The focus of discourse within the AI community has for a while now, been around the concept of “Beneficial AI”. 

Strictly speaking, applications that intend to enhance the human or societal condition are considered as being beneficial. This might be applicable to medical practices, like diagnosing illnesses, research, or technologies that enhance patient outcomes. Conservation of the environment, like sensors that identify hazardous emissions, in addition to humanitarian initiatives, like agricultural or transport enhancements to assist an expanding populace also reap advantages from these technologies. Even simplistic, useful applications that make our daily lives simpler and streamlined are beneficial. 

Usually when someone wishes to do the best, the wish is to do it as swiftly as possible and as efficiently as viable. So while there are innumerable ways through which AI can reap benefits, and several organizations are currently producing these functionalities, there are some obvious AI technologies that have the competitive edge with regards to providing beneficial AI. 

You might recall a game from your youth called “Where’s Waldo.” The basis of the game is to find Waldo in his patented red-white stripes sweater and glasses. This is a HOB (Hidden object game) developed as a massive visual puzzle. 

While it can prove to be difficult for the human eye, recognition of patterns and imagery is a typical AI activity – so commonplace, as a matter of fact, that even previous gen neural networks generate considerably precise results and find Waldo. 

• What is the time taken? 
• What is the level of effort and energy expenditure? 
• What are the number of computations executed in the cause? 
• What is the level of hardware that is needed? 

• What are the expenses? 

And what if Waldo changes the color of his sweater? Maybe the red and blue stripes isn’t cutting it for him anymore. The alters the playing field completely. The system has to carry out the search for Waldo one more time. 

So while there are several AI technologies that can execute this simplistic activity, in our scenario, finding Waldo, there are major differences in their strategy, and effectiveness.  

The BrainChip Akida processor, which leverages event-based Spiking Neural Networks (SNNs) is excellent in various ways, from reduced power consumption to learning in increments, and high-speed inferencing/one-shot training. 

The Akida engine can identify Waldo quicker, with reduced exertion, and a lot lower computational expenditure. 

Say, rather than tracking Waldo, we’re looking out for an endangered animal at risk from poaching, in a region renowned for such activities – an instance of Beneficial AI; we can observe why quickness and effectiveness is of paramount importance. Detecting members of that particular group of animals through image or the audio of their calls, and precisely categorizing ones with atypical features, like an absent ear enables for improved tracking and quicker intervention. 

Going back to Waldo, what if you could keep an ear out for Waldo? This would furnish another method to quickly detect and find his location in each puzzle. BrainChip has executed several tests in vibrational analysis: the capacity to “feel”, document, and process mechanical vibration noise, with the outcome surpassing the capacities of the human ear. 

With Akida’s speed, precision, and effectiveness in AI vibrational analysis, it can detect outliers swiftly. This is critical for detecting wear and tear in machines, for safety and preventive maintenance, or to enhance gas efficiency and minimize energy expenditure. 

There are several ways in which we observe Beneficial AI taking place at the edge. Geographic, industrial, medical, and biometric, to list a few examples. Akida does not need an external processor, RAM, or Deep Learning Accelerator (DLA) and is extremely conservative in terms of energy, so it is particularly apt in these applications. To safeguard species at-risk, we are required to be deploying unmanned aircraft, or drones, with cameras and CPUs in tow, carrying out on-device data analytics where there is no cloud connectivity, and eating up minimal battery power which enables for a longer life. To evaluate for COVID-19 and manage outbreaks, we require hand-held diagnostic evaluation gadgets that can carry out analysis of breath sensor data in the real world. 

Akida AI functionalities are superior, so we can generate beneficial outcomes not just quicker but more precisely – at almost 100% precision in various applications that have been evaluated.  

As it leverages SNNs, Akida knows how to distinguish between “good” data and bad data or trash data. Lately, analysts detected susceptibilities in a common image categorization system that leverages previous-gen neural networks. Interestingly, the system was tricked by imagery containing text, for example, a piece of paper with iPod written on it fooled the system into categorizing the image as an iPod device. 

Likewise, the background of imagery may consist of features that induce confusion in the neural network, leading it to a mis-categorization. If an object is part obscured, such errors often take place. A red-and-white striped curtain may be wrongly categorized as Waldo.  

Obviously, identifying Waldo isn’t a question of life and death, but some AI activities will be. Edge devices outfitted for Beneficial AI must hence provide speed, precision, agility, and efficiency to do the most good.