With the increasing demand for faster results and real-time insights, businesses are turning to edge artificial intelligence. Edge AI is a type of AI that uses data collected from sensors and devices at the edge of a network to provide actionable insights in near-real-time. While this technology offers many benefits, there are also risks associated with its use.
SEE: Don’t curb your enthusiasm: Trends and challenges in edge computing (TechRepublic)
Use cases of edge AI
There are many potential use cases for artificial intelligence at the edge. Some possible applications include:
- Autonomous vehicles: AI at the edge processes data collected by sensors in real-time to decide when and how to brake or accelerate.
- Smart factories: Edge AI monitors industrial machinery in real-time to detect anomalies or faults. Cameras also detect defects on the production line.
- Healthcare: Wearable devices can detect heart irregularities or monitor patients post-surgery.
- Retail: Store sensors that track customer movement and behavior.
- Video analysis: AI analyzes video footage in real-time to identify potential security threats.
- Facial recognition: Edge AI can be used to identify individuals by their facial features.
- Speech recognition: AI at the edge is now used to recognize and transcribe spoken words in real-time.
- Sensor data processing: Edge AI can process data collected by sensors to decide when and how to brake or accelerate.
Edge AI risks
Edge AI risks include data that may be lost or discarded after processing. One of the advantages of edge AI is that systems can delete data after processing, which saves money. The AI determines that the data is no longer helpful and deletes it.
The problem with this setup is that data may not necessarily be useless. For example, an autonomous vehicle may drive along an empty road in the remote countryside. The AI may deem most of the information collected useless and discard it.
However, data from an empty road in an outlying area can be beneficial depending on whom you ask. In addition, the data collected may contain information that may be useful if it makes it to the cloud data center for storage and further analysis. It could, for example, reveal patterns in animal migration or changes in the environment that would otherwise go undetected.
An increase in social inequalities
Another edge AI risk is that it can exacerbate social inequalities. This is because edge AI requires data to function. The problem is that not everyone has access to the same data.
For example, if you want to use edge AI for facial recognition, you need a database of photos of faces. If the only source of this data is from social media, then the only people who will be accurately recognized are those who are active on social media. This creates a two-tiered system in which edge AI accurately recognizes some people while others are not.
SEE: Artificial Intelligence Ethics Policy (TechRepublic Premium)
In addition, only certain groups have access to devices with sensors or processors that can collect and transmit data for processing by edge AI algorithms. This could lead to a situation where social inequality increases: Those who can’t afford the devices or live in rural areas where local networks don’t exist will be left out of the edge AI revolution. A vicious cycle could result, as edge networks are not simple to build and can be expensive, meaning that the digital divide may increase and disadvantaged communities, regions and countries may fall further behind in their ability to take advantage of the benefits of edge AI.
Poor quality of data
If sensor data is poor quality, then the results generated by an edge AI algorithm may also be poor quality. This could lead to false positives or negatives, which could have disastrous consequences. For example, if a security camera using edge AI to identify potential threats produces a false positive, this could result in innocent people being detained or questioned.
On the other hand, if data is of poor quality due to sensors that are not well-maintained, this could lead to missed opportunities. For example, if an autonomous vehicle is equipped with edge AI that is used to process sensor data to make decisions about when and how to brake or accelerate, poor quality data could lead the vehicle to make poor decisions that could result in an accident.
Poor accuracy due to limited computational power
In typical edge computing setups, edge devices are not as powerful as the data center servers that they are connected to. This limited computational power can lead to edge AI algorithms that are less efficient, as they have to run on smaller devices with less memory and processing power.
Edge AI applications are subject to various security threats, such as data privacy disclosure, adversarial attacks and confidentiality attacks.
One of the most significant risks of edge AI is data privacy disclosure. Edge clouds store and process a large amount of data, including sensitive personal data, which makes them attractive targets for attackers.
Another risk inherent in edge AI is adversarial attacks. In this attack, an attacker disrupts the input to an AI system to cause the system to make an incorrect decision or produce a false result. This can have serious consequences, such as causing a self-driving car to crash.
Finally, edge AI systems are also vulnerable to confidentiality or inference attacks. In this attack, an attacker attempts to uncover the details of the algorithm and reverse engineer it. Once the correct inference is made about the training data or the algorithm, the attacker can make predictions about future inputs. Edge AI systems are also vulnerable to various other risks, such as viruses and malware, insider threats, and denial of service attacks.
Balancing the risk and reward
Edge AI comes with benefits and risks; however, you can mitigate these risks through careful planning and implementation. When deciding whether or not to use edge AI in your business, you must weigh the potential benefits against the threats to determine what is suitable for your specific needs and objectives.