Computer programmer, researcher and inventor
Neural networks have provided us with a powerful and inexpensive tool to use to solve prediction, computer vision, and text analysis problems. However, at the same time, they raised the problem of inaccuracy, which is presented as the “norm” and “black box” of deep networks, the derivation of which is difficult to understand and improve.
It is not reasonable to set up a neural network to control complex and expensive objects, where the cost of an error is high or can cost lives or irreparable damage. This field also includes medicine, where neural networks have recently been used to analyze images and make a diagnosis. If the first is completely justified – we currently have no other technology comparable to CNN – networks (Attention + -), then the diagnosis with a network is not an optimal and very limited solution, leading us to In error.
This is due to the limited representation of the meanings of words and concepts in the neural network. The concepts we use in our heads are dynamic in nature and are more like streams connected to other flows than static objects with fixed connections, as depicted in modern GPT-2,3 neural networks, Bert, etc.
They do not better reflect the meaning that the shadow of an object reflects its actual geometry and color. However, if we take the strong side of neural networks – quickly obtaining a preliminary result and use it as an input for classical AI, which performs precise operations on semantic data and thus is able to compensate the errors of a neural network, then it is probably possible to obtain a result that is not separately accessible to networks or to classical AI.
Classic AI (hereafter cAI) is a rule-based inference system capable of providing any precision and depth of analysis on the data provided. The rules can be as complex as you want, even beyond human understanding, and at the same time give an accurate and explainable calculation, provided they are correct.
The downside of ICA is the complexity of rule making and system maintenance. All known fairly useful cAI systems were created from scratch and had their own representations of rules and inferences, which made it difficult to effectively reuse the experience of their creation and training in the future.
The overwhelming majority of these projects did not lead to any results due to incorrect timing (sometimes), lack of advanced experts able to clearly state their knowledge, as well as lack or low qualification knowledge engineers, a rare caste of programmers, which in theory should be knowledge of fuzzy logic, linguistics, semiotics and psychology.
They were a mediator between experts and computers, programming the knowledge gleaned from the experts. All these difficulties have led to the oblivion of cAI systems and the shift towards neural networks.
However, imagine that we have a cAI system that already has some idea of basic knowledge in the domain, as well as written logic rules describing exactly how logic works in that domain. If to broaden the knowledge, we would have nothing to program (- knowledge engineers), but it suffices to add the specific knowledge of the expert to the existing ones.
Suppose also that we have found a way to represent knowledge in such a way that it becomes understandable to any expert in this field. Then we can give the system to many specialists and then combine their knowledge into super knowledge, which should be richer and more precise than its parts. The system will be able to compare parts of super knowledge and give information on where and how they contradict each other, as well as resolve and correct the contradictions in automatic or expert mode.
This reasoning led to the start of the AI dermatologist project, which required the use of both types of AI in the same system. A session with a dermatologist begins with an examination of the problem skin. As a result, CNN will give us, by analogy, a list of possible diagnoses from a photo.
Then, using the classic deduction method, our CAI emulates the brain of a dermatologist, assuming they have brilliant analytical skills. It identifies the shortest path based on the number of questions asked, using data from the network to get the most reliable result. The answer to each of these questions changes the probabilities of possible diagnoses, and the best diagnosis is when competing diagnoses are much less likely.
The system automatically generates questions from the knowledge base data and conducts a conversation with a user in normal English.
A brief description of a DExpert personal AI dermatologist:
The user takes a photo or uploads a photo of the problematic skin to the program, after which the system determines a list of the most likely candidates, calculates the minimum number of questions that will clarify the diagnosis and starts a dialogue.
The user can receive a system recommendation by clicking on the button below, or investigate possible candidate diseases by requesting a photo and description from the system.
After receiving the responses, the system calculates a diagnostic table:
The user can get a system recommendation by clicking on the button below, or investigate possible candidate diseases by requesting a photo and description from the system.
A screenshot of the admin system (far right panel – knowledge views).
The described system is suitable for creating AI doctors both in related (cosmetology) and remote (radiology, tomography) specialties.
Create your free account to unlock your personalized reading experience.