You are asking yourself, so, we can sense raw data, create sequences, detect behaviours and create self-organising networks. They can be recursive in nature and provide depth and many other such capabilities. Yes, we can maybe program it. Sure, it can be distributed programming. But, how does this answer the questions we are trying to answer with current AI systems? How can such organic-based systems be applied to solve problems? What are the usage patterns of such an AI system?
In the current landscape of neural networks, finding answers to questions or decision making seems to be the primary goal of AI. There are various types of networks for different types of questions to be answered. So, a CNN recognises an image, an RNN decides whether the sentence is semantically correct and what the next word suggestion can be as we type in a sentence and so on and so forth. Typically, the output layer gives an output that can be mapped or used to associate a function based on the value. In a CNN, which is used for image recognition, it just maps the scalar array generated to some database map of images present to give an output. Some neural networks return probability values, others return true or false and many such numerical-based output that are mapped to some meaning and taken further in the application. In-fact rather than this being an AI, we need to ask ourselves, why is the image recognition not just a glorified search algorithm or a glorified mathematical computation algorithm.
We find that it is reiterated over and over again that AI should be used to augment human capability rather than replace them. While this maybe true and useful for some use cases, I find in many cases it is best to just create a totally automated AI based system. Obviously, this can only be done if we can create a true AI. The reason I think such an idea is reiterated is because we are not able to develop a reliable knowledge & intelligent system that can work independently. My view obviously is that we are in this state because of the technology we use for AI. As I have said in my first blog on AI, we do not have the necessary technology to create true AI. We need to go away from the traditional computer based thinking to create it.
The other aspect to consider here are the different applications of AI in the current world. It should be recognised that many use cases are considered a problem for AI rather than problems that were introduced because of the way certain things are implemented. They are not genuine problems for which we really need a fix. For example, using AI to spot bot-based tweets or fake tweets. This is the most apt example of introducing problems first, then solving it and claiming it to be technology. Seriously, all this talk about convey your thought as an elevator pitch, convey it in 250 characters, create a viral tweet, fake likes and followers, instant fame are all social media introduced problems, along with many many other cascading problems. Fake news, bot-based spread of misinformation and many such are making of social media which was not necessary in the first place. But, we introduced it and we now are trying to solve it as a grave problem. Hello, just let social media die, you do not have a problem! Wasting energy for an AI system to solve this as a problem is utter foolishness. A connected world is a borg my friends! Does it matter if it is mind connected or phone glue connected?
Let’s take another example of supply chain. Another human introduced problem. Seriously do I really need the laptop speakers I ordered today within EOD? Agreed there are items that are critical such as medicines. But hey, go to the shop and get it. No faster method. Instead, we have introduced a whole lot of problems making them online and having one-day delivery, one hour delivery. Adding algorithms to track, trace deliver faster and faster, storage requirements and many other such BLAHs introducing all sorts of related problems. No, AI should not be used here, pure waste of energy. More than energy waste it makes the problem more acute in all areas of life. Another is AI in finance! Seriously, we want to use AI to provide effective gambling? Aren’t we all already earning buckets without doing anything tangible? AI for marketing? Another junk mail? First junk delivered by my mailbox, then via email, now via internet powered by AI! Seriously? Isn’t this wasted energy? Marketing is a problem? No wonder we get no good ideas taking shape or being implemented. Everything is purely extensions of existing ideas propounded as a problem and a solution proposed for it, marketed by viral tweets and social media. But, I digress. A topic close to my heart of what the world of technology has become, trash!!
Coming back, the major question to ask ourselves is “What should be the output of any AI?” or “How should intelligence be represented?” or “Should it be represented in any other form other than action?”. If we look at ourselves, the only way we know that intelligence is present, is by looking at the actions of a person. So, why should it be different for an AI system we build? Why do we need an AI give an answer to a question in the language we understand? Sure, if we are looking at a NLP algorithm or an application that completes a sentence the output has to be represented in the language, because that is a language application of AI. For any other application, why does it need to be translated into a language we understand or a mathematical language?
So, what would be a typical implementation of an Organic-based AI look like?
Let’s take an example: Taking a medical use case example. Let’s say we have a scenario where we want to detect if a person is diabetic or not. For this, we can definitely create a AI system that detects various parameters such as blood sugar levels at various periods of time, collect them into external systems, analyse them and give an answer of “yes”, “no” or “maybe”. This output along with the analysis data can be subsequently confirmed by the doctor and a prescription of various treatments for the condition given. But, would this really be an AI system? Will this not just go back to being the external computer-based data analytics kind of an application. A true AI in my view should solve the problem intelligently. The question is “What are we trying to solve as a problem here?” The obvious answer is that “we are trying to diagnose and treat a person for diabetes. If they are diabetic we want to ensure that the right levels of insulin is maintained in the blood stream.” When we did not have a systemic way to solve the problem we needed to take tests, diagnose the problem manually based on the tests and manually inject insulin shots periodically where the detected cause is a deficiency of insulin. But, why should a systemic solution also follow the same process? Wouldn’t a better approach be to have an organic-based system that can run within the blood stream of the potential patient. In our current example, the pancreas produce insulin. Where the pancreas is unable to produce insulin, we want to introduce a organic-system that will augment the deficiency. Here, we need an a system that is a combination of an enzyme-detector + insulin-delivery-mechanism. The insulin-delivery-mechanism needs to react to the enzyme-detector to introduce controlled insulin into the bloodstream. The amount of insulin introduced should be based on the sugar level in the blood. Such a system will detect the blood sugar levels and trigger the delivery system to release the insulin into the bloodstream when the pancreas has not acted.
So, why is this a AI system? The process from the enzyme-detector to the insulin-delivery-mechanism is not straight-forward. We need the correct algorithm to understand that the blood sugar levels are increasing and decide the amount of insulin to be introduced into the blood stream. For example, what is needed here is a system that has a detector such as the bioenzyme sensor that senses the blood sugar level. This needs to trigger a protein that can pass through a signalling sequence to a target artificial cells introduced into the bloodstream that can identify the behaviour of the blood sugar level. This target cell should then start forming a network of proteins that can dissolve maybe material of an insulin-bag triggering the release of insulin into the blood stream. The beauty of such a organic systems is that we can introduce it into a person without diabetes and make it such that it is rejected by the body if blood sugar levels behave normally. This whole system becomes an Organic-based knowledge & intelligence system. As I have said before, artificial intelligence need not really be a single computer system with a series of training algorithms. Such systems cannot be used for anything else other than external, virtual simulations that get us nowhere. Systems where Ai is achieved as a part and parcel of the problem is what is required to solve the truly by AI.
Published on Java Code Geeks with permission by Raji Sankar, partner at our JCG program. See the original article here: How can an Organic-based AI system be used? Opinions expressed by Java Code Geeks contributors are their own. |
Hi everyone
Hi everyone in this forum