Over the past several decades, scientists, engineers, and mathematicians have developed a variety of AI implementations and models with one goal in mind: to mirror the thinking patterns of the human brain.
During this time, the concept of AI has made some incredible advances in areas from computer vision to natural language processing to advances in medical research. The industry has also endured “AI Winters” where progress slowed or even stopped. However, following AI Winters the industry traditionally experiences a new and refreshed way of approaching the future of AI and the next era of AI is born.
Today’s AI has reached its limits
When advancements in technologies slow, that doesn’t mean that the technology itself has peaked and there is no room left to innovate. In fact, it is the opposite. The AI industry is on the cusp of the next AI Winter. In order to come through this cold season we must rethink how hardware and software will work together to progress and move AI into the future.
The current era of AI advancements has been exciting. However, we are seeing these advancements begin to stall. The industry has been exclusively focused on accelerating the processing power associated with multi-layer neural networks. These AI processors are running headlong into the problem of a slowdown in Moore’s Law. With this, the industry leaders and startups’ ability to squeeze more performance from the existing AI model is under severe pressure.
The Future of AI is Natural Intelligence
To keep up with the demand and expectation of what AI is capable of doing, the future of AI models will need to more closely align with the biological model for human intelligence. The hardware and software must evolve in order to exhibit more natural intellect and processing capabilities.
But this hasn’t been done before, and many of today’s AI leaders are trying to figure out just how to get it done.
Well, we’ve been doing the work, the research, and the problem-solving and here’s what believe the future of AI will look like after this AI winter:
Patterns, not mathematics will be the basis for a new era of AI processing.
Today’s AI models – which are mathematical-based learning models – have very little in common with how the brain analyzes information and data. Mathematics is simply not the basis for the intelligence displayed by humans. The foundation for human intelligence is built on the capability of the human brain to understand the patterns received by our senses. Future AI systems based on pattern processing will be the most efficient means to deliver the human-like capabilities we seek from AI systems.
Data scientists will be able to work with the data they have.
Data scientists today spend 60% of their time preparing their data set for processing by an AI system. In some cases, the data is noisy and must be cleaned up. In others there may be missing values that must be imputed at the risk of injecting bias into the system.
The AI systems of tomorrow will be able to learn with small amounts of information – as little as 5% of the required data today. And just like humans, they will develop the capability to learn in the real world, without requiring tedious preprocessing or synthesizing of information.
AI systems will become trusted entities.
Existing AI models rely on convoluted, deep learning networks that regularly lose the ability to explain their decisions as millions of calculations are performed on the input data. They provide little insight into the ‘why’. This lack of ‘explainability’ has led to governmental regulations that prohibit the use of such opaque systems in certain use cases.
The only way to ensure that an AI system is making reliable, repeatable decisions that are free from unintended bias is to make the very basis for these decisions transparent and interpretable. The next era of AI systems lay in models that preserve the semantics of the dataset from input to output and will have the ability to explain themselves.
Training and inference will merge and become continuous.
Just as humans are continually learning, AI systems will develop the same capability. With this, the concept of training on big data sets that are curated by technology titans will yield to efficient and continual training that can be implemented by everyone. The future of AI will be democratized, allowing broader adoption than ever before.
AI systems will become energy efficient.
More mathematics is not the answer to improved AI. The chips being designed to handle the giant mathematical workloads do not have room left to grow. Training and retraining deep, multi-layer networks can cost thousands, if not millions of dollars, in electrical power and the problem is only getting worse. Next generation AI systems will dispose of their dependence on mathematical processing and adopt more efficient pattern processing techniques.
Pattern-based processing is the technology breakthrough that will lead the industry out of the next AI winter. This next era will clear a path for the future of AI, and provide the headroom needed for decades of improvements in both efficiency and capabilities to come.