News/Trends, Tech/Engineering

Emerging trends in artificial intelligence and machine learning – Part 1

January 30, 2020 Preethi Srinivasan, Technical Product Architect, Druva Labs

“Just like software, and the Internet from previous decades, public cloud and now AI are the megatrends of our generation.” — Jaspreet Singh, CEO of Druva

Artificial intelligence and machine learning (AI/ML) is driving breakthrough developments across industries such as Healthcare, Energy, Logistics, and more. Heliogen is using AI to optimize the next generation of solar technology to power energy intensive processes such as manufacturing steel which in the past was only possible with fossil fuels. Another example is Boston Dynamics’ HANDLE – an agile mobile robot that uses deep learning to autonomously unload trucks and move boxes in warehouses.

If someone tells you that AI/ML is hype, remind them that cloud computing was once called hype. Now organizations of various sizes across industries depend on the cloud to run their businesses. While AI/ML is fundamentally revolutionizing how technology is used across a variety of use cases, it continues to be received with skepticism by some, which is limiting its adoption. Still, AI/ML is already the norm for solving problems that require machine intelligence gained by learning from experience.

To understand the advancements in AI, lets focus on 2 areas in our 2 part series on AI / ML:

  • Breadth – Emerging trends in AI that can be leveraged across industries
  • Depth – Deep dive into new, exciting AI tech emerging in real world applications

Let’s take a look at some of the new focus areas emerging in AI/ML, and explore how they can be leveraged across a broad segment of industries.

AI democratization and automated machine learning

New tools and services are making AI/ML accessible to organizations that previously lacked the expertise required to solve their business challenges using AI. At the start, it was complex to build basic ML models due to the lack of efficient and simple-to-use tools, making AI accessible only to large tech-savvy companies that could hire the required talent. This gap in expertise and tools led to the rise of democratization of AI, helping organizations cross the chasm so that any developer can build AI powered solutions.

Automated machine learning (AutoML) is an example of AI democratization. AutoML simplifies the process of developing custom machine learning models. One example of the automation that AutoML includes is hyperparameter tuning. Hyperparameter tuning is a repetitive and time consuming process of selecting the optimal parameter values for the learning algorithm such that the model generalizes well for different patterns. When a model generalizes well for different patterns, it performs well in the real world. AutoML saves developers time by automatically selecting models or providing pre-trained models so that the developer does not need to build one from scratch.

Custom machine learning models can be easily built using AutoML libraries (for example, AutoKeras, an open source library for deep learning models) or leveraging cloud provider services. Amazon Web Services and Google Cloud offer a variety of services which require no ML/AI experience, so anyone can build AI powered solutions for their business needs. Cloud services such as Google Cloud Vision AutoML can detect objects in images. Amazon Comprehend can help you build custom text classification models. If you need more control and visibility into auto-generated models you can use Amazon SageMaker AutoPilot. These cloud services will expand in capability, prediction accuracy, and will be customizable and readily trained for specific industries. As your organization moves to the cloud, it will open opportunities in AI/ML.

Augmented intelligence

People often view artificial intelligence with apprehension because it threatens to replace human intelligence and automate work done by humans. But AI will be most useful when it complements human intelligence and improves human capabilities. Augmented intelligence is a harmony of computer intelligence with human creativity and empathy to propel better decisions.

Here are a handful of examples where AI complements human intelligence:

  • Analytics at scale such as detecting fraudulent or harmful activities from billions of events
  • Automating repetitive tasks or physically arduous tasks such as moving heavy boxes in a warehouse.
  • Improving human safety for dangerous tasks such as bomb detection
  • Improve the precision of decision making, such as healthcare and diagnosis
  • Improve human mobility through the use of brain-controlled prosthetics
  • To solve complex modeling problems such as understanding the origin of the universe

AI will enable people to do more.

Responsible AI/ML:

Building responsible AI/ML is a critical challenge that needs to be prioritized when developing AI/ML systems. AI systems make decisions based on historical data. Unfortunately, if the historical data is biased, the bias is propagated to the AI models. An interesting read on this topic comes from Joy Buolamwini’s research paper on her Gender shades project, which evaluated AI-powered gender classification products built by 3 major technology companies. It revealed that darker-skinned females were the most misclassified group with error rates of up to 34.7% while the maximum error rate for lighter-skinned males was just 0.8%, which indicated a bias in the data used to build the machine learning models

Without responsible AI, investments will be tangled in PR issues and lawsuits. Worse, people’s lives can be affected by irresponsible AI. The Gender shades project highlights that facial recognition technology is increasingly being used by law enforcement despite not being sufficiently tested for demographic accuracy. AI fueled automation now helps determine who is fired, hired, promoted, granted a loan or insurance, etc.

Diversity is a critical component to fight bias and a very important focus area. Not only does fairness qualify for responsible AI, it also includes privacy, security, and ethical nature of these systems. You’ll read more about data security and Druva’s recent research and development in part 2 of this blog. Responsible AI decisions will have to be made across several stages from use case definition, training data collection, sampling, feature selection, to model performance optimization.

Conclusion:

AI adoption will continue to grow rapidly. New techniques, tools, cloud services, and frameworks will rapidly grow and evolve. AI will be revolutionized by democratization. But the future of AI will be defined by responsible AI that augments human intelligence.

Next, let’s get deep and explore exciting AI tech that is emerging in real world applications in part 2 of this blog. Until then, read more about technology futures in our Top 10 predictions for data protection in 2020.