As machine learning (ML) capabilities advance, and with the advent of widely available low-cost cloud computing, AI will inevitably be applied to a wider range of more challenging problems, including those that affect the outcomes for millions of individuals throughout society. In high impact, complex settings, it simply isn’t realistic to train a model up front with a single batch of training data and expect it to perform well in all possible scenarios - such a naïve approach will almost certainly fail to capture some of the underlying nuance and edge cases of the situation, leaving gaps in performance and risk of failure during use. Active learning provides a promising way around the issue, empowering the AI to learn from human teachers in uncertain or novel settings and on new data. This architecture allows human experts to impart knowledge gradually as and when they become aware of AI shortcomings, improving performance through teaching and demonstration.
Training models with less data
In many cases, it’s possible to achieve good performance in a model without using large datasets. A large number of data points in most datasets encode information in common with other parts of the dataset and result in increased training cost, for comparatively little return in performance gain.
Active learning allows AI models to be deployed and start delivering value with less data, time and training investment, improving their performance on the fly through the oversight and involvement of human experts in a hybrid, human · AI collaborative system.
See part one of our blog series for an example of how active learning can reduce training time by 75%.
The benefits of active learning
Mind Foundry technology leverages active learning capabilities through our continuous metalearning and human • AI collaboration architectures, combining the speed and scalability of digital decision making with the contextual and situational awareness of human decision-makers. This provides several benefits:
Reduced risk. By including humans as collaborators in the system, close to the point of decisioning, they can quickly rectify any errors made by the AI, preventing issues from propagating through and impacting the wider system.
Improved performance over time through continuous metalearning. The AI system learns from human corrections and can ask better questions of human counterparts, increasing performance gains.
Bias mitigation through Human-AI collaboration. No single human is responsible for training the AI system. Instead, a diverse group of individuals with varying experiences and beliefs each have an impact, helping to reduce the impact of individual biases and blind spots.
Increased flexibility. Active learning lets you improve models over time, so you don’t need to throw all your data at the model at once. This reduces training time, emissions and lets you iterate on the solution as requirements change rather than requiring complete retraining.
"If you have a high stakes problem that affects human outcomes at scale... that last 1% can be someone's life."
Active learning in the field
Very few AI applications are static and capable of being left to their own devices without careful monitoring and relevant updates as problems change with time. Many AI applications resort to frequent benchmarking and performance assessments to detect and resolve data drifts and model degradation, but active learning provides a smoother alternative. By including the facility for algorithms to query human experts on areas of uncertainty, or for validation, it is possible to continually monitor and improve performance, removing the need for resource-intensive batch assessments and updates. This results in models that remain more up to date, a workforce that is better engaged with the problem, and more predictable resource requirements for the project as a whole. Mind Foundry Solutions frequently include active learning capabilities due to the increased reliability, predictability and safety they bring.
For some problems, it may not only be acceptable to train models to a percentage maximum performance, but it might also be the most ethical to do. For example, if you're running ML models to improve your ability to run email marketing, you may agree that it just doesn't make sense to use 75% of your resources to get 1% improvement. Not only would that cost your business a lot of money, if everybody did that (and we know that many people are doing that), it has profound consequences on our environment. But if you have a high stakes problem, an important problem that affects human outcomes at scale, you need that last 1%. That last 1% can be someone's life. With active learning, it doesn't need to be a binary decision between getting that last crucial bit of accuracy or utilising your resources more efficiently - you can do both.
Our work in the Public Sector is a great example of this, where the slightest change in results could have a significant impact on the lives of hundreds of thousands of people.