a free demo of their Natural Language API. Below you can find the results of a sentiment analysis from the API. The API has been fed the sentences “This site is really helpful” and “I have never been so annoyed by a website”. Subsequently, the API tries to classify the sentiment of them. While this seems like a dull service compared to computer vision, applications based on this technology could help customer service respond quicker to negative feedback and accelerate customer research.
Additionally, people that have some familiarity with machine learning can train their own models by using a machine learning engine in the cloud (Google, n.d.). Training your own model does not require any programming knowledge. You simply upload your dataset, select the labels and evaluate the models. This enables a lot of people for who programming might be scary to train and deploy their own solutions for their specific challenges.
The promise from the cloud
AIaaS has some major advantages compared to developing and deploying AI services in-house. First of all, the field of AI is rapidly growing and records in natural language processing and computer vision are being broken on a monthly basis. It would require tremendous investments in people with the right backgrounds and continuous training to keep up with state-of-the-art solutions (Deloitte, 2018). This is especially true for markets that rely heavily on AI to be competitive.
Secondly, training models and using them for prediction require substantial computational resources. GPU’s are the preferred hardware for doing these calculations and their prices, partly due to crypto boom, are now starting to stabilize (Niall McCarthy, 2018). Further, the state of the art models are only becoming more complex (Felix Laumann, 2018). Even if one would invest in the required infrastructure, it would be the company’s task to up- and downscale these assets accordingly. As a result, expensive equipment could be standing idly in a company’s basement most of the time. This argument is not only true for AIaaS, but also for infrastructure as a service. However, computational resources are an even bigger constraint for AI applications.
Finally, AIaaS removes the necessity of the gigantic heaps of data for training machine learning models that power most cognitive applications. In general: the more data, the better the model performs. Instead of training the models yourself from scratch, you can lever sophisticated pre-trained models that are trained on millions or billions of examples. It should be no surprise that Google, Amazon, Microsoft and IBM are one of the biggest players in AIaaS-market (Patrizio, 2018). These companies have trained their models on an almost exhaustive collection of examples. As a result, one model can be used for numerous prediction and classifications tasks.
While the promises of AI and AIaaS are immense, we should not start firing data scientists and AI-developers. We can only speculate when discussing the consequences of AI for the workforce. AIaaS will speed up and automate a substantial part of the functions of these vocations, but they are still needed. An increase in the accessibility of smart technologies will lead to an increase of people that can work with these technologies. Further, the off-the-shelf models work for generic use-cases, but do not outperform specialized models. Therefore, models still need to be trained or off-the-shelf models need to be fine-tuned. Finally, most people working in these vocations spend substantial time collecting and cleaning data, something AI applications can’t live without.
However, the role of data scientists and AI developers could shift from a technical centered to a more business centered position. Higher accessibility and faster development and deployment of AI enabled by AIaaS could help companies gain competitive advantage through the use of these services. This could result in a higher need of agility and flexibility in the use of AI and requires data scientists and AI developers to play a centric role between business and AI. On one side they need to be able to translate business needs to smart applications. On the other side, they need to be able to spot valuable developments in AI, translate the value to the business side and leverage these technologies in order to be competitive.
The advent of AI-services makes utilizing state-of-the-art cognitive and machine learning solutions possible without major investments in people, infrastructure and data collections and enables companies to enrich their products and services with general and more fine-tuned solutions. The service would not necessarily replace people currently working on AI applications, but could rather shift their function to more business centered functions.
Anthony Reisinger, A. (2019, January 9). Coastal classifiers: using AutoML Vision to assess and track environmental change | Google Cloud Blog. Retrieved May 2, 2019, from https://cloud.google.com/blog/products/ai-machine-learning/coastal-classifiers-using-automl-vision-to-assess-and-track-environmental-change
Deloitte. (2018, October 22). State of AI in the Enterprise, 2nd Edition. Retrieved May 2, 2019, from https://www2.deloitte.com/insights/us/en/focus/cognitive-technologies/state-of-ai-and-intelligent-automation-in-business-survey.html
Felix Laumann, F. (2018, November 26). When machine learning meets complexity: why Bayesian deep learning is unavoidable. Retrieved May 2, 2019, from https://medium.com/neuralspace/when-machine-learning-meets-complexity-why-bayesian-deep-learning-is-unavoidable-55c97aa2a9cc
Microsoft. (n.d.). How Uber is using driver selfies to enhance security, powered by Mi…. Retrieved May 2, 2019, from https://azure.microsoft.com/nl-nl/resources/videos/how-uber-is-using-driver-selfies-to-enhance-security-powered-by-microsoft-cognitive-services/
Niall McCarthy, N. (2018, October 19). Infographic: GPU Prices Stabilizing After Crypto-Mining Boom. Retrieved May 2, 2019, from https://www.statista.com/chart/15843/nvidia-and-radeon-gpu-pricing/
Patrizio, A. (2018, September 11). The Top Cloud-Based AI Services. Retrieved May 2, 2019, from https://www.datamation.com/artificial-intelligence/the-top-cloud-based-ai-services.html
Wildlife Insights. (n.d.). Get Involved | Wildlife Insights. Retrieved May 2, 2019, from https://wildlifeinsights.org/get-involved