Ravi Annavajjhala – CEO, kinara inc.
getty
Many applications require artificial intelligence (AI) these days. However, developers often underestimate the many crucial up-front decisions they must make to ensure their implementation works well within the budget they have allocated.
The first decision must be whether the AI will process the data it consumes in the cloud or at the edge of the network. There are speed and cost advantages to deploying AI at the edge, but the cloud has captured the imagination of most because of its massive computing power and storage capacity. Whether you use the cloud or the edge (or even a combination of both), there are things to consider.
The volume of data presents latency
Uploading large amounts of data to the cloud can often result in very high latency, in many cases exceeding hundreds of milliseconds, which can severely impact operations to the point of ineffectiveness at best and total loss of time and money at worst. The ability to handle large amounts of data almost effortlessly and consistently is a key, but too often overlooked, factor in the effective deployment of AI.
move data around
In terms of bandwidth to move data, you typically not only have to pay a lot, but you also have to pay to have it available at full performance 100% of the time. You don’t want to pay to process bad data. It costs the same, but bad data can be very expensive to eradicate and discard without ruining good data.
costs control
Video cameras generate vast amounts of data, and transferring and analyzing that data requires bandwidth and computing power with monetary values that can add up very quickly. Also, in addition to the time and money to “rent” equipment in the cloud to train an AI model, the cloud means big business for cloud service providers because that equipment must be continuously available for inference.
Privacy
Despite its advantages, cloud storage of sensitive and/or proprietary data is still a concern for many. The thought of transferring valuable data to and from the cloud, potentially subjecting it to network disruption or even hijacking and corruption by dark forces on the open internet, can still be troubling. Accidents are rare, but they still happen, often due to simple mistakes, and those seemingly minor mishaps can escalate quickly.
Is The Edge the best option?
Despite the shortcomings of cloud-based AI, most applications still rely on the cloud to perform end-to-end analytics, such as those associated with business or monetary decision-making. Where does AI come in at the edge? In my opinion, for use cases that require speed, scalability, and privacy, placing AI processing at the edge of the network can avoid many of the issues described. I think putting the AI at the edge is the best option, if not the only one.
For AI applications that don’t require millisecond speed, the cloud should work just fine. However, for an AI decision or reaction that needs to take place in the range of a few milliseconds or less, sometimes with very sensitive or valuable information, edge AI is the choice. While it may require slightly higher upfront costs to research and implement, the ROI should be demonstrably more favorable in a reasonably short period of time compared to cloud.
Example applications that benefit from edge processing
Cashierless stores, medical devices, and warehouse robots are just a few examples where the effectiveness of edge AI can be seen. Cashierless stores can use edge AI to provide greater flexibility when it comes to scaling to accommodate hundreds, if not thousands, of smart cameras. Medical devices with built-in AI capabilities can be used to monitor patients, for example by analyzing abnormal behavior or detecting falls. Warehouse robots need advanced artificial intelligence to enable highly precise instant action.
Consider this in advance
Although most companies of any reasonable size now use some type of cloud environment (whether private, public, or some other combination of cloud services as part of their business strategy), one size does not fit all.
The decision to go cloud or edge usually depends on the application. Many organizations will use a hybrid approach, performing the functions with the most time constraints at the edge while performing deeper analysis in the cloud. For example, warehouse robots can be a great environment to deploy edge AI, but the cloud is still used to consolidate and analyze the inputs from all the robots in the warehouse.
However, a hybrid edge/cloud approach can be daunting to the uninitiated because there is an inherent level of complexity. It is advisable to consider undertaking such a project with an experienced partner or supplier who can ensure that the right decisions are made in advance to avoid potential roadblocks.
However, getting both elements right in the planning stages can ultimately optimize performance and increase efficiency by balancing demand across multiple cloud systems and edge computing environments to precisely meet the specialized need of the business. .
What about storage? How much will you need, both now and in the foreseeable future? If you are dealing with large volumes, cloud computing and storage may be your answer. If data capture is relatively small, intermittent, or only temporary, edge computing devices are likely to not only cost less to deploy, run, and maintain, but may also require less power to operate.
seeing the future
In addition to the three examples described above, there are also many other applications that rely on machine vision and AI to identify someone or something. Although the cloud has been the default choice for running AI algorithms, especially for advanced analytics, new edge AI devices can enable applications to perform time-critical AI functions at the edge, while the role of cloud will shift towards advanced analytics.
The Forbes Technology Council is an invite-only community for world-class CIOs, CTOs, and technology executives. Do I qualify?