There are many perspectives on why artificial intelligence (AI) and machine learning (ML) continue to evolve. A recent McKinsey report identified the industrialization of ML and the application of AI as one of the major trends this year. At a conference this week at the AWS re:Invent conference, BraTIn, Vice President and General Manager of AI and Machine Learning at Amazon (AWS), outlined six key trends that the cloud giant sees driving innovation and adoption in 2022 and beyond.
AWS claims to have over 100,000 customers for its AI/ML services. These services are distributed across three tiers: ML infrastructure services, which enable organizations to build their own models; SageMaker, which provides tools for building applications; and dedicated services for specific use cases, such as transcription.
"Machine learning has gone from a niche activity to an integral part of how companies conduct their business," Saha said during the conference.
Trend 1: Model complexity continues to increase
Saha says that the complexity of ML models has grown exponentially in recent years. His use of the word "exponential" is not an exaggeration.
One way to measure the complexity of a machine learning model is to count the number of parameters it contains. Saha explains that parameters can be thought of as value variables embedded in the ML model. Saha says that in 2019, the then-state-of-the-art ML models had approximately 300 million parameters. Fast forward to 2022, and the best models now have over 500 billion parameters.
"In other words, the complexity of machine learning models has increased 1,600 times in just three years," Saha said.
These massive models are now often referred to as base models. Using the base model approach, an ML model can be trained once on a massive dataset and then reused and tuned for a variety of different tasks. Therefore, businesses can benefit from increasingly complex processes through a more easily adopted method.
"[Base Model] reduces the cost and workload of machine learning by an order of magnitude," Saha said.
Trend Two: Data Growth
More and more data, and different types of data, are being used to train ML models. This is the second key trend identified by Saha.
Organizations are now building models trained on both structured data sources (such as text) and unstructured data types (including audio and video). The ability to feed different data types into ML models has led AWS to develop various services to assist in model training.
Saha highlighted one such tool, SageMaker Data Wrangler, which helps users process unstructured data using a method that makes it suitable for ML training. This week at re:Invent, AWS also added new support for geospatial data to SageMaker.
Trend 3: Industrialization of Machine Learning
AWS also sees the trend of ML industrialization. This means that ML tools and infrastructure are becoming more standardized, making it easier for organizations to build applications.
Saha stated that industrializing ML is important because it helps organizations automate development and make it more reliable. As organizations build and deploy more models, industry-wide approach is crucial for scaling.
“Even within Amazon, we use SageMaker for industrial and machine learning development,” Saha said. “For example, the most complex Alexa voice models are now being trained on SageMaker.”
Trend 4: ML-enabled applications for specific use cases
ML is also growing due to dedicated applications for specific use cases.
Saha stated that AWS customers have been requesting vendors to automate common ML use cases. For example, AWS (and other vendors) now offer services such as speech transcription, translation, text-to-speech, and anomaly detection. These provide organizations with a simpler way to use ML-enabled services.
For example, sentiment analysis in real-time audio calls is a new and complex use case, which AWS now supports through its real-time call analytics capabilities in its Amazon Transcribe service. Saha stated that this feature uses speech recognition models to understand customer emotions.
Trend 5: Responsible Artificial Intelligence
There is also a growing trend and demand for responsible artificial intelligence.
"As artificial intelligence and machine learning develop, people realize that we must use them responsibly," Saha said.
From AWS's perspective, responsible AI needs to possess several key attributes. The system needs to be fair, operating equally for all users regardless of race, religion, gender, or other user attributes. The ML system also needs to be interpretable so that organizations can understand how the model works. Furthermore, governance mechanisms are needed to ensure that responsible AI is practiced.
Trend Six: Democratization of Machine Learning
The last key trend driving ML forward is democratizing technology, making tools and skills available to more people.
"Clients tell us that they... often have difficulty recruiting all the data science talent they need," Saha said.
According to Saha, the answer to the democratization challenge lies in continuing to develop low-code and use-case-driven tools, as well as education.
"AWS is also investing in training the next generation of machine learning developers," Saha said. "AWS is committed to helping more than 29 million people improve their technical skills through free cloud computing skills training by 2025."
