Teach and Test
Announced by Accenture today, the testing tools are meant to help developers create “safe, reliable and transparent” AI that effectively integrates into the scenarios it’s designed for. The tools are based around the two-stage Teach and Test algorithm and address ethics, bias, security and compliance.
The first stage of the process, “Teach,” is used to focus the machine learning models and algorithms which train the AI. The Teaching process involves experimentation and statistical evaluation of different models to determine which is the most performant. The decision is made by assessing the model’s accuracy, as well as additional factors such as its ethical basis and ability to withstand bias.
Once a model has been selected, the rest of the process continues with the trained AI. The “Test” phase is used to evaluate the completed AI’s performance against established baselines and performance indicators. The AI is expected to be capable of indicating how it arrived at a particular output, which may include explanation of the decisions and assumptions made along the way.
Accenture’s suite of cloud tools monitor the AI as it is tested. Information is stored in the cloud to evaluate changes in the AI’s performance as it develops. Accenture’s also added unique algorithmic tools to accelerate the testing of common AI workloads, such as computer vision and object recognition.
Training sustainable systems
Teach and Test is intended to help solve development issues in the current AI ecosystem. Skilled developers are hard to source and businesses can find it difficult to effectively test and evaluate their AI. However, assessment of an AI’s performance is a crucial prerequisite to deployment, helping eliminate sources of inefficiency and bias. Accenture’s cloud facilities can now assist firms in testing and monitoring their AI, while reducing the time taken to repetitively train new models.
“The adoption of AI is accelerating as businesses see its transformation value to power new innovations and growth,” said Bhaskar Ghosh, group chief executive, Accenture Technology Services. “As organizations embrace AI, it is critical to find better ways to train and sustain these systems – securely and with quality – to avoid adverse effects on business performance, brand reputation, compliance and humans.”
READ NEXT: Lenovo patent uses blockchain to authenticate physical documents
Accenture’s already deployed the technology in a few pilot projects. The Teach and Test approach was used to successfully train a new conversational assistant for a financial services company’s website.
Using Accenture’s tools, the bot was trained 80 percent more quickly than would otherwise have been possible. Operating results have indicated an 85 percent improvement in accuracy when making recommendations to customers.
The training service is available to businesses from today as part of Accenture’s wider portfolio of testing tools. The suite is currently used by over 1,000 clients worldwide.
