Examples of artificial intelligence (AI) are all around us. We probably use AI more than we think and in many ways take it for granted. Our smartphone assistant is an excellent example of AI, even though we may not think of it as such. In many cases, we take for granted our interaction with Siri or Google Assistant because they consistently work. Likewise, face recognition has become a standard unlock feature on new smartphones.
AI, for which machine learning is a subset, works by training a computer-based neural network model to recognize a given pattern or sound. Once the neural network has completed training, it can then infer a result. For example, if we train a neural network with hundreds of images of dogs and cats, it should then be able to correctly identify a picture as either being a dog or a cat. The network model determines an answer and indicates the class probability of its prediction.
As machine learning-based applications become more deeply ingrained in our daily lives, system developers have become more aware that the current way neural networks operate is not necessarily the right approach. Using the above example, if we showed the neural network a picture of a horse, the neural network, trained only to infer either a cat or a dog, would have to decide which one it is going to pick within the class for which it was trained. Of more concern, the likelihood is that it would give an incorrect prediction with a high degree of probability; something you may not even notice. The model has failed silently.
As humans, our approach to a similar scenario would be very different. We use a more reasoned decision-making approach. We would expect the neural network to answer that it didn’t know or had not seen an image of a horse before. The example above, while very simple, serves to illustrate a flaw in how a neural network has to operate in the human world of surprises and uncertainties. The reality is that many industrial and automotive systems continue development even though there are concerns about how it might operate in certain situations.
At NXP, we’ve been investing in building our AI capabilities for many years and are concerned with such shortcomings. Your smartphone assistant incorrectly inferring a spoken word is far removed from the consequences that might occur in an industrial or healthcare environment. We are delivering advanced machine learning solutions for our customers and have been working on an approach termed ‘explainable AI’ (xAI). xAI expands on the inference and probability capabilities of machine learning by adding a more reasoned human-like decision-making approach and the additional dimension of certainty. xAI combines all the benefits of AI with an inference mechanism that is closer to how a human would respond in a situation.
Consider the following example. Imagine you were a passenger in an autonomous vehicle. If the vehicle was proceeding slowly and cautiously, you would naturally wonder why the vehicle was being so careful. If the driver was human you could ask why are you going so slow, to which the driver would explain that with the heavy rain the visibility was poor and that they were uncertain what hazards lay ahead. The explanation is based on uncertainty. xAI decision making behaves in a similar way by communicating the aspects of inference that the model is uncertain about.
At NXP we are already investigating ways we might incorporate xAI capabilities in the machine learning solutions we are developing for automotive, industrial and healthcare systems.
With the unprecedented global COVID-19 pandemic situation, our xAI research teams believe that NXP xAI might help enable the rapid detection of the disease in patients. It is still early days, but we are encouraged by the proof points we have seen and have established interactions with some leading hospitals to see how our xAI technology might aid the healthcare challenges our planet currently faces.
The use of CT radiology and X-ray imaging provides a fast alternative detection capability alongside the prescribed PCR testing and diagnosis protocols. CT and X-ray images could be processed by a suitably trained xAI model to differentiate between clean and infected cases. xAI allows for real-time inference confidence and explainable insights to aid clinical staff in determining the next stage of treatment.
Our xAI research team believes they are well-advanced with a mature model and are engaging in discussions with medical and AI experts in Europe and across the Americas. However, to further our research, we must have access to larger anonymized datasets and would welcome hearing from researchers and potential partners engaged with COVID-19 who would like to collaborate with us to advance this detection technique. If you would like to contact us about collaborating on the use of xAI for COVID-19 detection, please contact our research team.
xAI gets us closer to how humans react in situations where decision-making involves uncertainty. It adds certainty and confidence to class probability-based decisions. NXP sees opportunities for xAI across safety-critical systems for automotive, industrial and healthcare applications.