Getting AI from lab to production

The business is eager to push AI out of the lab and into production environments, where hopefully it will usher in a new era of productivity and profitability. But it’s not as easy as it seems because it turns out that AI tends to behave differently in the test bed than in the real world.

Breaking this hump between lab and actual applications is fast emerging as the next major career goal to deploy AI. Because smart technology requires a continuous flow of reliable data to function properly, a controlled environment does not necessarily mean this proof is for traditional software. In AI, the uncontrolled environment is now the real test, and many models have failed.

The ‘Valley of Death’

Crossing this “Valley of Death” has become so important that some organizations are elevating it to an executive-level core competency. Valerie Bécaert, senior director of research and scientific programs at ServiceNow’s Advanced Technology Group (ATG), is now leading the company’s research on bridging this space. As he explained in Workflow recently, it’s not just a matter of training AI properly, but changing the organizational culture to improve AI skills and foster greater risk acceptance.

One technique the group employs is to train the AI ​​with limited data so that it can learn new facts on its own. Real -world data environments, after all, are much larger than the lab, with data coming from countless sources. Rather than just throwing initial models into this chaotic environment, low-data analysis provides a simplified path to more effective models that can extrapolate more complex conclusions based on their acquired knowledge.

A recent report by McKinsey & Co., highlighted some of the ways leading AI practitioners – defined by the company as those who can attribute 20% of EBIT to AI – are driving production projects seamlessly and reliably. Among the key best practices, the company identifies the following:

• Use design thinking when making tools
• Test performance internally before deployment and monitor production performance to ensure results show continuous improvement
• Establish well -defined data management processes and protocols
• Develop the AI ​​skills of technology personnel

Other evidence seems to suggest that the cloud provides an advantage when deploying AI in production environments. In addition to extensive cloud scalability, it also offers a wide range of tools and capabilities, such as natural language understanding (NLU) and facial recognition.

of AI Accuracy and precision

However, part of the problem of putting AI into production is the AI ​​model itself. Android developer Harshil Patel told Neptune recently that most models make predictions with high accuracy but low accuracy. This is a problem for business models that require exact measurements with minimal tolerance for errors.

To counteract this, organizations need to be careful about eliminating outlier data sets in the training process, as well as implementing continuous monitoring to ensure that bias and variability do not creep into model over time. Another issue is class imbalance, which occurs when the instances of one class are more common than another. This can divert results from real experiences, specifically data sets from new domains being introduced.

In addition to technological inhibitors in production-ready AI, there are also cultural factors that must be considered, says Andrew NG, adjunct professor at Stamford University and founder of deeplearning.ai. AI tends to disrupt the work of many stakeholders in the enterprise. Without their purchase, hundreds of hours of development and training are wasted. This is why AI projects should not only be effective and useful to its users, but it should also be explainable. The first step in any project, then, should be to define the scope, where technical and business teams meet to determine the intersection of “what AI can do” and “what matters most to the business.”

The history of technology is full of examples of solutions to finding problems. AI has the advantage of being so flexible that a failed solution can be quickly configured and re-deployed, but it can be costly and futile if the right lessons are not learned from those failures.

As the business progresses with AI, the challenge is not to push the technology to its conceivable limits, but to ensure that the effort in developing and training AI models is focused on solving today’s real problems while ensuring they can pivot. to problems that will arise in the future.

#lab #production #Source Link #Getting AI from lab to production