Tiny AI is a set of principles that leverages the latest advances in data, hardware, and software to reduce the overall economic and ecological costs of artificial intelligence (AI). AI has the potential to change the world, or at least add $13 trillion to the global economy, according to the McKinsey Global InstituteIt faces many technological issues preventing its wider adoption.
AI technology comes at a high environmental cost, requiring a significant amount of computing resources and bandwidth when running on the cloud. According to data from the International Energy AgencyData centers are responsible for 2.5 to 3.7 percent of global greenhouse gas emissions. This figure exceeds the total greenhouse gas emissions of the commercial aviation industry, which stands at 2.4 percent, a sector with a well-documented history of environmental criticism.
These emissions are grouped into three categories:
- Group 1: Emissions accounts for refrigerants (for cooling), diesel (emergency power generators) and natural gas (for heating and fuel supply).
- Group 2: Accounts for emissions generated by electricity consumed or purchased in the context of data center operations.
- Group 3: It takes into account indirect greenhouse gas emissions, such as those generated by IT equipment.
MIT Technology Review corroborates this point, noting that the process of training an AI “can emit more than 626,000 pounds of carbon dioxide,” or “nearly five times the lifetime emissions of an average American car.”
Even without the physical infrastructure issues, according to the IEA, storing 100GB of data in the cloud each year would produce 0.2 tonnes of CO2. Given the capacity of Open Vault Broadband Outlook Report By 2024, the average internet user is estimated to consume 641GB of data per month, so the cloud is hardly an environmentally friendly solution for AI emissions.
Beyond its environmental cost, AI is also very costly financially. For large language models to understand the meaning of a query and generate text that reads like a human response, they must study billions of text documents from all over the internet. The cost of training GPT-3, an AI capable of answering general knowledge questions and writing basic news articles, was $12 million.
Additionally, AI training requires significant physical infrastructure, as the algorithms used for deep learning must process an implausible amount of data. The sheer amount of computing resources required can lead to reduced efficiency and privacy limitations of AI applications. Tirias Research predicted that, if we follow its current course, the infrastructure and operating costs of generative AI data centers will exceed $78 billion by 2028.
While you can swap out physical infrastructure for cloud services like the latest offerings from Google Cloud, AWS, and Azure to relieve the demand for local memory and compute power, you’ll need significantly more bandwidth to do so.
That being said, you might be wondering why the entire tech world seems to be moving towards AI in the future, but the truth is that Pandora’s box has been opened and there is no closing it. National Institute of Standards and Technology According to a study by IHS Markit, more than 75 billion IoT devices will be in use by 2025. With the number of devices in circulation, the need for processing power could explode due to the sheer amount of new data, unless we shift some of the computing load to edge devices that require less processing power and resources.
How does Tiny AI solve these problems?
AI models don’t have to scale to hundreds of gigabytes or take up thousands of dollars of physical server infrastructure. MobileNet models, for example, are based on a simplified architecture that uses just 20MB and is capable of delivering low latency for mobile and embedded devices.
Tiny AI models use model conditioning techniques such as knowledge distillation, network pruning, and quantization, all of which reduce the number of parameters that should be fed into an AI model without reducing its accuracy. Running Tiny AI on new hardware, better designed to handle compute-intensive tasks, can also reduce the strain on cloud services and mitigate security concerns related to transferring data to the cloud in the first place.
Miniature AI has many applications across a wide range of industries and, while still in its development phase, is expected to have a significant impact in the near future. In healthcare, miniature AI can help deliver faster results for medical tests. It allows healthcare providers to easily access comprehensive research and use deep learning to quickly analyze test results, make an informed diagnosis, and address potential issues. Miniature AI will also enable self-driving cars to respond more quickly and reliably and improve the image processing capabilities of cameras.
Once fully operational, Tiny AI would allow simple, compact devices like your smartphone to deploy complex training algorithms without connecting to the cloud. Developers have already started making progress in this area, as Google in May discovered a way to run Google Assistant locally on smartphones, and Apple can use Siri’s voice recognition app offline on iPhones with iOS 13 and beyond.
