AI would be able to generate an additional 13 trillion dollars for the global economy between 2020 and 2030. However, its environmental cost remains worrying: AI already has a significant carbon footprint – more important than that of the airline industry. Today, the design of AI models is unquestionably very expensive and to address its energy-intensive nature, engineers are looking into the development of “minimalist AI”. A concept that reduces costs and energy requirements by drastically reducing models. This technology allows AI to be trained on edge networks rather than cloud servers.
Small AI models for small devices
AI models don’t have to be hundreds of gigabytes to be effective. Smaller models, such as MobileNet (20 MB), work just as well, if the input data, peripheral hardware and model architecture are selected appropriately. Additionally, model compression techniques can reduce the number of parameters that go into an AI model, without causing a significant loss of precision in the results.
25 to 30 billion. This is the number of IoT devices we anticipate by 2025. Processing power requirements will then explode due to the amount of data they generate. It has become imperative to shift some of the computational load to peripheral devices. Such small AI models can be pushed to edge IoT devices, which require minimal processing power and capacity.
Minimalist AI Enablers
The AI ecosystem is changing rapidly with advances in federated learning, decentralized web, and batteryless IoT devices. These developments are likely to serve as catalysts for the adoption of minimalist AI.
Federated learning, a concept pioneered by Google, involves training an algorithm on an application’s users’ machine and sharing the learning done on each user’s machine. Thus, the devices train their own models using local data and only share a periodic summary update in order to train the centralized model. This reduces the overall processing requirements – and therefore the energy consumed to train the AI models.
Also, the future of the web seems to be decentralized. Blockchain-based networks, such as Helium, which can make anyone a network coverage provider (in unlicensed spectrum), are turning the Internet’s cards down.
Finally, the appearance of battery-less IoT devices will fully contribute to environmental protection: these devices draw their processing power from their ambient environment rather than from batteries which generate waste.
AI is everywhere. Companies must ensure that it is deployed responsibly and sustainably. Responsible AI means that all initiatives must respect fundamental rights (e.g., privacy and equality) associated with humans. Sustainable AI aims for carbon neutrality. Minimalist AI will therefore play a major role in achieving this goal. As regulators around the world enforce stricter ethical and environmental standards, companies that make sustainability a founding principle will be more successful in AI than ever before.
We would like to thank the author of this short article for this remarkable content
Minimalist artificial intelligence: towards a sustainable AI future? – AI News
Our social media profiles here as well as other pages on related topics here.https://yaroos.com/related-pages/